Journal of the British Interplanetary Society

VOLUME 72 NO.4 APRIL 2019 General interstellar issue

POSITRON PROPULSION for Interplanetary and Interstellar Travel Ryan Weed et al. SPACECRAFT WITH INTERSTELLAR MEDIUM MOMENTUM EXCHANGE REACTIONS: the Potential and Limitations of Propellantless Interstellar Travel Drew Brisbin ARTIFICIAL INTELLIGENCE for Interstellar Travel Andreas M. Hein & Stephen Baxter

www.bis-space.com

ISSN 0007-084X PUBLICATION DATE: 30 MAY 2019 Submitting papers International Advisory Board to JBIS

JBIS welcomes the submission of technical Rachel Armstrong, Newcastle University, UK papers for publication dealing with technical Peter Bainum, Howard University, USA reviews, research, technology and engineering in astronautics and related fields. Stephen Baxter, Science & Science Fiction Writer, UK James Benford, Microwave Sciences, California, USA Text should be: James Biggs, The University of Strathclyde, UK ■ As concise as the content allows – typically 5,000 to 6,000 words. Shorter papers (Technical Notes) Anu Bowman, Foundation for Enterprise Development, California, USA will also be considered; longer papers will only Gerald Cleaver, Baylor University, USA be considered in exceptional circumstances – for Charles Cockell, University of Edinburgh, UK example, in the case of a major subject review. Ian A. Crawford, Birkbeck College London, UK ■ Source references should be inserted in the text in square brackets – [1] – and then listed at the Adam Crowl, Icarus Interstellar, Australia end of the paper. Eric W. Davis, Institute for Advanced Studies at Austin, USA ■ Illustration references should be cited in Kathryn Denning, York University, Toronto, Canada numerical order in the text; those not cited in the Martyn Fogg, Probability Research Group, UK text risk omission. Raghavan Gopalaswami, Aerospace Researcher, India ■ Captions must be labelled with their Fig. number and should be as short as possible. Lamartine Guimarães, Institute for Advanced Studies, Brazil Mark Hempsell, Hempsell Astronautics Ltd, UK Illustrations should be: Takuto Ishimatsu, Massachusetts Institute of Technology, USA ■ Colour or mono, but should be as close to print Les Johnson, Marshall Space Flight Center, USA resolution (300 dpi) as possible. Poor-quality illustrations may compromise the acceptance of Terry Kammash, University of Michigan, USA paper for publication. Images embedded in Word Kelvin F. Long, Initiative for Interstellar Studies documents may be acceptable, but JBIS reserves Inoue Makoto, Institute of Astronomy & Astrophysics Academia Sinica, Taiwan the right to request separate higher-resolution Gregory L. Matloff, City University New York, USA image files from the author prior to publication. Koichi Mori, Nagoya University, Japan ■ Responsibility for copyright clearance of images rests entirely with the author. Richard Obousy, Richard Obousy Consulting LLC, USA Robert Parkinson, BIS, Aylesbury, UK Submission of papers George Schmidt, NASA John H Glenn Research Center, Ohio, USA ■ Papers for consideration should be sent by Paul Schuch, The SETI League Inc, USA email to [email protected] as both a Word document and as a Word PDF file (in order to Tabitha Smith, Bifrost, USA check for font anomalies), together with any Andreas Tziolas, Variance Dynamical Corporation, USA separate image files. Chris Welch, The International Space University, Strasbourg, France ■ If a paper is accepted for publication, the Friedwardt Winterberg, University of Nevada, Reno, USA author will be asked to sign a License to Publish form. This can be downloaded at www.bis- space.com/wp-content/uploads/2012/08/ WebsiteLicense.pdf. ■ Authors will receive a complimentary copy of the issue in which their paper appears. Editor Roger Longstaff Deputy Editor Duncan Law-Green Associate Editors Stephen Ashworth, We respectfully ask authors to adhere to these Keith Cooper, Stephen Gamble, Paul Gilster, Rob Swinney, Production MP3 Media guidelines. Failure to do so will result in the Promotion Gill Norman JBIS Office British Interplanetary Society, Arthur C. Clarke House, delay of acceptable papers for publication. 27-29 South Lambeth Road, London, SW8 1SZ, United Kingdom tel +44 (0)20 7735 3160 email [email protected] www.bis-space.com

Our full Guidelines for Authors can be downloaded DISTRIBUTION from www.bis-space.com JBIS is distributed worldwide by mail and may be received by annual subscription or purchase of single copies. It is available through membership of the British Interplanetary Society at much reduced rates. Subscription details for members, non-members and libraries are available from the above address.

JBIS is a publication that promotes the mission of the British Interplanetary Society. Opinions expressed in signed articles are those of the contributors and do not necessarily reflect the views of the Editor or the Council of the British Interplanetary Society. Security clearance, where necessary, is the responsibility of the author.

Published by the British Interplanetary Society. Registered Company No: 402498. Registered Charity No: 250556. Printed by Latimer Trend & Company Ltd, Estover Road, Plymouth, PL6 7PY, England. © 2019 British Interplanetary Society. No part of this magazine may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying or recording by any information storage or retrieval system without prior permission from the Publishers. contents VOLUME 72 NO.4 APRIL 2019

110 POSITRON PROPULSION for Interplanetary and Interstellar Travel Ryan Weed et al.

116 SPACECRAFT WITH INTERSTELLAR MEDIUM MOMENTUM EXCHANGE REACTIONS: the Potential and Limitations of Propellantless Interstellar Travel Drew Brisbin

125 ARTIFICIAL INTELLIGENCE for Interstellar Travel Andreas M. Hein & Stephen Baxter

OUR MISSION STATEMENT The British Interplanetary Society promotes the exploration and use of space for the benefit of humanity, connecting people to create, educate and inspire, and advance knowledge in all aspects of astronautics.

JBIS Vol 72 No.4 April 2019 109 JBIS VOLUME 72 2019 PAGES 110–115

POSITRON PROPULSION for Interplanetary and Interstellar Travel

RYAN WEED1, BALA RAMAMURTHY1, JOSH MACHACEK1, DANTE SBLENDORIO1, MASON PECK2 1Positron Dynamics Inc, Livermore, California, USA; 2Sibley School of Mechanical and Aerospace Engineering, Cornell University, New York City, USA

Email [email protected]

Current state of the art in-space propulsion systems fail to meet requirements of 21st century space missions. Antimatter propulsion has been identified as a candidate mechanism that could safely transport humans and/or robotic systems with drastically reduced transit times, providing quicker scientific results, increasing the payload mass to allow more capable instruments and larger crews, and reducing the overall mission cost. Propulsion systems based on antimatter have been considered in several manifestations. Despite the substantial performance advantages, significant technical barriers have kept the cost of usable antimatter well outside the realm of propulsion applications. Each design is a trade-off between

mass and complexity, but they all share high specific impulse (Isp) well above those obtained from even the most ambitious electric ion-propulsion. We describe a positron-based propulsion system utilizing radioisotope positron sources combined with annihilation-catalyzed fusion and include a basic design for a propulsion demonstration employing the CubeSat architecture as well a scaled propulsion system that utilizes an ‘on-board’ radioisotope breeding technique.

Keywords: Antimatter, Positrons, Fusion propulsion

1 INTRODUCTION thrust levels and transit times required for interplanetary and interstellar travel. We present initial analysis of a positron The primary challenge of an antimatter propulsion system is source generation concept based on Deuterium-Deuterium conversion of the annihilation products into propulsive force. (DD) fusion neutron capture reaction 78Kr(n,γ)79Kr [8, 9, 10]. One way to do this is by catalyzing a fusion reaction(s), re- This type of radioisotope ‘breeder’ fuel cycle would allow for sulting in fast charged particle products that can be guided to much higher positron source intensities and thrust levels re- produce thrust [1]. Traditional laser or particle fusion-driver quired for an interstellar mission. systems have high mass and power requirements that are not practical for any near-term space applications [2]. 1.1 Why Positrons?

Positrons are the easier form of antimatter to obtain - over Antiprotons have been the antimatter particle of choice for the past 20 years the cost of usable positron production has de- most propulsion system studies over the last several decades creased, and the techniques have become more widely known [11,12,13]. While antiproton annihilation does release approx- [3]. Our solution to antimatter propulsion is based on using imately 2,000 times more energy per annihilation, a large accel- electron/positron annihilation induced fusion reactions, first erator (e.g. CERN, FermiLab) is required to reach the energies proposed in the 1990s [4], but never experimentally meas- required for pair production of antiprotons. Such large systems ured. Recent advances in cold positron production [5], cre- are unrealistic in a spacecraft, therefore, antiproton propulsion ation of dense deuterium clusters on metallic substrates [6], concepts rely on storage of these charged particles in magnetic and measurement of positron catalyzed fusion reaction cross bottles (e.g. Penning traps). These charged particle traps run section [7] show that a Na-22 radioisotope positron catalyz- up against fundamental number density limits (Brillouin limit) ed fusion propulsion system is possible, capable of >10 mN that require large mass and volume to trap a tiny amount of 5 thrust with >10 s Isp. Initial design of an in-orbit demonstra- antimatter, leading to poor propulsion system performance. tion flight spacecraft utilizing the Cubesat architecture is pre- sented. Using commercial GPS, better than 10m orbital-posi- The positron, or anti-electron, is the antimatter counterpart tion accuracy can be obtained, ensuring 99.99% confidence in of the electron. It has the same mass as an electron but oppo- the measurement of 1 deg. inclination change, decoupled from site charge. Positrons are the ‘easier’ form of antimatter and are any in-track perturbations (e.g. atmospheric drag). While ra- produced by several readily available radioisotopes (e.g. Na-22, dioisotope sources are sufficient for demonstration missions, Co-58) in large number and broad energy spread. A radioiso- a regenerative source of positrons will be required to reach tope such as Na-22 has a specific activity of 6,243 Ci/gram. Five micrograms of Na-22 produces a billion ‘hot’ positrons per second. In condensed matter, an energetic positron, like those This paper was presented at the Foundations of Interstellar Studies produced from a radioisotope, will rapidly thermalize (cool) Workshop at CUNY City Tech, New York City, 13-15 June 2017. through inelastic collisions. Eventually, the positron annihi-

110 Vol 72 No.4 April 2019 JBIS POSITRON PROPULSION for Interplanetary and Interstellar Travel

propriate density to catalyze fusion. A permanent magnet array and/or electromagnet provides the field required for each sec- tion, ultimately forming a nozzle to direct the fusion particles. The fusion process (described below) produces charged parti- cles with >MeV energy, travelling at nearly 10% of the speed of light, which are magnetically guided to the engine exit produc- ing thrust.

1.3 The Fuel

Recent work on dense states of deuterium in metal substrates could lead to a further increase in the positron/deuterium Fig.1 Original ‘Photonraket’ design of Sanger [15]. overlap and therefore fusion rate, leading to a corresponding increase in thrust. Ultra-dense deuterium (UDD) states on the lates with an electron in the material and emits several gamma surface and subsurface of Palladium have been reported by two rays. groups, and correspond to a number density between 1027 and 1029 cm-3. [18, 19, 20, 21]. In fact, the University of Gothenburg When a positron annihilates with an electron in free space to group reported on laser induced fusion events in UDD states produce gamma rays, these gamma rays are emitted (approx- on the surface of metals [22]. imately) back-to-back, but in any orientation. Using photons directly for propulsion, as envisioned by Sanger in the 1950’s This work on dense deuterium and UDD, in addition to ear- [14], requires the ability to efficiently reflect photons (Fig. 1). lier work on laser-initiated Coulomb explosion in D clusters [23] inspired us to investigate the existence of a regime of pos- No currently known material is known to be capable of re- itron flux that can produce similar coulomb explosion fusion flection of gamma-rays, like those emitted from antimatter an- process in dense D clusters and surface states. Utilizing these nihilation. In general, the absorption of photons scales with the ultra-dense states results in significantly higher thrust for a giv- mass of the material. Thus, it is likely that a gamma-ray mirror en amount of positron radioisotope by relaxing the conditions would be prohibitively massive for a spacecraft. necessary to produce a fusion burn.

Without a gamma ray mirror, the only mechanism known 2 THEORY to produce a well collimated beam of gamma-rays is the gam- ma-ray laser [16]. This requires the production and control of It has been postulated that positron-electron annihilation a Bose-Einstein Condensate (BEC) of positronium atoms (a could induce fusion reactions in hydrogen isotopes. In this bound state of a positron and an electron). To-date, efforts have process, the positron-electron annihilation couples to a lat- yet to produce a BEC of positronium [17]. tice-trapped deuteron giving it a kinetic energy kick [4]. This 'knock-on' process provides a mechanism to accelerate deute- 1.2 Propulsion Mechanism rium ions using positrons. The accelerated deuteron, in some cases, has sufficient kinetic energy to fuse with other deuterons Positron catalyzed fusion propulsion works by injecting cold in the substrate. positrons onto a region of high deuterium density. Typically, the deuterium is located on the metal surface or inside vacan- Other mechanisms for transferring energy from annihilating cies and other point defects or surface states of the metal lattice positron/electrons to the atomic nuclei have been investigated. as shown in Fig. 2 (C). Fig. 3 (p.112) shows the Feynman diagram for nuclear excita- tion of nuclei via an inelastic (non-resonant) channel. This en- A radioisotope positron source produces isotropic ‘hot’ pos- ergy transfer mechanism is similar to positron nuclear excita- itrons which are cooled by an array moderator to near thermal tion process described by Raghavan and Mills [24] and has been energies (~eV). The resulting ‘cold’ positron beam is guided shown experimentally to produce excited metastable states of and focused into the deuterium-loaded fuel material at the ap- surrounding nuclei, with the cross section increasing with in-

A B C D

Fig.2 (A) Radioisotope source of hot positrons (red). (B) positron moderator produces high flux of cold positrons. (C) Deuterium loaded metal. Deuterium clusters (blue) have formed in vacancy type point defects or surface states. Injected cold positrons will diffuse into the same defect sites before annihilation. (D) Fusion reaction products (red) are magnetically guided to produce thrust along magnetic field line direction.

JBIS Vol 72 No.4 April 2019 111 RYAN WEED ET AL

Fig.3 Feynman diagrams for Inelastic Nuclear Excitation Process (left) and the Morioka Process (right). Qint describes the sudden momentum transfer interaction with the surrounding crystal (analogous to the Mossbauer effect). creasing atomic number of the absorbing atom [25, 26, 27]. the positrons can be used as such, they must be cooled down to < keV energies in a process called moderation. The efficien- Fig. 3 shows the Feynman diagram for a similar nuclear cy for the moderation process to date has been <1%. Positron excitation process, whereby a virtual photon imparts a large Dynamics has developed new methods to increase modera- amount of kinetic energy to a trapped deuteron, theorized by tion efficiency by several orders of magnitude [5], combining Morioka. In the case where positron and Deuterium atoms are a technique called Field-Assisted-Moderation [28, 29, 30] in co-located in the potential well of the lattice point defect, we wide bandgap semiconductor (Silicon Carbide) arrays with a consider the knock-on Deuteron process induced by the anni- charged particle extraction technique using crossed electric hilating positron inside the metal vacancy. Typically, the fusion and magnetic fields (ExB drift) (Figs. 4 and 5). Using such an rate is defined by: efficient moderator will allow for production of intense and -fo cused pulses of positrons that are able to deposit substantial amounts of power into fuel targets. where and are the velocity and cross section of the deu- teron-deuteron system. However, in the case where we consid- 3 SPACECRAFT DESIGN AND PERFORMANCE ESTIMATES er momentum transfer from the annihilating positron to the surrounding Deuteron the fusion rate becomes: Using measured value for momentum transfer probability R of 10-1 [7] we estimate that a 20um diameter, ns pulse of 1011 positrons could deposit >1023 W/cm3 onto a thin film substrate surface covered in 50 nm of UDD material of number density where R is the non-relativistic spin averaged differential cross 1029/cm3, leading to hot spot ignition [31]. Using a conservative section for the momentum transfer process described above. estimate for burn fraction (10%) in the implantation volume, Although the value for R will depend on the substrate material this leads to a thrust of approximately 2mN/Ci. Thrust pro- characteristics, it has been calculated at R~10-3 for reasonable duced by the >MeV DD fusion reaction products is given by: atomic parameters [4].

It is important to note that the fusion rate depends only on the overlap in the positron and Deuteron number densities. In where is the mass flow rate. The specific impulse is relat- order to ensure sufficient overlap with reasonable fuel geome- ed to the exhaust velocity of the fusion products and resulting tries the positrons must be cooled, or moderated. plasma expansion in the magnetic nozzle. Depending on fuel/ substrate geometry and makeup, magnetic nozzle properties, When positrons are born, they are extremely energetic or ignition burn characteristics, radiative and ablative effects, this ‘hot’ (mean energy ~250keV) and thus, difficult to control. One would lead to a specific impulse upper limit of 2 x106 seconds. significant challenge to date is the ability to control these very hot positrons using realistic electric and magnetic fields. Before Total impulse will be limited by the half life τ of the radi-

Fig.4 Planar Array moderator extraction concept (single element). Fig.5 Planar Array moderator extraction concept (single element).

112 Vol 72 No.4 April 2019 JBIS POSITRON PROPULSION for Interplanetary and Interstellar Travel oisotope and the reduction in Deuterium density as the fuel is burned, described by decay rate , where is the initial positron flux and ε is the moderation and transport efficiency. Therefore:

We consider an example propulsion system using a Na-22 source with a half-life of 2.6 years and activity of 14 Ci /cm2 [27] over a 4 cm2 surface. This gives an initial total thrust of 132 mN and total impulse of ~9.4x 105 Ns over a mission du- ration of tf (Fig. 6). We define the mission duration as the time at which thrust has dropped to 5% of initial thrust, which, de- pending on the mission thrust profile requirements and can range from 0.7 to 6 years for a Na-22 positron source. Fig.6 Thrust available for positron catalyzed fusion propulsion 3.1 Cubesat Demonstrator system based on Na-22 source of 56 Ci activity. The reduction in thrust in the solid line (100% duty) is mainly due to burn-up of the In order to validate this concept, a small propulsion subsys- Deuterium fuel, while the thrust decay in the dotted line (5%) is tem (<6kg) will is being designed in partnership with PD and primarily due to the half-life of the positron source. the Space Systems Design Studio at Cornell. The propulsion unit will be housed inside a 6U CubeSat (a standardized small satellite form factor) weighing approximately 12kg (Fig. 7). lifetime of positron emitting radioisotopes, these higher thrust Initial estimates show this mass and volume is sufficient for a levels require on-board production of positrons. Such a system six-month demonstration mission. Launch, ascent and orbit would allow for an approximately ~50 year transit to Alpha environments are well-characterized and flying fragile science Centauri, with the spacecraft reaching nearly 10% of the speed hardware is not a unique challenge. Standard methods to test of light. and mitigate damage (vibration table, thermal-vacuum testing, etc) during these phases of flight test are well understood and 3.3 79Kr Breeding* (on-board positron source) will be utilized. Fortunately, the Deuterium-Deuterium fusion process produc- The spacecraft will change its inclination over the course of es an abundance of fast neutrons. In this case, we may devise a the six months using the positron-catalyzed propulsion subsys- radioisotope breeding technique* that utilizes a high neutron tem. The volume and mass of the flight demonstration space- capture cross section [15] in Krypton-78 to produce Kryp- craft will require a 1 Ci Na-22 positron source, with a half-life ton-79, a positron emitting radioisotope (see Figure 8). This of 2.6 years, to provide a sufficient number of positrons for ap- fuel cycle will allow for scaling of Thrust to approaching 100N. proximately 2mN of instantaneous thrust. The Na-22 source would be classified as radiotoxicity group III “minor sources” according to the FAA’s Office of Commercial Space Transpor- tation [32]. Recent advances in composite shielding materi- als will be utilized in order to minimize additional spacecraft weight due to shielding requirements [33]

3.2 Scaling to Alpha Centauri

The Alpha Centauri mission would require to scale thrust lev- els between 10N-100N. Due to limits in specific activity and

Fig.8 Production channels for 79Kr, including the neutron capture cross section 78Kr(n,g)79Kr considered for breeding positron emitting radioisotope 79Kr. Reproduced from Ref. 15.

*A traditional fission breeder reactor creates more fissile material than it uses. Here, in the context of a fusion reactor and with the addition of an intermediate step (positron production/annihilation) we will use the same term ‘breeder’ to indicate that the fusion Fig.7 Rough sketch of Cubesat design for 1U positron source and reactions generate neutrons that produce more positron emitting moderator subsystem. radioisotope, which in turn produce more fusion reactions.

JBIS Vol 72 No.4 April 2019 113 RYAN WEED ET AL

Fig.10 A numerical solution to the 79Kr breeding cycle over approx. Fig.9 The 79Kr breeding cycle. 10 years. The initial 79Kr breeding period lasts ~2 months.

The fuel cycle [10] is summarized in Fig. 9. Hot positrons mass of 100kg, including 100ug of 79Kr. The model also in- are generated from layers of Krypton (79Kr-rich) frozen onto cludes an artificial ‘limiter’ that maintained 79Kr total mass metallic surfaces in an array structure described in section 2. below 100g and thrust at a constant level. In practice, this Fortunately, Krypton not only serves as the source of positrons, would be accomplished by controlling the repetition rate of the it also makes an excellent positron moderator [30]. In the en- pulsed positron beam. The 79Kr in the source layers will decay gine core, large positron pulses generate fusion reactions on to 78Br, from which CBr4 can be generated and would act as an the fuel target substrate, generating a high flux of fast neutrons. efficient positron cooling gas in a buffer gas type accumulator It is estimated that 10-100atm blanket of pressurized Kr will be stage. sufficient to thermalize these hot neutrons within a reasonable length scale (<1m). The reference design including a rough mass budget and delta-V estimates is included in Table 1 below. The mass frac- The Krypton then passes into the cryogenic isotope en- tion for this design is 0.7, slightly lower than 0.8 mass fraction richment stage. This allows for source specific activities (Ci/g) of the retired Space Shuttle system. The total mass of the ref- high enough to generate the number of positrons required for erence design at approximately 1.5mT could be launched into higher thrust levels. A candidate isotope separation method LEO by a small or medium lift launch vehicle (Minotaur-C, is described by Mills [9,16], where 79Kr created in the engine Falcon 9, Delta II, etc) and would not require in orbit assembly core is preferentially separated from the 78Kr using a liquid (approximate size 6m length x 2m diameter). The fast neutron nitrogen immersed tube with an array of heater elements and a flux in Table 1 is high enough to cause damage to surrounding carrier gas that oscillates pneumatically. materials (hardening / embrittlement / creep / phase instabil- ity). However, such a high neutron flux is seen in the core of A numerical solution to the Krypton breeding fuel cycle is high flux fission reactors [9] without catastrophic structural shown in Fig. 10. This model assumed a total initial Krypton failure over the course of a decade.

TABLE 1 Key parameters for reference design with scaled propulsion system based on 79Kr breeding and upper limit of specific impulse (2x106 secs). Krypton Deuterium Structure/Payload Thrust Fast neutron flux Delta-V Alpha Cen 100 kg 1000 kg 400 kg 60N 5E15 n/s×cm2 9E6 m/s 60 years

Fig.11 Reference design for scaled propulsion system based on 79Kr breeding.

114 Vol 72 No.4 April 2019 JBIS POSITRON PROPULSION for Interplanetary and Interstellar Travel

4 CONCLUSIONS sor and other payloads. The current estimates of performance provide an upper bound based on uncertainties in fusion burn Interstellar travel within the span of a human lifetime is per- parameters. Non-ideal effects such as radiation, ionization, ab- haps the greatest technological challenge that humanity faces. lation and collisional losses can lead to a significant reduction The vast distances, incredible cold, harsh radioactive and mi- in performance (e.g Isp). Future work is needed to determine cro-particle environment requires novel solutions if we are to practical performance parameters. venture outside of our . Conquering these chal- lenges will prove to be difficult, but the rewards are profound. Antimatter based propulsion is a game-changing propulsion technology. We have laid out the path to an antimatter propul- Unlike beam-powered propulsion, this propulsion concept sion technology demonstration based on an available radioiso- would allow a spacecraft to carry large scientific or mission tope positron source that does not require long term antimatter payloads and would also allow the spacecraft to slow down trapping and can be demonstrated in the near term. This con- once it has reached its target destination. The fuel cycle de- cept is scalable to fast transit, high delta-V missions with small scribed in section 4 could also provide abundant source of to medium sized spacecraft. thermoelectric power to run avionics, communications, sen-

REFERENCES

1. R. Keane, “Beamed Core Antimatter Propulsion”, Journal of the British 19. F. Winterberg, “Ultra-dense deuterium and cold fusion claims”, Physics Interplanetary Society 2012 Letters A, Volume 374, Issue 27, 14 June 2010, Pages 2766-2771 2. J. Lindl, et al. "Review of the national ignition campaign 2009-2012. 20. L. Holmlid, “High-charge Coulomb explosions of clusters in ultra-dense Physics of Plasmas 1.2 (2014): 020501. deuterium D(−1)”, International Journal of Mass Spectrometry, Volume 3. R. Krause-Rehberg, “Positron Annihilation in Semiconductors: Defect 304, Issue 1, 15 June 2011, Pages 51-56 Studies”, 1999 21. P. Andersson, “Ultra-dense deuterium: A possible nuclear fuel for 4. S. Morioka, “Nuclear Fusion Triggered by Positron Annihilation in inertial confinement fusion (ICF)”, Physics Letters A, Volume 373, Issue Deuterated Metals”, Il Nuovo Cimento, Vol107a,1994 34, 17 August 2009, Pages 3067-3070 5. R. Weed, et al, “Array Structures for Field Assisted Positron Moderation 22. L. Holmlid, “Laser-induced fusion in ultra-dense deuterium D(−1): and Corresponding Methods”, WO Patent App. PCT/US2012/042,049 Optimizing MeV particle emission by carrier material selection”, Nuclear Instruments and Methods in Physics Research Section B: Beam 6. B. Shahriar, P. U. Andersson, and L. Holmlid, "High-energy Coulomb Interactions with Materials and Atoms, Volume 296, 1 February 2013, explosions in ultra-dense deuterium: Time-of-flight-mass spectrometry Pages 66-71, with variable energy and flight length.” International Journal of Mass Spectrometry 282.1 (2009): 70-76. 23. T. Ditmire, J. Zweiback, V. P. Yanovsky et al, “Nuclear fusion from explosions of femtosecond laser-heated deuterium clusters”, Nature, 398, 7. R. Weed, et al., Positron catalyzed fusion reactions in deuterated pp.489-492, 08 April 1999. palladium, Unpublished work, Livermore, CA 24. R. S. Raghavan and A. Mills, “Nuclear Excitation by positron 8. A. P. Mills Jr, “Suitability of 79 Kr as a Reactor-Based Source of Slow annihilation: Comments on theory vs experiment” Physical Review C Positrons.” Nuclear Science and Engineering 110.2 (1992): 165-167 Vol 24 Num 4, October 1981 9. A. P. Mills Jr, “Physics with many positrons.” Rivista del Nuovo Cimento 25. R. D. Present, S. C. Chen, “Nuclear Disintegration by Positron-K della Societa Italiana di Fisica 34.4 (2011): 151-252. Electron Annihilation,” Physical Review, Vol 85, Num 3, February 1952 10. R. Weed et al, “Positron-emitting Noble-gas Fusion Breeder Reactor, 26. D. B. Cassidy et al, “Resonant versus nonresonant nuclear excitation Patent date Filed”, Dec 8, 2016 US 52254426 of 115In by positron annihilation,” Physical Review C, Vol 64 054603, 11. G. Schmidt, H. Gerrish, J. J. Martin, “Antimatter Production for Near- October 2001 term Propulsion Applications,” 1999 Joint Propulsion Conference. 27. A. Ljubičić, “Nuclear excitation in 176Lu by positron annihilation on 12. R. L. Forward, “Antiproton Annihilation Propulsion,” Journal of K-shell electrons” Journal of Radioanalytical and Nuclear Chemistry V Propulsion, 1 (5), 370-74 (1985). 272 2007 13. G. A. Smith, G. Gaidos, R. A. Lewis, K. Meyer and T. Schmid, “Aimstar: 28. K. G. Lynn and B.T.A. Mckee, “Some Investigations of Moderators for Antimatter Initiated Microfusion for Precursor Interstellar Missions,” Slow Positron Beams”, Applied Physics, 1979. 19(3): p. 247-255. Acta Astronautica, 44 183-86 (1999) 29. Al-Qaradawi, I.Y., P.A. Sellin, and P.G. Coleman, “Tests of a diamond 14. E. Sanger, “Zur Theorie der Photonenrakteten”, Ing. Arch. 21, 213, 1953. field-assisted positron moderator”, Applied Surface Science, 2002. 194(1- 15. R. Hoffman et al. “Neutron and charged-particle induced cross 4): p. 29-31. sections for radiochemistry in the region of bromine and krypton”, No. 30. J. P. Merrison, et al., “Field Assisted Positron Moderation by Surface UCRL-TR-205563. Lawrence Livermore National Laboratory (LLNL), Charging of Rare-Gas Solids”, Journal of Physics-Condensed Matter, Livermore, CA, 2004. 1992. 4(12): p. L207-L212. 16. D. B. Cassidy and A. P. Mills, (2007), Physics with dense positronium. 31. M. D. Rosen, “The physics issues that determine inertial confinement Phys. Status Solidi C, 4: 3419–3428. fusion target gain and driver requirements: A tutorial.” Physics of 17. K. Shu et al, “Study on Bose-Einstein condensation of positronium.” Plasmas 6.5 (1999): 1690-1699. Journal of Physics: Conference Series. Vol. 791. No. 1. IOP Publishing, 32. Space Applications of Radioactive Materials, Office of Commercial 2017 Space Transportation Licensing Programs Division, Federal Aviation 18. A. Lipson, B. J. Heuser, C. Castano et al, “Transport and Magnetic Administration, 1990 Anomalies below 70K in a Hydrogen Cycled Pd Foil with a Thermally 33. S. Chen, M. Bourham, A. Rabiei. “Attenuation efficiency of X-ray and Grown Oxide”, Physical Review B 72, 212507 2005 comparison to gamma ray and neutrons in composite metal foams” Radiation Physics and Chemistry, 2015

Received 9 April 2018 Approved 30 October 2018

JBIS Vol 72 No.4 April 2019 115 JBIS VOLUME 72 2019 PAGES 116–124

SPACECRAFT WITH INTERSTELLAR MEDIUM MOMENTUM EXCHANGE REACTIONS: the potential and limitations of propellantless interstellar travel

DREW BRISBIN Universidad Diego Portales, Núcleo de Astronomía, Av. Ejercito 441, Santiago, Chile

Email [email protected]

All known interstellar transportation methods encounter monumental technological or engineering roadblocks, or even rely on speculative unknown science. In particular light of the recent public excitement and ensuing disappointment regarding the exotic “EM drive” it is worthwhile to point out that propellantless space travel is eminently possible based on well established physical principles. Here a new mode of transport is proposed which relies on electric-field moderated momentum exchange with the ionized particles in the interstellar medium. The application of this mechanism faces significant challenges requiring industrial-scale exploitation of space but the technological roadblocks are different than those presented by light sails or particle beam powered craft, and may be more easily addressed depending on the uncertain march of technology. This mode of space travel is well suited to energy efficient travel at velocities <0.05c and compares exceptionally well to light sails on an energy expenditure basis. It therefore represents an extremely attractive mode of transport for slow (~multi-century long) voyages carrying massive payloads to nearby stellar neighbors. This would be useful for missions that would otherwise be too energy intensive to carry out, such as transporting a generation ship or bulk materials for a future colony around Alpha Centauri A.

Keywords: Propellantless space travel, ISM, Electric sail, Magsail, Light sail, SWIMMER

1 INTRODUCTION cle-beam powered spacecraft. This hinges on a sail formed by an extended electric or magnetic field which is able to deflect The tyranny of the rocket equation has long been recognized as a remotely-beamed stream of charged particles. Since charged an impediment to becoming a truly spacefaring species. Due particles carry much more momentum per unit energy than to the exorbitant reaction mass required for traditional rockets photons this could have much lower power requirements than in interstellar travel, there has been considerable attention to light sails. This concept has its origins in the Magsail, a large methods of space travel that circumvent the rocket equation. loop of current carrying wire which deflects passing charged Laser-driven light sails are a prominent and long-standing particles in the interstellar medium (ISM), eliciting a drag idea (see for example [1] and references therein). While light force which could be used as a brake to slow spacecraft down sails are well established and also the engines of the widely to rest with respect to the ISM after a high speed journey [6]. publicized “Breakthrough Starshot” program [2] and “Project To provide acceleration, one could simply replace the ISM Dragonfly” [3], their thrust is fundamentally limited to 6.67 with a beamed source of high velocity charged particles [7]. N GW-1. For comparison, the Three Gorges Dam, the largest Providing a long distance beam of charged particles is, how- capacity power plant currently in operation, has a capacity of ever, quite difficult because of beam divergence due to particle about 22.5 GW. If this power was transmitted with perfect ef- thermal motion, interaction with interplanetary or interstellar ficiency to a light sail it would provide thrust equivalent to the magnetic fields and electrostatic beam expansion in the case force required to lift a 15 kg mass on Earth. Scaling light sails of non-neutral particle beams. Andrews (2003) suggests that it up to larger-than-gram-scale spacecraft therefore necessarily would be necessary to construct a highway of beam generators depends on humanity's ability to harness incredible power. Al- at least every AU or so along the route on which the craft accel- ternatively, direct sunlight could be used as a source of photon erates [8]. The related concept of the electric sail instead uses pressure. Unfortunately, the material properties suggested to be an electric field generated by a positively charged grid of wires necessary for a practical interstellar solar sail require extremely or wire spokes extending from a central hub to push against the low areal density materials with σ 10-3 g m-2 [4]. Current state- outward streaming solar wind [9]. This concept has the near of-the-art reflective films developed for light sails reach areal term potential to allow travel within our own stellar neigh- densities of ~10 g m-2, or four orders of magnitude too dense borhood with very low energy costs. The electric sail, like the even without including any support structure or payload, so it Magsail however, ultimately relies on a drag force, decelerating is uncertain when if ever suitable materials will be developed the spacecraft to rest with respect to the surrounding medium for a solar sail [5]. (the outward moving solar wind in this case). It is therefore unable to accelerate beyond the heliosphere, nor can it accel- Another idea using external reaction mass is the parti- erate directly inwards towards the sun while in the heliosphere

116 Vol 72 No.4 April 2019 JBIS SPACECRAFT WITH INTERSTELLAR MEDIUM MOMENTUM EXCHANGE REACTIONS: propellantless interstellar travel

(though tacking at an angle to the wind along with gravitation- like the submerged paddle on a paddle steamer boat pushes al attraction do allow it to more slowly reduce its radial helio- on the surrounding water. This process will also slow down centric distance). the spin rate of the overall vehicle, but this can simply be spun up again by use of the central reaction wheel (using some on- It would be possible to overcome these obstacles by actively board or beamed power source). Although this is an illustrative pushing against the charged particles of the ISM, rather than example of a SWIMMER, it would likely pose extreme diffi- passively coming to rest with respect to the medium. These culties in implementation for fast (v~0.05 c) space travel. Note Spacecraft With Interstellar Medium Momentum Exchange that each sail can only push on the ISM when its velocity is neg- Reactions (SWIMMERs) can accelerate with respect to the ative with respect to the ISM. To accelerate the spacecraft up to ISM, are significantly more energy efficient than light sails, do 0.05 c then, would require that the electric sails be moving at not require pre-established infrastructure along the route and a speed of at at least 0.05 c with respect to the vehicle center of are based on elementary physical principles. Recently Robert mass. Based on the requirement that the sails be as low mass Zubrin discussed his independent work on a “dipole drive” as possible, likely constructed out of thin strands of supercon- concept which bears a striking resemblance to the SWIMMER ducting wire, they are likely to be fragile, wispy structures and concept described here [10]. Although the two ideas are re- it seems unlikely that such sails and tethers could be robust lated and even share a similar geometry, they were arrived at against the strains involved while retaining their low mass. independently. Furthermore the dipole drive, as described by Robert Zubrin, suffers from a flaw which prevents its successful Another option with fewer moving parts, could operate by acceleration in the ISM. The work presented here concerns the first setting up inhomogeneities in the ISM arranged especially conceptual mechanism which allows SWIMMERs to acceler- so that an electric field can push asymmetrically in one direc- ate through a stationary ISM. The mathematical properties of tion. There is an arbitrary number of configurations that could the idealized governing equations are developed and potential achieve this. One simple implementation, illustrated in Fig. 1, future SWIMMER missions are explored. This work adopts the could feature a pusher plate made of a large grid or long tethers nomenclature that Log(x)≡Log10(x). of wires moving face-on through the ISM (much like the pro- posed geometry of a standard electric sail). Unlike a standard 2 SWIMMER DRIVE

Both the Magsail and electric sail concepts rely on the fact that there is significant mass in the ISM (or the heliosphere) which can interact with relatively low mass structures consisting of charged or current carrying wires. How, then, could a space- craft interact to accelerate rather than decelerate with respect to the surrounding medium?

Generally this will require a time varying electric field which can do work on the surrounding particles of the ISM. As a thought experiment, one can imagine a large paddy-wheel structure, a bit like the paddy-wheel of a paddle steamer boat, with two electric sails mounted opposite each other at the ends of two long tethers which are electrically connected and across which can be applied a potential difference. The tethers are mounted to a reaction wheel in the center and the whole sys- tem is set spinning with the axis perpendicular to the direction of travel (defined as the positive direction). If the spin rate is fast enough, there will be portions of the cycle during which the sails have negative velocity with respect to the ISM. As one of the electric sails (sail “A”) approaches the portion of the cycle when its velocity is negative, a potential difference is applied, charging sail A positively and sail B negatively. In the frame of sail A, ions in the ISM are streaming towards it and pushing it in the desired direction of travel. Simultaneously the nega- tively charged sail B will be reflecting electrons in the positive direction causing some drag. Since ions out-mass electrons by at least a factor of mp/me=1836, the electrons contribute negli- gible drag and can be ignored throughout the analysis. As the Fig.1 A schematic representation of a SWIMMER in operation. rotation cycle continues, sail A moves into the portion of the Tethers are shown branching off from each other. Laser light energy, cycle where its velocity is no longer negative with respect to represented by the blue squiggly arrow, is beamed from a power the ISM and the electric sails are neutralized. Then as sail B station located out of the figure to the right. It is absorbed and approaches the negative velocity portion of the cycle, the po- converted into electrical energy in the hub located at the center of tential difference is turned on and reversed, charging sail B the tether network, which would also be the location of the payload. positively. By charging the sails positively only when they have The inset illustrates the bi-layer braided structure of a single tether. negative velocities with respect to the ISM, they can operate In its current charge state, with the front layer charged positively like standard electric sails, exchanging momentum with the and the back layer negative, the SWIMMER is in a primer stage, ambient medium and slowing down, while giving the overall pushing on positive ions in front to set up a clump in its immediate spacecraft a positive momentum boost. In this way the posi- path. The braided inset figure is adapted from a creative commons tively charged electric sail pushes on the ambient ISM much file [11]. It has been cropped and edited for 3-d effect.

JBIS Vol 72 No.4 April 2019 117 DREW BRISBIN electric sail, however, the grid of wires would actually be two the pusher plate and the cycle repeats. Fig. 2 shows the electric identical layers of wire sandwiching a strong insulator between potential and ion density at various cycle stages for a simple them to keep the two layers physically apart and electrically model which represents the electric potential due to the pusher isolated. These wire grids or tethers could be made from very plate as two potential ramps extending out 20 m. The ions are fine superconducting wire and the entire ensemble could be assumed to initially be travelling rightward at 0.001 c, and their spun to create tension and keep the wire grids extended with- effect on the electric potential is ignored. out heavy support structure. In the “primer” portion of the op- eration cycle, the front layer in the pusher plate is raised to a This process is conceptually straightforward and obeys all positive potential φ, modestly above the stopping potential of conservation laws. The spacecraft gains momentum by giving 2 the ions, φstop≡vion mion/(2e) where e represents the elementary backward momentum to the ISM (pushing ions to the right in charge, mion is the mass of the ion species and vion represents the Fig. 2). The source of the potential difference does work in the maximum of either the ion thermal velocity or the streaming primer stage when it sets up the positive potential, raising the velocity (the velocity of the spacecraft itself, assuming travel electrical potential energy of the ion clump and again in the through a stationary medium). The back layer is charged to -φ. push stage when it raises the clump to a higher potential. In Due to edge effects of the finite plates and the self-shielding this idealized one-dimensional case, it is also very energy ef- behavior of plasmas, this results in a decaying electric poten- ficient. By appropriately tuning the cycle timing and the elec- tial of opposite sign on either side of the plates. Ions streaming trical potential levels the SWIMMER will avoid sending any towards the front positively charged layer slow down, building ions initially in the vicinity of the plates forward to infinity (to up an overdense clump in front of the pusher plate while an un- the left). Electrons encountering the negative potentials will be derdensity forms at the immediate location of the pusher plate. reflected but this causes only a negligible momentum drag. In a Then in the “pull” stage of the cycle the potential difference real three dimensional case, there will also be loss of efficiency across the layers is reversed and significantly increased. The due to particles which do not interact perfectly in one dimen- ion clump that was formed in front of the plate will be attracted sion, but instead are pushed off to the side as they pass by the to the negative front layer, pulling the spacecraft forward. Any charged wires. ions that transition through the pusher plate at this moment will cause significant drag as they encounter the strongly rising This qualitative conceptual analysis does not account for electric potential crossing the thickness of the plate. Fortunate- the self influencing behavior of plasmas. This will undoubtedly ly, the underdensity set up in the primer stage ensures there strongly influence the ion (and electron) distributions and the will be very few if any ions which will encounter the pusher extended electric potential. Detailed particle-in-cell simula- plate before the clump. As the clump approaches the pusher tions will be necessary to investigate the optimal tuning of cycle plate, the potential difference is turned off and the clump is al- timings and electrical potentials, and these will be affected by lowed to coast through the plate to the other side. In the final ISM density, pusher plate size, plasma (spacecraft) bulk veloci- “push” stage the same potential difference is applied and the ty, plasma temperature and the available power. These simula- clump is further pushed backwards by the positive back layer of tions are beyond the scope of this paper but these collective ef- the pusher plate. The clump drifts away beyond the influence of fects are unlikely to eliminate the features created by the primer

Fig.2 An example showing snapshots of the electrical potential and ion density in one dimension cutting across the SWIMMER pusher plate at various times. The two-layered pusher plate is located at 45-46 m. Black lines indicate the electrical potential and red lines indicate the ion density (where the average ISM ion density is n0). In the primer stage, a positive potential ~φstop sets up an ion density gap at the location of the pusher plate and an overdensity on the upwind side. In the pull stage the potential difference is reversed and increased (note the change in the y-axis). This pull stage persists until just before the ion clump passes through the plate, at which point the pusher plate layers are neutralized. Once the clump passes through, the potential difference is restored beginning the push stage. This example assumes an initial uniform ion velocity of 0.001 c and does not account for the electric potential contributed by the ions or electrons.

118 Vol 72 No.4 April 2019 JBIS SPACECRAFT WITH INTERSTELLAR MEDIUM MOMENTUM EXCHANGE REACTIONS: propellantless interstellar travel stage including a leading ion clump with an underdensity at the Now assume the energy donated to the ions is given by some pusher plate. These inhomogeneities are the crucial feature that power, P, applied over a small amount of time, Δt: ΔE=P Δt. allow SWIMMERs to push on the ISM asymmetrically. Fur- The mass of ions that the energy is applied to is given by the thermore, this is only one possible configuration of a SWIM- mass of ions swept out in time Δt by some cross sectional in- MER. It would also be possible to use multiple pusher plates to teraction area of the pusher plate, A: mion=A v x n mp where n accelerate ion clumps across a series of potential differences to is the ion density and mp is used as the individual ion particle gain more thrust per ion, at the expense of a more complex and mass under the simplifying assumption that all the ions are massive pusher plate. The effectiveness of a SWIMMER will -ul protons. The acceleration of the spacecraft is found by making timately need to be tested by simulation, small scale laboratory these substitutions in Eq. 1, subtracting the initial velocity v tests and real world application. from v', dividing by Δt and then taking the limit as Δt→0. The force on the ship, FSWIMMER, is simply acceleration times the ship The configuration described here, a large pair of wire grids mass and is given by: with opposite charges to push on the ambient ISM, is very similar to that recently described by Robert Zubrin as the di- pole drive [10]. In the case of the dipole drive, however, the (2) electric field is apparently static rather than pulsed, the wire grids are separated by a significant distance and they push on The argument of the square root is real and positive. Choos- the charged particles as they pass between the plates. At first ing the negative root corresponds to the situation in which the look, this seems like a reasonable and simpler approach. Two spacecraft gives up some momentum and sends the ions in the oppositely charged infinite plates produce a strong electric field positive direction while slowing itself down, a braking force. between them and no electric field outside, so by simply push- Choosing the positive root corresponds to the spacecraft send- ing the heavy ions between the plates in the correct direction ing the ions in the negative direction and accelerating itself for- this static electric field should create thrust. Unfortunately, the ward. A braking force could be generated by reversing the po- approximation of infinite plates leads one astray here. In fact, larity of the pusher plate during the push and pull stages shown a finite system of parallel plates will produce an electric field in Fig. 2. Eq. 2 represents the ideal limit of the force generated outside the plates pushing in the opposite direction. Although by any system generating thrust by pushing on the surrounding these fields will be weaker than the field between the plates, ions with perfect efficiency. they will also extend over a larger region, cancelling out the thrust gained from particles between the plates. Indeed, any The power referred to throughout this work is the delivered system of charges over a finite area must leave the electric po- electrical power. Thus far the source of power for a SWIMMER tential zero at infinity. Without any change in the electric field, has been ignored. There is no reason a SWIMMER could not any particles coming from far away and leaving far away begin use an onboard power source, making it totally independent and end with zero electric potential energy and no change to of external infrastructure. This, of course, would require an their kinetic energy. At most their velocity vector may change exceptionally energy dense fuel source as well as a very effi- direction and lead to a change in momentum, but this change cient generator to achieve useful velocities for interstellar travel in momentum could only be used to decelerate (with respect to (note that if the spent fuel mass rate is high and used fuel is the ambient medium) or change direction. continuously ejected, this would alter Eqs. 1 and 2 as the ship's mass decreases throughout the interaction, but if spent fuel is 3 MATHEMATICAL EXPRESSION OF AN IDEALIZED CASE held on board these equations would not change). Beaming power remotely to the SWIMMER is possibly a more viable To examine the limits of SWIMMERs, consider a spacecraft strategy for interstellar travel, which invites a direct compari- of mass, m, moving with some velocity, v, in the frame of the son to light sails. In this case an additional P/c term is included surrounding medium (this would be the stationary ISM frame in Eq. 2, corresponding to the photon pressure of the beamed in interstellar space or a frame that is comoving with the solar energy being absorbed by the spacecraft. The total force is then: wind within a heliosphere). The spacecraft’s direction of trav- el is defined to be positive. Ignoring the details of operation, (3) at a given moment the spacecraft is able to inject some small amount of energy, ΔE, into a collection of ions in its vicinity of The photon force is added or subtracted depending on mass mion, increasing the system's kinetic energy and changing whether the beamed energy is directed in the same direction as the momentum of the spacecraft and the ions. the SWIMMER velocity or opposed to it respectively. Explicitly then, there are four different modes of operation for a SWIM- By conservation of energy and momentum the final ship ve- MER depending on the orientation of the velocity, the photon locity, v', and the ion velocity vion' is: force and the interaction force from pushing on the ions in the surrounding medium (FSWIMMER). These are illustrated in Fig. 3 overleaf along with their corresponding implementations of Eq. 3 with explicit positive and negative sign choices. These four modes of operation are referred to respectively as the (1) “normal” mode when the velocity, FSWIMMER and photon force are all oriented in the same direction; “tractor beam” mode when the velocity and FSWIMMER are aligned with each other but opposed to the photon force; “destination braking” when the photon force and velocity are aligned but opposed to FSWIMMER; and “home braking” when the photon force and FSWIMMER are aligned but opposed to velocity. The ability to operate in these different modes is one of the advantages of SWIMMERs. Un- like light sails they are able to decelerate at their destinations

JBIS Vol 72 No.4 April 2019 119 DREW BRISBIN

Fig.3 The four modes of operation for a SWIMMER. Red arrows indicates velocity, black arrows indicate the photon force and FSWIMMER, green squiggle arrows indicates the direction of the energy beam. The equation for the total force in each mode is given at the bottom of each panel with explicit positive and negative signs. By definition velocity is in the positive direction (so in the two left panels positive is to the right, in the two right panels positive is to the left). or even be accelerated back towards their origin without addi- ferent missions, however, for any SWIMMER mission to an- tional infrastructure such as a laser array or giant reflector pre- other stellar system, the normal mode will be used for the bulk viously prepared at the destination or launched along with the of the journey. Therefore the mathematical description of nor- spacecraft and allowed to travel ahead. Since SWIMMERF is veloc- mal mode, which uses the positive root and the positive photon ity dependent and drops to zero at zero velocity, a SWIMMER force deserves further consideration. It will also be useful to would never be able to completely come to a stop and reverse in consider the ratio, R, of the force on a SWIMMER in normal the midst of the ISM. Additional means of propulsion such as a mode to the force on an ideal light sail with equal delivered modest rocket engine could be carried along to provide a small power (F=2 P/c), which can be written as: Δv to reverse the direction after most of the braking was ac- complished in destination braking mode, and then the SWIM- MER could begin operating again in tractor beam mode. If the (4) destination were not the midst of the ISM, but a star with an active heliosphere, however, this would be unnecessary. Upon entering the alien heliosphere a SWIMMER could begin des- In Fig. 4 R is shown as a function of velocity for a few values tination braking and reduce its velocity to near zero. Since the of A/P. There is some uncertainty surrounding the structure velocity referred to by Eq. 2 is the spacecraft's velocity with re- and properties of the local ISM, but there is general consen- spect to the interacting medium, within the alien heliosphere sus that a journey to Alph Centauri A will involve passage destination braking would allow the SWIMMER to approach through some combination of the Local Interstellar Cloud, the the velocity of the outward streaming solar wind. The SWIM- Circum-Heliospheric Interstellar Medium and the G Cloud. MER could then coast along with the solar wind until exiting Therefore, a conservatively low ion density of n=0.07 cm-3, con- the heliosphere, at which point its velocity with respect to the sistent with the estimated densities in these clouds, is used in interacting medium would change from nearly zero to whatev- Fig. 4 (eg. [12]). Fig. 4 shows the force initially rising with ve- er the solar wind velocity was (~5 x 105 m s-1 around our sun locity due to the increasing volume of ISM swept out. The force and likely similar for other stars of similar type). At that point peaks at some velocity, vpeak, and then decreases as the ratio of the SWIMMER could operate in tractor beam mode until ap- the change in momentum to change in energy shrinks with proaching its origin. A SWIMMER launched within our heli- faster ion velocities. Due to this initial rise in force with veloc- osphere at velocities slower than the solar wind would operate ity, it may be useful to give SWIMMERs operating in normal in home braking mode, braking with respect to the solar wind mode an initial velocity boost through other means (such as but gaining velocity in the heliocentric frame. A SWIMMER conventional rockets, gravitational assists, particle beam assists which was sent to the edge of our heliosphere which needed to or through home braking SWIMMER mode operations allow- return swiftly could accelerate directly inward in tractor beam ing electric sail-style passive interaction with the solar wind) to mode, unlike electric sails. take advantage of the forces at higher velocities.

All four of these modes of operation could be useful for dif- Taking the derivative of Eq. 3 with respect to v and setting it

120 Vol 72 No.4 April 2019 JBIS SPACECRAFT WITH INTERSTELLAR MEDIUM MOMENTUM EXCHANGE REACTIONS: propellantless interstellar travel

Fig.4 The total force on a SWIMMER in normal operation (FSWIMMER + Fphoton) divided by force on an ideal light sail (Eq. 4), as a function of velocity. Trends are shown for Log(A/P)=6, 4 and 2 in solid, dash dotted and dashed lines respectively. In general, high values of A/P give superior performance relative to light sails. The straight red line indicates (c+v)/(2v) the ratio approached at high velocities. equal to zero gives the velocity of this peak force: is always negative indicating that higher acceleration could be achieved by decreasing the power and the mass associated with (5) the power system.

4 POTENTIAL SWIMMER MISSIONS Larger A/P values give significantly better performance at lower velocities, but trend together as velocity increases, with To illustrate the potential of SWIMMERs for interstellar travel, the force approaching (P/v) x (c+v)/c (the ratio R approaches it is helpful to consider some possible future missions. The sta- (c+v)/(2v), shown by the red line in Fig. 4). This high velocity tionary infrastructure associated with the power beaming sta- limit implies an order of magnitude larger force for SWIM- tion has already been explored in some detail by other authors MERs relative to light sails up to v=c/19 or about 5% c. This regarding light sails, and the associated strategies and techno- also indicates that it is not advantageous to increase the pusher logical obstacles for this component are equally applicable to plate area arbitrarily. If, for instance, a SWIMMER began oper- SWIMMERs. Therefore the primary focus of this investigation ation with an initial velocity of 106 m s-1 with Log(A/P)=3, then is on the SWIMMER itself rather than its remote power source. increasing A by a factor of 100 will not dramatically increase In general, the specific power of the onboard SWIMMER pow- the total force, but significantly increasing the pusher plate area er converter is an important parameter as it determines the would presumably require an increase in the spacecraft mass mass devoted to the onboard power systems. Photovoltaic cells and thus a net decrease to the acceleration. It might even be have fairly poor specific power, ~80 W kg-1 in space based ap- useful to adjust pusher plate area en route by discarding bits of plications [13]. Although technological advances such as in- the pusher plate as the SWIMMER reaches higher velocities. flatable solar arrays might improve their specific power, even Also note that increasing A/P by simply reducing the power these foreseen developments may not increase photovoltaic will increase R, as seen in Fig. 4, but it decreases the overall specific power sufficiently for use in a SWIMMER. Rectennas force. may be more promising, with near-term estimates of specific power as high as 4 kW kg-1 [13]. Although there is ongoing Finally, note that just as SWIMMERs should not have arbi- work to extend rectennas to the optical regime (eg. [14]), cur- trarily large pusher plates, they also should not be completely rent rectennas are only able to convert light at ~cm wavelengths dominated by the mass of the power system which converts to electrical power. At such long wavelengths however, the dif- beamed light into electrical energy. If the SWIMMER’s mass fraction of the light beam would be too large to provide useful were dominated by the power conversion system, the mass power at interstellar distances without an interstellar highway grows with P. If the power system is considered as a thermody- of booster beams or lenses along the route of travel. Perhaps the namic heat engine, absorbing the remotely beamed energy and ideal power converter would be a simple reflector consisting of converting it to electrical power through a temperature differ- a thin aluminized membrane stretched across an aperture and ential with a heat sink radiating to empty space, then to convert electrostatically curved to a focus by a grid of charged wires more power at the same efficiency, the surface area of the heat behind it. Beamed optical or UV light could then hit the re- sink, S, must increase proportional to P. Exactly how the mass flector and converge towards a focus at the hot side of a heat of the heat sink increases with surface area depends on its ge- engine. The heat engine would need to be very low mass, but ometry, but it is safe to assume m Sγ Pγ where 1 ≤ γ ≤ 3/2. not necessarily efficient in terms of electrical output power to Considering the most optimistic case, m P, the acceleration incident radiative power. In fact, an inefficient heat engine with is given by dividing the force by κP (where κ represents the a relatively hot “cold” side will radiate to space more efficiently specific power, W kg-1, of the power conversion system). The and require a lower mass radiative heat sink, at the cost of re- first derivative of the acceleration with respect to P in this case quiring more energy output from the remote power beam sta-

JBIS Vol 72 No.4 April 2019 121 DREW BRISBIN tion. The practical limits of such a system are not well known TABLE 1 A Summary of Mission Parameters and estimating them is beyond the scope of this paper. Instead, Ark ship it is assumed that a specific power of 4 kW kg-1 is achievable. If heat engines are unable to achieve this, and neither rectennas Vmax 0.020 c 0.014 c nor photovoltaics are able to be sufficiently developed, then the tcruise 260 years 370 years achievable travel times will be longer than anticipated here. For tcruise/tsail 0.33 0.16 the following examples a simplified model of the ISM and the Pdelivered 10 MW 10000 GW heliospheres of the Sun and Alpha Centauri A is adopted. The -3 9 ISM is assumed to be uniform with a density of 0.07 cm , a Mpay 1000 kg 8×10 kg 9 temperature of 7000 K and therefore an electron Debye length Mpower 2500 kg 1×10 kg λD=21.8 m. The heliospheres of both the Sun and α Cen A are Initial pusher plate parameters assumed to have a density of 7.3 cm-3, a temperature of 140000 Summed tether length 4.1×10 9 m 2.0×1016 m K and therefore an electron Debye length of λD=9.5 m. Fur- 10 thermore, the solar wind in both heliospheres is assumed to be Mpusher 7400 kg 3.7×10 kg 5 -1 uniformly streaming outward at a velocity of 5 x 10 m s out Final pusher plate parameters to a distance of 100 AU at which point the surrounding medi- 8 15 um abruptly transitions to a stationary ISM. A summary of the Summed tether length 2.9×10 m 1.0×10 m 9 mission parameters is shown in Table 1. Mpusher 520 kg 1.9×10 kg

4.1 SPACE PROBE RENDEZVOUS AT α CEN A Table 1 Mission parameters for the missions discussed in sections 4.1 and 4.2. The parameter cruiset represents the time to traverse 1 pc A relatively lower mass SWIMMER mission might have the in the ISM, ignoring initial time spent in the sun’s heliosphere as goal of transporting a modest space probe, mpay=1000 kg to well as time spent decelerating near α Cen A. The fraction cruiset /tsail α Cen A and then decelerating to allow gravitational capture compares this cruise time to the time required for an ideal, massless for a permanent orbital space telescope. A modest electrical light sail with an equal payload pushed by an equal amount of power delivered to the SWIMMER of 10 MW is assumed. The power. Mpay, Mpower, and Mpusher represent the mass devoted to the pusher plate will be made up of several long tethers. In prac- payload, the power conversion systems, and the pusher plate tethers tice these tethers will consist of very fine braided filaments to respectively. Note that Mpusher decreases throughout the journey as prevent failure due to micrometeoroid and interstellar dust the SWIMMER discards pusher plate mass. Tether length gives the collision, as described for the electric sail [9]. From a material summed length of all the tethers used in the pusher plate. This can mass standpoint these are considered to be single wires with be converted to a pusher plate cross sectional area by way of Eq. 6. an effective diameter of 30 μm. This is equivalent in material to eight filaments with diameters of about 10 μm. Strategically weak breakpoints in the tethers are included which can be acti- the moment it is unclear how much mass to devote to mpusher, vated by simply increasing the pusher plate spin rate such that however a mass of 7400 kg will be shown to be a useful choice. the centripetal force exceeds the break point capacity. As the The mass for the tethers could be mined in situ from asteroids. SWIMMER reaches higher velocities then, it may leave behind This mass provides for a total summed tether length of 4.1 x 109 mass from the pusher plate. Given the pulsed nature of the m. While this is seemingly a very long tether, it does not in any SWIMMER electric field, the wire tethers should be made out way represent the spatial scale of the SWIMMER, as the pusher of superconducting materials. A full analysis of the material re- plate will be made up of of several thousand tethers, possibly quirements is beyond the scope of this paper, and it will depend splitting off from each other at greater radial distances. The on the necessary current density based on the geometry of the summed length is merely a useful value for determining the pusher plate as well as the timescale of the primer, pull and total cross sectional area in plasmas of different temperatures push stages. For this example MgB2 is used to represent one and densities. possible wire material. MgB2 is a well known super conductor with a density of ρ=2570 kg m-3 and a high critical temperature The SWIMMER begins at rest with respect to the sun near (Tc=39 K) which should passively reach super conductivity be- its creation site by the asteroid belt at 3 AU. Within the Sun’s yond ~5-50 AU depending on its surface emissivity. A single heliosphere the SWIMMER will be able to operate in home charged wire will interact with charged particles passing within braking mode by producing a drag force with respect to the ~λD on either side of it. The total cross sectional interaction solar wind. The pusher plate tethers will not be super conduct- area is given by: ing in the inner solar system, but even while operating totally passively with P=0, a static charge on the plates will produce a (6) significant drag force accelerating the SWIMMER towards the velocity of the solar wind. Based on the summed tether length, where L is the summed length of all the tethers. This cross sec- our SWIMMER will have a total cross sectional interaction tional interaction area is somewhat of an idealization as the area of 7.7 x 1010 m2 within the heliosphere. Using simple code Debye length does not represent a hard cut off where parti- written in Interactive Data Language (IDL) (available upon re- cles suddenly cease to be effected by an electric field, and in quest) the SWIMMER path is iteratively tracked according to regions where tethers intersect, part of their cross sectional ar- Eq. 3 while also introducing a gravitational attraction inwards eas will overlap. Nonetheless it is a sufficient estimate for our toward the sun. After 1.5 years the SWIMMER enters the ISM rough calculations. The mass devoted to this pusher plate will at 100 AU with a velocity of 4.0 x 105 m s-1 2 be mpusher=ρ x L x π rwire . Upon entering interstellar space, the SWIMMER begins The total mass of the SWIMMER ship is comprised of normal mode operations. Simultaneously the ion density drops mpay=1000 kg, mpower=2500 kg (given by the 10 MW supplied and the cross sectional area of our tethers increases by a factor electric power and its assumed specific power) and pusherm . At of λD(ISM) / λD(helio)=2.3. At this distance from the sun the SWIM-

122 Vol 72 No.4 April 2019 JBIS SPACECRAFT WITH INTERSTELLAR MEDIUM MOMENTUM EXCHANGE REACTIONS: propellantless interstellar travel

MER tethers will be superconducting, and 10 MW of electrical are able to passively operate beyond 3 AU. Delivered electrical 9 power are supplied. The SWIMMER will also begin discarding power will be 10000 GW, thus mpower=1 x 10 kg. With a pusher 10 mass from the pusher plate as it accelerates. The optimal rate plate of mass mpusher=3.7 x 10 kg the summed tether length to discard mass will change based on the specific details of any is 2.0 x 1016 m. As before, this pusher plate mass is based on given mass distribution, power and journey length. Analysis optimization of the travel time during the normal SWIMMER of mass discard rate is not necessary for a conceptual under- operation as a function of velocity, ψ and χ. standing of the SWIMMER mission, but is investigated briefly for completeness. To consider this situation the problem can In the initial stage the ark SWIMMER accelerates in home be parameterized by assuming that at any given moment, if braking mode from rest at 3 AU, with the full benefit of the the pusher plate mass is a significant fraction of the total mass, beamed power. In the heliosphere the pusher plate has a cross 17 2 mpusher/mtot >χ, mass will be discarded from the pusher plate sectional interaction area of 3.0 x 10 m . Although it requires until A=ψ Apeak where Apeak is the pusher plate area that cor- relatively little mass, this is, admittedly, very large (~20% of the responds to the A/P value which lets the current SWIMMER Sun's cross sectional area). Care would need to be taken during velocity match vpeak. The parameters ψ and χ were experimen- construction to ensure tidal forces with any nearby asteroids do tally varied over a range of values to find the minimum travel not disrupt the pusher plate. This results in an eight-year jour- time. In this example a minimum 1 pc travel time of 260 years ney to the edge of the heliosphere at 100 AU, where it enters is found with χ=0.13 and ψ=0.53. The SWIMMER arrives with the ISM at a velocity of 1.3 x 105 m s-1. Due to the larger Debye a velocity of 6.0 x 106 m s-1 (0.02 c). Without allowing the push- length, the cross sectional interaction area in the ISM is 6.8 x er plate mass to be discarded en route, the journey would take 1017 m2. Operating in normal mode with ψ=0.070 and χ=0.17 slightly longer at 340 years. For comparison, an ideal light sail it takes 370 years to travel 1 pc, at which point it has a velocity 6 -1 9 dominated by mpay=1000 kg (i.e. ignoring the light sail mass of 4.3 x 10 m s and a remaining pusher plate mass of 1.9 x 10 and assuming perfect reflectivity) pushed with the same deliv- kg. For comparison, this 1 pc long journey through interstellar ered power, would take 790 years to complete the same journey space would require 2300 years for an ideal light sail pushed by and it would not be able to stop at the destination without very 10000 GW and with the same payload and negligible sail mass. complicated optics such as a detachable mirror that sails out ahead [1]. The very large pusher plate of the ark ship allows it to decel- erate even faster than the previously considered space probe. As the SWIMMER approaches Alph Centauri A (α Cen A) If it begins destination braking at a distance of 6500 AU from it begins destination braking. This would begin in nearby inter- Alpha Centauri A, then after 12 years it will reach the edge of stellar space at a distance of ~13000 AU from α Cen A. By this the heliosphere with a velocity of 1.6 x 106 m s-1. Entering the point the SWIMMER has significantly reduced the mass of its heliosphere, the cross sectional area changes as before and the pusher plate to 520 kg, with a corresponding interaction area of SWIMMER continues destination braking with power. After 1.3 x 1010 m2. After 23 years of braking in the ISM, the SWIM- another year the SWIMMER arrives at a a distance of 3.7 AU MER enters the α Cen A heliosphere at a distance of 100 AU and from the star and has braked to escape velocity at 2.3 x 104 m a velocity of 1.1 x 106 m s-1 relative to α Cen A. Again, the veloci- s-1 with respect to α Cen A. Slight variations in the onset of ty of the surrounding medium changes, as does the density. The braking and the applied electrical power will allow it to reach changed Debye length reduces the cross sectional interaction any orbit within the heliosphere in comparable times. The full area of the SWIMMER tethers to 5.4 x 109 m2. Within the heli- journey takes just under 400 years. osphere gravitational attraction toward α Cen A is incorporated into the total force and power is reduced to zero. After a further 5 SUMMARY 1.2 years of passive braking the SWIMMER reaches a distance of 2.8 AU with a velocity of 2.6 x 104 m s-1 with respect to α SWIMMERs represent a new mode of interstellar transport. Cen A. At this distance the SWIMMER velocity is equivalent By disposing of onboard reaction mass they circumvent the to the escape velocity and after another moment of braking the rocket equation, and by exchanging momentum with ions in SWIMMER can either continue braking to eventually enter a the ISM they improve by orders of magnitude over the energy circular orbit or stop braking and enter a highly elliptical orbit efficiency of traditional light sails at relatively low velocities. that will allow it to pass through the inner and outer regions of The key to this momentum exchange is the changing electric the α Cen A system. The full journey takes just under 290 years. field which allows SWIMMERs to create inhomogeneities in This is a significant amount of time for a scientific endeavor, but the surrounding plasma and then push on these inhomogenei- there is good precedent for multi-century science projects for ties to create thrust. SWIMMERS perform exceptionally well at worthwhile investigations (cf. [15-17]). lower velocities, with their advantage over light sails diminish- ing quickly at v>0.05 c. Furthermore, by relying on the ambient 4.2 Ark Ship ISM as a momentum exchange medium, they are quite versa- tile, able to accelerate either away or towards a beamed energy Due to their extremely favorable performance at lower power source, opening up myriad opportunities to serve as one-way and velocities, SWIMMERs would make excellent transporters transport, roundtrips or even immobile statites hovering in sta- for large masses that can take long timescales. This could be tionary positions with respect to the Sun and serving as useful used as the engine of a generation ship or perhaps a transporter waypoints with infrastructure for other potential space trans- for bulk colony materials sent out ahead of time before a fast portation networks. moving, low mass, people transporter arrived. For this exam- 9 ple assume a payload mass, mpay=8 x 10 kg, equivalent to the The examples discussed here only scratch the surface of the Super Orion ship discussed by Dyson (2002) [18]. Since such possible roles for SWIMMERs in our spacefaring future. Their a mission would likely only be attempted after significant tech- characteristics make them ideal for any mission with large nological advances, slightly enhanced material properties are masses in which relatively low velocities are acceptable. They assumed including a specific power of the power conversion are unlikely to be the sole mode of space transport due to their systems of 10 kW kg-1 and superconducting materials which diminishing advantages at high velocities and their structural

JBIS Vol 72 No.4 April 2019 123 DREW BRISBIN complexity which requires onboard power conversion systems eral issues ignored here. Areas of further investigation, include with significant mass. Nonetheless, SWIMMERS will play an the efficiency of the SWIMMER drive in three dimensions; important role in future space exploration and augment oth- the electrical potential and cycle timings during the pulsed er modes of transport. They might, for instance, also be well SWIMMER operation and how they effect the required current suited to aiding the construction of a fast interstellar highway density of the tethers; the expected impact of interstellar dust by transporting massive particle beam stations along with their collisions and redundant tether configurations to avoid cata- fuel supply out to stationary positions between us and our tar- strophic damage from tether breakage and realistic limits on get destinations. These particle stations could be used to swiftly power conversion system capabilities. carry low mass Magsails along the path or augment the power of future SWIMMERs by replacing the stationary ISM with a As our understanding of interstellar travel develops, we corridor of fast moving beamed particles. must face the realization that, not only is it difficult, but there is no one-size-fits-all solution. Where SWIMMERs excel in one The missions analyzed here regard one way interstellar trips. metric, other methods may excel in another. Ultimately our While they do push the limits of current technology by as- best strategy is to develop all possible methods in the hope that suming relatively high specific power electrical systems, very their synergy will provide a means to accomplish our goals. thin mass-produced super conducting wire, and light weight electrical insulators which can resist large potential differences Acknowledgements (as well as very large laser array optics which are addressed in other works regarding light sails) there is no obvious material Thanks to the paper reviewer for insightful comments improv- or theoretical limits which would prevent these missions from ing this work. Drew Brisbin acknowledges support from FON- realization. Future work in this vein will need to examine sev- DECYT postdoctorado project 3170974.

REFERENCES

1. R.L. Forward, "Roundtrip Interstellar Travel Using Laser-Pushed 10. R.M. Zubrin, "Dipole Drive for Space Propulsion", Presented at Lightsails", J. Spacecraft and Rockets, 21, pp.187-195, 1984. Breakthrough Initiatives conference, Stanford, California, April 2018. 2. P. Lubin, "A Roadmap to Interstellar Flight", JBIS, 69, pp.40-72, 2016. 11. Stilfehler, “technique of 4 strand braiding”, Wikimedia Commons file 3. N. Perakis, L.E. Schrenk, J. Gutsmiedl, A. Kroop, M.J. Losekamm, (licensed for sharing and adaptation), https://commons.wikimedia.org/ “Project Dragonfly: A feasibility study of interstellar travel using laser- wiki/File:4_Strand_Braiding.png, last accessed on 18 March 2019. powered light sail propulsion”, Acta Astronautica, 129, pp.316-324, 2016. 12. I.A. Crawford, "Project Icarus: A review of local interstellar medium 4. R. Heller, and M. Hippke, "Deceleration of High-velocity Interstellar properties of relevance for space missions to the nearest stars", Acta Photon Sails into Bound Orbits at α Centauri", Astrophysical Journal Astronautica, 68, pp.691-699, 2011. Letters, 835, pp.L32, 2017. 13. J. Dankanich, C. Vassallo, and M. Tadge, "Space-to-space power 5. D. Spieth, and R.M. Zubrin, "Ultra-Thin Solar Sails for Interstellar beaming enabling high performance rapid geocentric orbit transfer", in Travel – Phase I Final Report", NASA Institute for Advanced Concepts, 51st AIAA/SAE/ASEE Joint Propulsion Conference, 2015. Pioneer Astronautics Inc, 1999. 14. G. Moddel, and S. Grover (eds), Rectenna solar cells, Springer, New 6. D.G. Andrews, and R.M. Zubrin, "Magnetic sails and interstellar travel", York, NY, 2013. JBIS, 43, pp.265-272, 1990. 15. A. Kivilaan, and R.S. Bandurski, "The one hundred-year period for 7. G.A. Landis, "Interstellar flight by particle beam", in AIP Conference Dr. Beal's seed viability experiment", American Journal of Botany, 68, Proceedings, vol. 552, pp.393-396, 2001. pp.1290-1292, 1981. 8. D.G. Andrews, "Interstellar Transportation using Today's Physics", 16. R. Johnston, "World’s slowest-moving drop caught on camera at last", Conference proceedings, American Institute of Aeronautics and Nature News, 18, 2013. Astronautics, 4691, 2003. 17. C. Cockell, "The 500-year microbiology experiment", Microbiology 9. P. Janhunen, "Electric sail for spacecraft propulsion", J. of Propulsion and Today, 95, pp.95-96, May 2014. Power, 20, pp.763-764, 2004. 18. G. Dyson, Project Orion: The True Story of the Atomic Spaceship, Henry Holt and Co., New York, NY, 2002.

Received 13 August 2018 Approved 4 April 2019

124 Vol 72 No.4 April 2019 JBIS JBIS VOLUME 72 2019 PAGES 125–143

ARTIFICIAL INTELLIGENCE for Interstellar Travel

ANDREAS M. HEIN1 & STEPHEN BAXTER2 1Initiative for Interstellar Studies, 27/29 South Lambeth Road, London, SW8 1SZ, United Kingdom; 2c/o Christopher Schelling, Selectric Artists, 9 Union Square 123, Southbury, CT 06488, USA

Email [email protected]

The large distances involved in interstellar travel require a high degree of spacecraft autonomy, realized by artificial intelligence. The breadth of tasks artificial intelligence could perform on such spacecraft involves maintenance, data collection, and designing and constructing an infrastructure using in-situ resources. Despite its importance, existing publications on artificial intelligence and interstellar travel are limited to cursory descriptions where little detail is given about the nature of the artificial intelligence. This article explores the role of artificial intelligence for interstellar travel from an engineering perspective by compiling use cases, exploring capabilities, and proposing typologies, system and mission architectures. Estimates for the required intelligence level for specific types of interstellar probes are given, along with potential system and mission architectures, covering those proposed in the literature but also presenting novel ones. Finally, a generic design for interstellar probes with an AI payload is proposed. Given current levels of increase in computational power, a spacecraft with a similar computational power as the human brain would have a mass from dozens to hundreds of tons in a 2050–2060 timeframe. Given that the advent of the first interstellar missions and artificial general intelligence are estimated to be by the mid-21st century, a more in- depth exploration of the relationship between the two should be attempted, focusing on neglected areas such as protecting the artificial intelligence payload from radiation in interstellar space and the role of artificial intelligence in self-replication.

Keywords: Interstellar travel, Artificial intelligence, Artificial general intelligence, Space colonization

1 INTRODUCTION pected events. Satisfactory schedules are searched based on a timeline model of the spacecraft state and its resources using Robotic deep space exploration and interstellar travel require AI. Interesting areas of future development are collaborative high levels of autonomy, as human intervention is very limited spacecraft and rovers that use sensor webs fusing data from with signals taking years to travel to the probe and back. For various sensors [9, 30, 133]. robotic interstellar travel, autonomy is required for exploring the target star system, developing an infrastructure using local Chien et al. [29] and Hein et al. [75] further explore the role resources, and even colonization [71, 70, 67]. High levels of au- of AI in human space exploration. AI-based mission opera- tonomy in spacecraft are associated with performing cognitive tions scheduling can help crews to interactively schedule their tasks such as image recognition, reasoning, decision-making activities [29]. Managing the spacecraft systems such as the etc. For example, current planetary rovers are able to auton- power subsystem is also considered an area where AI can sup- omously identify scientifically interesting rock formations via port humans, in particular in off-nominal situations. In such a feature recognition and take decisions to analyse them [30, 149, case, AI can perform problem analysis, perform repair actions, 27, 89, 165]. A program that is able to perform such and other and evaluate the impact on future operations [29, 75]. Further- cognitive tasks is referred to as artificial intelligence (AI) in the more, operations on planetary surfaces is another area where following [79]. We will provide a brief overview of the current AI can assist in developing operations plans. literature on AI for space exploration in general and then focus on AI for interstellar travel. Looking into the far future, robotic probes with sophisticat- ed artificial intelligence capabilities have been proposed such An overview of the current state of the art of artificial in- as self-replicating space probes and probes that are capable of telligence in space exploration has been provided in Chien et communicating with extraterrestrials [5, 10, 22, 39, 150, 152]. al. [29] and Chien and Wagstaff [30]. According to Chien and The scenario of an interstellar probe encountering an extrater- Wagstaff [30], the main goals of AI on space probes is to detect restrial intelligence has been explored by Baxter [10]. Brace- and characterize features of interest such as usual and static well [22] proposed an intelligent interstellar probe, a so-called (snow, water, ice, etc.) or unusual and dynamic (volcanic activ- Bracewell probe that is able to perform sophisticated commu- ity, fires, floods, dust devils, active jets); autonomous collection nication with extraterrestrials and contains large amounts of of interesting samples, autonomous creation of environmental knowledge of a civilization. Combining the Bracewell probe maps, on-board analysis of data is desirable for reducing the with a self-replicating capability has been explored by Frei- data that needs to be stored and transmitted, on-board sched- tas [52], O’Neill [126] and Jones [92]. Such advanced probes uling where mission scheduling needs to be adapted to unex- would require levels of artificial intelligence that are similar to

JBIS Vol 72 No.4 April 2019 125 ANDREAS HEIN & STEPHEN BAXTER human intelligence or even superior in broad task categories TABLE 1 Current, near-, mid-, and far-term use cases for AI [46]. An artificial intelligence that is able to perform a broad in space exploration range of cognitive tasks at similar levels or better than humans AI use case Reference is called artificial general intelligence (AGI). Estimates for the Current and near-term advent for AGI differ [46], however, their median is somewhere in the middle of the 21st century. The estimated launch date Detect and characterize features of interest (usual [29, 30] for the first interstellar probe falls into a similar time frame. / unusual; static/dynamic) Given these estimates, it is plausible to assume that AGI and in- Autonomous collection of interesting samples [29, 30] terstellar travel might materialize at similar points in time and Autonomous creation of environmental maps [29, 30] implications of one on the other are worth of being considered. On-board analysis of data [29, 30] The use of AI for maintenance and housekeeping of interstel- Mission operations planning and scheduling [29, 30] lar probes and crewed interstellar spacecraft has been explored Maintenance (problem analysis, perform repair [19, 29, 75] actions, and evaluate the impact on future for the Project Daedalus study [19] and world ships [75]. Pre- operations) cursors for such technologies have been developed in the con- text of using augmented reality and intermediate simulations Mid- and far-term for space stations [97, 98, 101, 100, 99, 163]. Design and construction of artifacts (spacecraft, [52, 54, 70, 81, 55] infrastructure, colonies) Past publications have mostly dealt with the principle fea- In-situ resource utilization [52, 54, 70, 81, 55] sibility of interstellar probes with an AI without providing en- Self-replication [52, 54, 70, 81, 55, gineering details of such a probe. For example, Tipler assess- 150, 152, 5] es the principle feasibility of mind-uploading into an artificial Communication with extraterrestrials [10, 22] substrate and how a fusion-propelled Daedalus-type interstellar probe could transport the AGI to other stars and gradually col- Educate humans (transmission of knowledge) [36, 75] onize the universe [19, 150]. Ray Kurzweil in ”The Singularity is Near” describes nano-probes with AI payloads that could even Categorizing and measuring capabilities of artificial intel- traverse small worm holes for colonizing the universe [103]. The ligence is considered challenging and none of the proposed most sophisticated analysis of AI probes is provided by Brad- frameworks has been generally accepted [78]. Existing tax- bury who introduces the concept of ”Matrioshka Brain” where a onomies categorize artificial intelligence with respect to its large number of spacecraft, producing power for AIs, orbit a star abilities (weak vs strong AI; narrow / general AI, super intel- [24]. Bradbury imagines whole layers of orbital rings around a ligence) [78], working principles [38, 62], internal processes star harnessing its energy, similar to a Dyson Sphere [7]. Hein [80], and embodiment [60]. AI metrics are either task-oriented introduces several potential mission architectures based on AI or ability-oriented [78, 79]. Most existing metrics fall into the interstellar probes with the main objective of paving the way task-oriented category, where the performance of an AI sys- for human interstellar colonization by creating space or surface tem is measured with respect to a task such as playing chess colonies in advance to their arrival [71, 70]. Using AGI to grow and autonomous driving. Such an evaluation is appropriate and raise humans from individual human cells or embryos at for specialized AI systems for specific tasks. By contrast, abil- another star and thereby avoiding the transport of grown-up ity-oriented metrics focus on the set of tasks that would indi- humans has been proposed by Crowl et al. [36]. cate the presence of a more general AI ability, for example, the AI decathlon [159, 3, 79, 120, 119, 139]. Such an evaluation is These mid- and far-term use cases of AI for space explora- appropriate for AI systems that are not characterized by a set tion can be categorized into building artifacts in space, com- of tasks such as cognitive robots, assistants, and artificial pets. municating with extraterrestrials, and growing / educating hu- mans. Building artifacts in space encompasses diverse activities We propose a mix between formal and qualitative frame- such as in-situ resource utilization, design, manufacturing, ver- work elements. Formal approaches permit the generation of ification, validation, and testing, and self-replication. A sum- sufficiently general results that might remain valid, even with mary of the AI use cases for space exploration are presented the large uncertainties associated with future progress in AI. in Table 1. These use cases also apply to the specific case of We specifically use thepragmatic general intelligence metric interstellar travel. [61] for formally comparing different AI capabilities. The qual- itative approaches such as literature surveys of both, the scien- The existing literature on AI and AGI in interstellar travel tific literature and fiction, generation of mission architectures, and colonization seems to be limited to high-level concepts and and design of a generic AI probe, allow for exploring specific there is a lack of a systematic analysis of the role of AI/AGI and scenarios and concepts. synergies with other technologies. This article addresses these gaps from an engineering perspective by analysing the capa- Regarding formal elements of the analysis framework, we bilities of AI, AGI for different interstellar missions, explores argue that any AI-based interstellar mission is based on one synergies with other technologies that could result in radically or more agents. According to Franklin and Gaesser [50], an different mission architectures. Concepts for different AI in- “agent” is a system situated within and a part of an environ- terstellar probes and a generic AI probe design are presented, ment that senses that environment and acts on it, over time, using the methodology of explorative engineering [40, 41]. in pursuit of its own agenda and so as to effect what it senses in the future.” An agent is distinguished from computer pro- 2 ANALYSIS FRAMEWORK grams in general by their autonomous, adaptive nature. We can define an agent more formally as a function π which takes An analysis framework for AI probes is developed, in order to an action history as input and outputs an action. An action compare the capabilities required by such probes for fulfilling history is defined as agent’s actionsa , observations o of the specific mission objectives. environment, and rewards r:

126 Vol 72 No.4 April 2019 JBIS ARTIFICIAL INTELLIGENCE for interstellar Travel

(1) We distinguish between four types of AI probes:

The history up to a point in timet can be abbreviated by Explorer . • capable of implementing a previously defined science mission in a system with known properties (for instance According to Legg and Hutter [106], the utility function V, after remote observation); [10, 77] which expresses the expected total reward E for an agent π and • capable of manufacturing predefined spare parts and environment µ over its entire lifetime T is: components; • Examples – the Icarus and Daedalus studies [19].

(2) Philosopher • capable of devising and implementing a science program in unexplored circumstances; Goertzel [61] extends this framework by adding functions • c apable of original science: observing unexpected phe- that indicate the complexity of goals and environments in which nomena, drawing up hypotheses and testing them; the agent operates, thereby formalizing pragmatic general intel- • c apable of doing this within philosophical parameters ligence, defined as achieving complex goals in complex environ- such as planetary protection; ments. The expected goal-achievement is defined as: • c apable of using local resources to a limited extent, e.g. manufacturing sub-probes, or replicas for further explo- ration at other stars. (3) Founder • c apable of using local resources on a significant scale, with the interaction sequence m1a1o1g1r1m2a2o2g2r2…, where such as for establishing a human-ready habitat; m is a memory action, and T=i (i,…,t). Each finite interac- • c apable of setting up a human-ready habitat on a target tion sequence I(g,s,t)=aorg(s:t) with gs corresponding to a goal g, is object such as part of an embryo space colonization pro- mapped by each goal function to a ‘raw reward’ rg (I(g,s,t)) [0,1], gramme; perhaps modifying conditions on a global scale indicating the reward of achieving the goal during that interac- (terraforming) [49, 70, 75]. tion sequence. The agent’s total reward rt is the sum of the raw rewards from all goals obtained at time t, where the symbols for Ambassador these goals appear in the agent’s history before t. • equipped to handle the first contact with extraterrestrial intelligence on behalf of mankind, within philosophical According to Goertzel [61], the pragmatic general intelli- and other parameters, e.g. obeying a Prime Directive and gence of an agent Π, relative to the distribution ν over envi- ensuring the safety of humanity [10, 22] ronments and the distribution γ over goals, is its expected per- formance with respect to goals drawn from γ in environments A more detailed description of each of the probe types is drawn from ν. given in the following. Besides the scientific literature, the sci- ence fiction literature has equally elaborated on different probe (4) types and will also be considered.

3.1 Explorer This formal framework of pragmatic general intelligence al- lows for a comparison of the intelligence of agents as a sum of An Explorer probe is an extension of the model of modern-day the expected rewards these agents would obtain with respect automated space probes, which have limited on-board AI and to environments and goals. For example, an agent that is ex- well-defined missions. Because of remoteness from Earth mod- pected to obtain a reward 0.2 in a single environment µ with ern probes are capable of some independent decision-making. respect to goals g1 and g2 has a total Π of 0.4, whereas an agent Probes may put themselves into ‘safe’ modes in case of navi- that is expected to obtain a reward 0.4 for goal g1 in a single gation failures or other issues; Mars rovers will stop before or environment µ has the same value for of 0.4. Hence, the metric back up from unexpected obstacles. But essentially, in the event allows for taking both, breadth and specificity of an agent’s per- of novelty, the probes wait for further orders from an Earth- formance into account. bound mission control.

Apart from this quantitative framework for comparing an Because of light speed time delays, this would not be an op- agent’s capability, we further use a qualitative maturity scale tion with a probe like Icarus and Dragonfly to Alpha Centauri for analysing task-specific capabilities and general capabilities [73, 69, 109, 110, 130]. During the long flight, the AI would need with respect to AI probe missions, drawing heavily from Her- to deal with routine systems operations like course corrections nandez-Orallo [79, 78] and Hein [68]. The results of this qual- and communications, and also maintenance, upgrades, and itative analysis are presented in Section 4. dealing with unplanned incidents like faults. On arrival at Al- pha Centauri, coming in from out of the plane of a double-star 3 ARTIFICIAL INTELLIGENCE PROBE CONCEPTS system, a complex orbital insertion sequence would be needed, followed by the deployment of sub-probes and a coordination AI probes can be distinguished with respect to their objectives. of communication with Earth [12]. It can be anticipated that It could be classic exploration where AI serves only as a means the target bodies will have been well characterised by remote for realizing autonomous exploration of a star system. It could inspection before the launch of the mission, and so objectives also be more sophisticated such as preparing an infrastructure will be specific and detailed. Still, some local decision-making for human colonization or even entirely AI-based colonization. will be needed in terms of handling unanticipated conditions,

JBIS Vol 72 No.4 April 2019 127 ANDREAS HEIN & STEPHEN BAXTER equipment failures, and indeed in prioritising the requirements as there are cases where on-board manufacturing could lead (such as communications) of a multiple-sub-probe exploration. to a globally lower value on pragmatic general intelligence. Imagine the case where manufacturing the aperture leads to 3.1.1. Technology and capabilities a shift in the centre of mass of the spacecraft which leads to a higher consumption of fuel, leading to a shorter lifetime of the The AI could be based on already existing technologies such probe and hence lower performance. However, we can imagine as deep learning [105, 136] for feature recognition and genetic a special case where an action ai enabled by on-board manu- algorithms for task sequencing. Such an AI would not be con- facturing substitutes for an action aj. The substitution leads to a sidered an agent according to the definition of Franklin and higher reward ri > rj but has no other effect on the entire history Graesser [50], where a distinction is made between programs aor1:T of the agent from its first cycle to its end of lifeT . For this that just interact with the environment and agents that show a case, inequality 6 would hold. level of autonomy and adaptability with respect to the environ- ment. Using a pre-trained deep learning algorithm would have However, on-board manufacturing or other means of ex- a limited ability to adapt to a changed environment due to its tending the set of actions would not change the fundamental dependency on large training data sets. The genetic algorithm’s limitation that the available set of actions is pre-defined via the performance depends on a carefully crafted set of objective training data and the system would perform poorly in environ- functions, which are hard-wired and not changed during the ments that have no resemblance with the training data. mission. The general problem of a machine that constructs some- Consequently, using the pragmatic general intelligence met- thing has been treated by von Neumann [162] and Myhill [122] ric from Section 2, we claim that the reward of such an AI for µ with universal constructor theory. A universal constructor is a is close to 0 whenever the environment does not resemble the machine Ma that can construct another machine Mb given an test data set. Keeping things simple, we can define a distribu- instruction I: tion of environments νsolsys, which represents the distribution of environments within the solar system. We assume that any (7) environment that is sufficiently outside this distribution results in a reward close to 0. Hence, the more ν differs from νsolsys, Π where “ ” indicates the inputs to a construction process on for πexplorer will approach 0. We can express the similarity of the the left side and the created object on the right side. The simple distribution of environments by a similarity function sim with constructor would be equipped with an initial set of instruc- sim(ν1, ν2) = 1 for ν1 = ν2 and sim(ν1, ν2) < 1 for ν1 = ν2. With tions to build infrastructure elements, instruments, replace- the similarity function approaching 0, the pragmatic general ment parts, etc., using given instructions: intelligence of the explorer probe would reach 0. (8)

Ma could be a 3D-printer or any other machine for manu- facturing. (5) 3.1.2 Mission architectures A more advanced Explorer probe could make use of on- board manufacturing capabilities for creating mechanisms, An Explorer type probe would arrive in the target star system instrument components, and tools for having the flexibility to and start its exploration program either using its pre-existing react to unexpected situations. The existing literature on using hard- and software or could use its capabilities of modifying or manufacturing technologies in space can serve as a source of manufacturing hard- and software components, depending on inspiration [51, 35, 141, 128, 127, 84, 155]. One possibility is the encountered situation. A standard mission architecture for to carry bulk material stocks and a 3D-printer to manufacture an Explorer type probe is shown in Fig.1. components during its trip and during exploration [51]. Used components could be recycled to close the material loop once 3.2 Philosopher components are no longer needed. Existing and near-term in- space manufacturing technologies rather focus on manufac- In contrast to the Explorers with their specific and well-de- turing structural elements [141] using bulk material sent from fined missions, a Philosopher probe is capable of supporting Earth, processed in-situ materials on planetary surfaces [128], an independent, open-ended exploration strategy. This may or on small bodies [42]. include devising and implementing its own science and explo- ration programme from goal- setting to execution and exploit- On-board manufacturing capabilities would increase the ing local resources to manufacture, for example, subsidiary flexibility of the probe, i.e. it would increase the space of poten- equipment, sub-probes, or even replicas of itself for further tial actions with respect to an observation. An Explorer probe interstellar exploration. without on-board manufacturing would have a set of actions (a1, ...an) to choose from and an Explorer probe with on-board manufacturing a set of (a1, ...am) actions, where m > n. For ex- ample, the probe could perform an action to manufacture a larger aperture for its telescope and allow for higher-resolution observations. The following inequality does not hold in gener- al for Explorer probes with on-board manufacturing πexplorerobm and explorer probes without πexplorer

(6) Fig.1 : Star system exploration via Explorer probe.

128 Vol 72 No.4 April 2019 JBIS ARTIFICIAL INTELLIGENCE for interstellar Travel

3.2.1 Philosopher probes in fiction tion, which means that:

An interesting advanced probe of the Philosopher type, depict- (9) ed in Greg Bear’s 1990 novel Queen of Angels [13], is AXIS (for Automated eXplorer of Interstellar Space), a probe to Al- which means that the Philosopher AI based on a Gödel ma- pha Centauri. An advanced on-board ‘biologic thinker system’ chine has a practical general intelligence, which is equal or (Chapter 4) can design its own science programmes at the larger than that of the Explorer: target. As an example, AXIS observes circular structures on a planetary surface which it hypothesises are artefacts of ETI, a (10) hypothesis it later disproves. An example of a wider Philoso- pher-class probe strategy is Tipler’s [151] suggestion that the The Philosopher AI would not only be able to manufacture use of self-replicating von Neumann machines [162] as probes artifacts as the Explorer but go a step further in actually de- could reduce the costs of a large-scale interstellar exploration signing artifacts during its mission. Using the notation from programme drastically – the originating culture need only bear von Neumann [162] and Myhill [122], a designing machine is the costs of sending out the first probe, and allow descendants a machine Mc that creates instructions Ii: constructed of local resources to explore the Galaxy step by step. As a near-term example of this idea, Freitas [52] described (11) a Daedalus probe with a self-replicating payload to establish an industrial infrastructure that allows for building another Programs that can synthesize designs are numerous and Daedalus self-replicating probe. However, Freitas did not pro- already exist today, for example, for generating complex ge- vide details about the AI required for this task. ometric shapes from geometric primitives [28]. For complex soft- and hardware, a design Gödel machine [72] could be im- Philosopher-class probes are capable of devising and im- agined, where the machine analyses its environment (available plementing an open-ended exploration program. Philoso- resources) and a set of design requirements (provide an air- pher-class probes may also be particularly useful on pioneering tight volume), synthesizes a set of designs (aluminium hull) voyages to develop necessary infrastructure for follow-up mis- and assesses its feasibility. Feasibility is assessed via simulations sions. In “StarCall” [12] a smart probe called Sannah III is sent or testing. We can imagine that the machine conceives small on an eighty-year mission to Alpha Centauri, using for acceler- prototypes to test key feasibility areas with minimal expendi- ation a mass-beam propulsion system in Earth orbit, and decel- ture of available resources before embarking on the construc- erating using an on-board inertial-confinement fusion drive. tion of the real artefact. Once at Centauri, the mission is to construct another mass- beam propulsion station from local resources; future probes, An open research question concerning self-improving AI is with no need to carry fuel for deceleration, will be capable of the safety problem [46]. If the AI can modify itself and specifi- delivering cargoes orders of magnitudes larger. cally its utility function, how can we assure that it will not take harmful actions? 3.2.2 Technology and capabilities 3.2.3 Self-replicating probes One key aspect of the Philosopher probe will be how the AI can learn from the data at the target star system to adapt and learn Self-replicating probes have been proposed in the literature for from new findings. It is quite obvious that optimal problem solv- decades [52, 55, 54, 116]. In the following, we refer to the the- ers such as practical implementations of AIXI [87, 106, 157, 158] ory of self-replicating machines which can be found in the the- and the self-referential Gödel machine could be used [146, 145]. oretical computer science literature such as [4, 122] and [162]. Another possibility is to use genetic algorithms to automatically generate programs adapted to new findings at the star system [14]. A self-replicating machine can be understood as a machine M, e.g. a Turing machine or equivalent that accepts some input To take the Gödel machine as an example, it consists of two data I that includes a description of M and is able to construct parts. The first part is a program that interacts with its environ- M. However, this is not self-replication, as the instruction is not ment. The second part includes a proof searcher that searches copied. Hence, we introduce two machines Ma and Mb where for proofs that a modification to the Gödel machine is expected Ma creates a copy of I and Mb creates a copy of Ma and Mb. The to yield higher rewards during its lifetime. Once such a proof is input data I needs to include a description of Ma and Mb. found, the modification is implemented and the Gödel machine modified. Schmidhuber [135] argues via his Global Optimality (12) theorem that the Gödel machine performs optimally in the set of environments ν and is not restricted by the Free Lunch theorem. (13) The Gödel machine can modify any part of its code, including the proof searcher itself and the utility function which sums up the Where “ ” can be interpreted as ”creates”. Hence, combining rewards. Hence, a Philosopher probe based on a Gödel machine the two yields a self-replicating machine: would, in principle, not be bound by the limitations of the Explor- er probe and could modify its soft- and hardware with respect to (14) a specific environment and even set its goals. A version of the Gödel machine for solving design problems has been proposed Although the existence of a self-replicating machine has by Hein and Condat [72]. Such a design Gödel machine could be been formally proven, an actual construction turned out to be used for building infrastructures in the target star system. more difficult. Programs that can take their own code as in- puts have been around for years and are called “Quines” [82]. The Gödel machine only switches to a modified version if it Self-replicating machines based on cellular automata have can prove that it would yield better results on the utility func- been developed but turn out to be computationally very ex-

JBIS Vol 72 No.4 April 2019 129 ANDREAS HEIN & STEPHEN BAXTER pensive, as they simulate the assembly or the machine from elementary parts [140]. Robotic self-replicating machines have been proposed by Zykov et al. [170, 171], Yim et al. [167], and Griffith et al. [64]. However, they use prefabricated parts that are assembled to form copies of themselves. Several NASA NIAC studies [107, 31, 20, 153] have concluded that at least “cranking” self-replicating machines are feasible. Nevertheless, for any practically useful application, physical self-replicating machines would need to possess considerable computing pow- er and highly sophisticated manufacturing capabilities, such as Fig.2 Star system exploration via multiple sub-probes. described in Freitas [52, 55, 54], involving a whole self-repli- cation infrastructure. Hence, the remaining engineering chal- Self-replicating star system exploration lenges are still considerable. Possible solutions to some of the A mass-efficient exploration strategy would comprise the pro- challenges may include partial self-replication, where complete duction of self-replicating probes using in-situ resources of the self-replication is achieved gradually when the infrastructure is star system. Only the mass for the initial spacecraft is required, built up [117]. Furthermore, the development of generic min- thereby exponentially reducing the required mass for explora- ing and manufacturing processes, applicable to replicating a tion. The success of such an exploration strategy will depend on wide range of components, automation of individual steps in the ease of identification, reachability, and extraction of resourc- the replication process, and supply chain coordination. es in the star system. Various AI architectures can be imagined. Computing on the sub-probes could be limited and the main The main challenge for the AI of such a probe is rather how to probe would be responsible for sophisticated computations. adapt the design of the probe to the given resources in a star sys- Such an architecture might be more efficient but also more risky, tem. Depending on the chemical composition and reachability in case of the failure of the main probe. Alternative architectures of resources in the star system, different mining and manufac- could be based on distributed computation between sub-probes turing processes are needed. For example, resources on asteroids, and the main probe, where the computing power of sub-probes comets, exoplanets, and might be quite different in and the main probe would not differ significantly, which would composition and ease of mining them [96, 15, 114]. Using min- increase the reliability of the overall system. However, such an ing and manufacturing processes that are applicable to a broad architecture would require the replication of computing hard- range of resources would significantly facilitate the challenge, ware in-situ, which might be difficult to achieve. e.g. sintering can be applied to a broad range of regolith material, whereas high purity metals and alloys require highly specialized processes, which are limited to a specific type of metal and alloy. However, products from general processes might suffer from lower performance characteristics compared to products from a highly specialized process, e.g. in terms of tensile strength.

A special case of self-replicating machines are self-replicating machines that improve on each generation. Myhill [122] pro- vides an existence proof for such machines with the properties: where ”<” indicates that the machine on the right has larger Fig.3 Star system exploration via self-replication.

(15) Adapted biome creation Another mission architecture for a Philosopher probe could theorem-proving capabilities than the machine on the left and consist of the preparation of habitable planets for subsequent i indicates the generation and settlement. A crucial element for human habitability is the where a machine produces a machine with greater capabilities existence of a human-compatible biome, i.e. microorganisms [37], which is vital for human survival. In case the exoplanet (16) is sterile, such a biome could be engineered, taking the local environmental conditions into consideration. For example, a of the subsequent generation. Hence, in principle, machines higher level of stellar radiation might lead to different envi- that create improved versions of themselves in a sequence are ronmental pressures on the biome than on Earth, leading to possible. Myhill sees implications of this proof for biology, a biome which is no longer sufficient for human survival. -En where it implies that a finite genetic code opens up the possi- gineering and cultivating an adequate biome would be a task bility of open-ended evolution. The difficulty is rather, as for which would require sophisticated AI capabilities. The corre- self-replicating machines in general, the practical implementa- sponding mission architecture is shown in Fig. 4. tion of such machines.

3.2.4 Mission architectures

Surface exploration, including astrobiology Similar to an Explorer type probe, the most basic mission archi- tecture for a Philosopher type probe would consist of the probe arriving in the star system, as shown in Fig. 2 and deploying a number of sub-probes for exploration. The difference to an Ex- plorer probe is that the exploration strategy is developed in-si- tu, depending on the observations the probe will make. Fig.4 Creation of adapted micro biome for future colonization.

130 Vol 72 No.4 April 2019 JBIS ARTIFICIAL INTELLIGENCE for interstellar Travel

World model creation ‘seedship’ colony strategy. Crowl et al. [36] gave a recent sketch Another possible Philosopher probe objective could be the of possibilities for ‘embryo space colonisation’ (ESC). The pur- generation of a ‘world model’, as shown in Fig. 5. A world pose is to overcome the bottleneck costs of distance, mass, and model [65] in the context of AI is similar to a mental model in energy associated with crewed interstellar voyages. Crowl et al. humans in which they can perform reasoning without directly [36] suggested near-term strategies using frozen embryos, and acting on the real world. Here, we can think of world models more advanced options using artificial storage of genetic data as either such AI mental models that are transmitted back from and matter-printing of colonists’ bodies, and even ‘pantropy’, the star system to the solar system and could allow for running the pre-conception adaptation of the human form to local con- experiments and simulations. In other words, world models ditions. Hein [70] previously explored the possibility of using could be used for virtually exploring the star system. AI probes for downloading data from probes into assemblers that could recreate the colonists. Although appearing specula- tive, Boles et al. [18] have recently demonstrated the produc- tion of genetic code from data.

A seedship’s AI would, at a minimum, need to create a hu- man-ready habitat from local materials – though more ad- vanced options up to terraforming could be considered. And, crucially, it must raise the first generation of colonists to adult- hood, and perhaps beyond, without adult-human support.

Fig.5 Creation and submission of world model. 3.3.2 Founder probes in fiction

Traveling AI In fiction embryo/genetic space colonization had been hinted Another possible mission architecture consists of a traveling AI, at as long ago as 1930, by Stapledon in Last and First Men [144]. as shown in Fig. 6. Once the target star system has been ex- Some billions of years in the future, the Last Men on , plored, the AI which has been interacting with this environment threatened by solar destabilization (p.238), disseminate ‘among could transmit a copy of itself back to the solar system. Alterna- the stars the seeds of a new humanity’. These will ‘combine to tively, an AI could be transmitted to the star system. This would form spores of a new life, and [will] develop, not into human be interesting, in case the evolution of AI in the solar system is beings, but into lowly organisms with a definite evolutionary advancing quickly and updating the on-board AI would lead bias toward the essentials of human nature’. (Earlier in this saga to performance improvements. Updating on-board software on the genetic adaptation of human species to new environments, spacecraft is already a reality today [45, 111]. Going even a step on Venus and Neptune, was also depicted.) further, one can imagine a traveling AI which is sent to the star system, makes its observations, interacts with the environment, Clarke’s 1986 novel The Songs of Distant Earth [34] contains and is then sent back to our solar system or even to another a classic modern description of seedship colonization, with a Philosopher probe in a different star system. An AI agent could typically elegant summary of its challenges. Driven by the im- thereby travel between stars at light speed and gradually adapt pending nova explosion of the sun, a ‘seedship’ carrying ‘gene to the exploration of different exosolar environments. patterns’ (Chapter 2) colonized an Earthlike planet called Thal- assa, fifty light years from Earth. The first generation of Thalas- sans was manufactured and raised by machines. The ship had to ‘rear these potential humans, and teach them how to survive in an unknown but probably hostile environment. It would be useless – indeed, cruel – to decant naked, ignorant children on to worlds as unfriendly as the Sahara or the Antarctic. They had to be educated, given tools, shown how to locate and use local resources. After it had landed and became a Mother Ship, it might have to cherish its brood for generations. Not only humans had to be carried, but a complete biota.’ (p13). Seven hundred years later, ‘the Mother Ship [was] the oldest and most revered monument on the planet’ (p14). Vinge’s Long Shot Fig.6 Traveling AI between probes in different star systems. (1972) [160], perhaps more realistically, hints at the challenges posed even by the journey component of such a mission. When 3.3 Founder the Earth is threatened by a lethal increase in solar luminosi- ty, a 10,000-year embryo space colonization mission to Alpha A Founder probe is capable of much more ambitious missions, Centauri is hurriedly mounted as a last-resort species survival including the significant modification of its environment, and option. The story is told from the point of view of the on-board perhaps even establishing human colonies. Not only does the AI, called Ilse. Ilse is trained in Earth orbit in such disciplines Founder need the capability to collect and analyse data such as as orbital manoeuvres and planetary survey, manages the long the Philosopher but also the capability to deliberately alter its mission itself, observes the target stars and selects a planet for environment and to verify that the conceived interventions and landing, and survives a final drastic atmospheric re-entry. designs actually work. Hence, sophisticated simulation, opti- mization, and reasoning capabilities are required. But during the long journey component failures degrade Ilse’s mentation and memory, to the extent that she struggles 3.3.1 Founder probes in the literature to complete her tasks, and even forgets the primary mission. A last-resort backup memory enables the nurturing of the The classic application of a Founder-class probe may be the embryos to go ahead – and this convincing story tantalisingly

JBIS Vol 72 No.4 April 2019 131 ANDREAS HEIN & STEPHEN BAXTER ends before the AI’s next great challenge: raising the first gener- curred to anybody. There wasn’t any reason to. We’ve carried on ation of colonists. Crowl et al (perhaps optimistically) suggest- that way ever since. You’ll get used to it’ (p129). Thus they have ed it would be sufficient to use androids as surrogate parents: naturally evolved a post-scarcity society. AIs embodied to enable physical contact, and equipped with ‘a When more conventional follow-up missions are sent to type of expertly programmed expert system, with sophisticated Centauri, Hogan hints at still deeper cultural clashes between natural language abilities’. We may, however, need a more com- ‘normal’ folk and the seed-grown: ‘”It’s not really their fault be- plete understanding of what contribution other human beings cause [the seed-born are] not really people like us ...”’ (p44). make to our development from infancy before we can be sure There may even be religious prejudices. The ship carries an -or how to supplant natural parenting with surrogates. dinance proclaiming that the seed-born have souls – rather as the sixteenth-century Popes had to decree that the inhabitants This contribution may even include a biological input. The of the New World had souls, like Europeans (p44). seedship would need the capability of synthesising far more than ‘human’ cells. The human body contains ten thousand In terms of more advanced seeding technologies, the term times as many microbes of specialised kinds as eukaryotic hu- ‘pantropy’ [137], meaning the pre-conception manipulation of man cells; together this ‘human microbiome’ has a gene set far human stock to adapt it drastically for survival in novel envi- larger than the human. Furthermore, the development of the ronments, was coined by SF writer James Blish in his Seedling microbiome in an individual’s body is not well understood; Stars stories (1952-6, coll. 1957) [17] (though Blish acknowl- perhaps the microbes are transmitted from others, like infec- edged Stapledon’s prior exploration of related ideas). If seed- tions [134] (pp142-3). ships reduce the cost of the travel to a new home, pantropy should reduce the cost of adaptation in a new environment Cultural learning would also need to be assured. As an ex- – pantropy, changing people, will be cheaper than terraform- treme case study, Kemp [95] speculated on how a group of ing, changing worlds. Blish’s story starts with a rogue pantro- infants, isolated from any adult contact at all, might develop. pist who has adapted humans to survive on ; ulti- Is culture hard-wired into our consciousness? In one relevant mately a relatively low-cost, open-ended interstellar ‘seeding example, a sign language spontaneously developed among programme’ (p54) succeeds so well that interstellar colonists an isolated group of deaf children in Nicaragua in the 1970s. return to create pantropes to recolonise ‘the vast and tumbled Necessary if primitive tools might be invented from scratch, desert of the Earth’ (p192). and a lifestyle equivalent to hunter-gathering might emerge, depending on the environment. Sexual differentiation of be- It would appear that on Clarke’s Thalassa [34] the educa- haviour and roles might arise when the first wave of pregnan- tion of the initial generations and cultural transmission from cies occurred – and the first deaths might lead to religious im- the terrestrial precursor went well; the Thalassans understand pulses. Groups limited by the ‘Dunbar number’ of 150 close where they came from, how they got there, and the meaning of personal contacts might emerge, leading to differentiation later visitors from Earth. But this transmission is a challenge, of culture, perhaps even war. Clearly, the contribution of the and perhaps more so if the pantropes’ own physical form is wider environment of our human society to our development drastically modified. A cautionary tale of cultural discontinuity will have to be well understood and replicated if humans are and amnesia is Blish’s story Surface Tension [16], in which a to be manufactured ‘from scratch’ beginning with nothing but crash-landed seedship crew on Hydrot, world of Tau Ceti, hast- genetic data. Perhaps relevant case studies such as that of the ily creates microscopic pantropes to share mud-flat puddles Nicaraguan deaf children could give an indication as to the with a menagerie of algae, diatoms, protozoans, and rotifers. minimum support, physical and psychological, required of any With no knowledge of their origin, even of their basic cosmo- AI surrogate-parent to raise successfully any seedship children. logical context, after sixteen generations the pantropes try to However, Kemp quotes paleoanthropologist Ian Tattersall as break out of a puddle-world into which they don’t seem to fit, predicting that the stranded group would die out, with the chil- delivering the mother of all science-fictional conceptual break- dren developing pathologically without the presence of adults: throughs: ‘the two-inch wooden spaceship and its microscopic that is, deprived of the social nurturing with which we have cargo toiled down the slope towards the drying little rivulet’ evolved. Ethical questions also arise. It would presumably be (p175) [17]. possible to imprint the infants with cultural values of a specific type, such as religious or libertarian. Would it be right to do The embryo space colonization idea remains imaginative- so? After all, such values are transmitted within any ‘normal’ ly alive. In the recent movie Interstellar (2014, dir. C. Nolan), human society from parents and teachers to children. with the Earth becoming uninhabitable due to a blight, em- bryo space colonization through a wormhole was presented as On the other hand, a weaker purposeful conditioning may a ‘Plan B’ to save mankind if a ‘Plan A’, involving the transport cede unanticipated influence to the unusual initial conditions of mature humans, failed. of the seedship colonists. In Hogan’s Voyage from Yesteryear (1982) [83], a limited nuclear war in 1992 triggers a panicky 3.3.3 Technology and capabilities attempt to seed Chiron, a planet of Alpha Centauri. Genera- tions later, humanoid robots with an essentially nurturing role Founder probes will certainly use some form of self-replication, continue to permeate Chironian society – just as during the which has been presented in Section 3.2.3 in order to bootstrap upbringing of the first generation (p125). Just as there was no the infrastructure needed for building up habitats (free-float- obvious hierarchy of human authority presented to the first ing or surface colonies, or even terraforming). However, we can cadre of children in an adult-free world, they have developed expect that the breadth of tasks required for building a habitat a society which continues to be hierarchy-free and self-organ- is much larger and their safety-criticality much higher. It is also ising (p110). And, still more profoundly, the Chironians have reasonable to assume that the complexity of a space colony is continued to regard material goods as free and as abundant as higher than that of a Philosopher probe, for the simple reason they were when provided by the founding robots in the begin- that a space colony would likely also contain a sophisticated AI ning: ‘the idea of restricting the supply of anything never oc- for environmental control and maintenance [75]. Furthermore,

132 Vol 72 No.4 April 2019 JBIS ARTIFICIAL INTELLIGENCE for interstellar Travel engineering a proper biome for the ecosystem and fine-tuning ertheless, this would come quite close to the notion of ’telepor- the overall system to the local conditions is a task that would be tation’ [70], as illustrated in Fig. 9. challenging for human engineers.

Hence, it is reasonable to assume that the pragmatic general intelligence of the AI of a Founder probe is equal or higher than for a philosopher probe, given the larger number of goals to be achieved and higher required performance on these goals (rg yields values close to 1 only if a number of strict safety con- ditions are satisfied). Hence, the expected goal-achievement for the Founder is equal or higher than for the Philosopher: Fig.9 On-site production of genetic material via a data to DNA (17) converter.

If we assume that the distributions ν over environments and It is not that outlandish to assume that, for example, stem over goals γ are the same as for the Philosopher probe, we yield: cells could be transported to the star system, DNA data is sent to the converter, DNA is printed out and the stem cell ”repro- (18) grammed”. Using far more advanced forms of bio-printing [121, 90] than exist today, entire organisms could be created This does not exclude that the Philosopher probe AI could on-site. Such an approach would circumvent potential radia- achieve higher raw rewards for individual goals such as devis- tion-related and age-related degradation during transport in ing scientific hypotheses. However, we assume that most of the interstellar space. Philosopher probe’s goals are part of the Founder probe’s goals. We can even imagine that using self-replication technology 3.3.4 Mission architectures and advanced manufacturing systems, a design for an up-to- date digital to DNA converter could be sent to the target star Fig. 7 shows a mission sequence for the Founder probe, which system, the converter would be built by the advanced manu- begins with exploring and harvesting the star system to design facturing system. Hence, a combination of these advanced and manufacture habitats. In-space colonies and surface colo- technologies would allow for significant flexibility in how the nies could be constructed, depending on the judgment of the colonization operation is performed. probe’s AI. 3.4 Ambassador

3.4.1 Ambassador probes in the literature

The idea of using smart space probes as a specific means to make contact with extraterrestrial civilizations dates back to Bracewell [22, 23], who proposed the idea in 1960 as an alterna- tive to the then-nascent ‘conventional’ Search For Extraterres- trial Intelligence (SETI) model (detection of EM signals). Brace- well imagined a culture sending out many minimal-cost probes Fig.7 A Founder probe building habitats in a star system. equipped with artificial intelligence at least at the human level. On encountering a target culture with radio technology, a probe The inhabitants could be transported to the star system via a would initiate contact, perhaps by echoing back native signals. world ship [75, 70]. Due to the extremely high cost of a world ship, two alternatives can be imagined: The genetic material This approach has distinct advantages, at least for a long-lived for creating humans or other organisms is transported via the culture, in a universe in which technological cultures are sepa- Founder probe or a separate probe, as shown in Fig. 8. rated by large distances (Bracewell suggests 1000 light years or more), or, indeed, such cultures are typically short-lived. A lo- cal probe would allow rapid dialogue, compared to an exchange of EM signals which might last millennia. The probe might even be able to contact cultures lacking advanced technology, through recognizing surface structures for example [11]. And if technological cultures are short-lived, a probe, if robust enough, can simply wait at a target star for a culture ready for contact to emerge – like the Monoliths of Clarke’s 2001 [32]. In Bracewell’s model, the probe would need to be capable of distinguishing Fig.8 A Founder probe building habitats and growing a population between local signal types, interpreting incoming data, and of from transported genetic material. achieving dialogue in local languages in printed form – perhaps through the use of an animated dictionary mediated by televi- As an alternative, the genetic material is recreated from data, sion exchanges. In terms of message content, perhaps it would using an advanced version of a digital to DNA converter [18]. discuss advances in science and mathematics with us, or ‘write The latter approach would have the advantage that up-to-date poetry or discuss philosophy’ (p79). DNA data could be transferred at light speed to the star sys- tem. One of the caveats of the first instance of a digital to DNA However, any engagement with an alien culture on behalf of converter is its extremely low efficiency (99.999%) [129]. Nev- humanity might require a sophisticated political and ethical un-

JBIS Vol 72 No.4 April 2019 133 ANDREAS HEIN & STEPHEN BAXTER derstanding. Bracewell suggested his probe might need to han- from those of previous probe types, as the focus of the former dle political complications, such as avoiding rivalries between is on communication, i.e. agent to agent interaction. This inter- contacted groups by selecting a ‘competent worldwide entity’, action can be broken down into performing an action (sending like Earth’s NASA [23] (p79), to speak through. As suggested by a visual signal, moving an object, etc.) and interpreting the ac- Baxter [11], such a probe would presumably need mandates not tion of the other agent. This generic framework would apply to to harm the extraterrestrial intelligence culture by the violation various forms of organisms / Extraterrestrial Intelligence (ETI). of a ‘Prime Directive’, and not to risk harm to humanity, for in- stance by revealing the existence of Earth and its location to a An existing formal framework we refer to in the following is potentially hostile culture; it might choose to conceal or mask its multi-agent reinforcement learning [85, 26, 161, 147, 108]. It is approach trajectory, for example. Using such precedents as plan- an extension of reinforcement learning to the multi-agent case etary protection protocols and the First SETI Protocol, before with two or more agents [26, 161]. the launch of any such probe a publicly debated and agreed pol- icy on the balance between the opportunities offered by contact Using the notation from Busoniu et al. [26], a single agent and the risks posed by exposure could be developed as a protocol reinforcement learning is with X as the finite set to guide the AI in its decision-making. of environment states, U as the finite set of agent actions, as the state transition probability func- Bracewell went into no details of the probe’s AI, beyond tion, and as the reward function. speculating that ‘presumably the computing part need only be the size of a human head, which is, we know, large enough to Multi-agent reinforcement learning is a generalization of store an immense amount of information’ [10] (p79). Brace- the single agent case and called stochastic game. A stochastic well’s argument was developed further over the years [53, 57, game is a tuple with n the number 56, 102, 154]. Tarter [148] speculated on the use of nanotech- of agents, X the finite set of environment states,U i,i = 1,…,n the nology to send out extremely small smart probes. finite sets of actions available to the agents, yielding the joint ac- tion set U = U1 ×…× Un, f:X × U × X [0,1] is the state transition 3.4.2 Ambassador probes in fiction probability function, and the reward functions of the agents. In Clarke’s 2001: A Space Odyssey (1968) [32] the rogue com- puter Hal was effectively an Ambassador. The true mission of Muti-agent reinforcement learning distinguishes between the spacecraft Discovery, to investigate the alien Monolith or- cases where the agents cooperate, compete, and mixed cases. For biting Jupiter (in the movie and sequels; Saturn in the novel), the cooperative case, the reward function of the agents are the was kept secret from the pilot crew of Bowman and Poole, and same (ρi = ρj, i, j 1,…,n). For the competitive case, the reward was known only to a hibernating team of specialists – and to functions of the agents are distinct (ρi ≠ ρj, i, j 1,…,n, i≠j). The the HAL 9000 unit, the on-board AI. The need to perpetuate mixed case is neither fully cooperative nor competitive. this dishonesty caused Hal to break down. But if the crew were incapacitated and contact with Earth lost, Hal himself had been Based on this basic multi-agent framework, we can already instructed to continue the mission of alien contact (pp.98-9). draw a few conclusions for an encounter between an Ambas- sador probe AI and an ETI. Whether or not the ETI is an AI is Clarke’s The Fountains of Paradise (1979) [33], a novel of the secondary for applying the formal framework, however, recent building of a space elevator, features a kind of cut-price Brace- publications in the field of Search for Extraterrestrial Intelli- well probe, a visitor to the solar system called Starglider by hu- gence (SETI) have argued for the case of an alien AI [39, 25]. man observers. Arriving in the 2060s, the probe was launched First, it seems very unlikely that the reward functions of these from a red dwarf system 52 light years away; proceeding by agents are the same, as there is a vast space of possible reward means of gravity assists it has hopped from star to star, taking functions and the probability of the agents having the same re- 60,000 years to reach Sol. When it arrives in the solar system ward function should be very low unless there is some form of Starglider initiates conversation using English and Mandarin universal convergence. It follows that the interaction between acquired from our leakage broadcasts. Starglider’s primary the agents is very likely not cooperative. In case the encoun- function seems to be the acquisition and sharing of informa- ter is between a single Ambassador probe AI and a single ETI, tion: ‘Starglider combines the functions both of Ambassador the stochastic game is necessarily competitive if more than one and Explorer’ (p83). probe AI or ETI are involved, we either have a competitive or mixed case. But Starglider may have a wider agenda of cultural manipu- lation. It reveals ‘almost no advanced technology, and so [has] Drawing from game theory, the case of an encounter of an minimal impact upon the technically-orientated aspects of our Ambassador probe AI with an ETI can be interpreted as the culture’ (p174). But on the other hand it appears purposefully case of coupled learning [161], where each agent attempts to to demolish religion, for example by logically deconstructing model the other agent(s), i.e. their transition function(s) and St Thomas Aquinas: as Clarke puts it, Starglider ‘had put an reward function(s). Depending on the type of game, specifi- end to the billions of words of pious gibberish with which ap- cally zero-sum, general-sum, or coordination game, different parently intelligent men had addled their minds for centuries’ solutions can be calculated by the agents: Nash equilibrium, (p.94). If intentional, this may amount to a very subtle cultural correlated equilibrium, or a co-ordinated joint action. Such manipulation (towards a cautious development of technology solutions cannot always be calculated but have been success- and away from religion?), implying a deep apprehension of our fully applied in practice [161]. culture and a very high level of cognition. The Ambassador probe AI’s actions would correspond to 3.4.3 Technology and capabilities strategies in game theory. We can imagine actions such as ’ob- serve’, ’contact’, ’send message’, ’withdraw’, or even ’self-destruct’, The capabilities of an Ambassador probe AI are quite distinct in case of a hostile ETI. In each time step, the model of the

134 Vol 72 No.4 April 2019 JBIS ARTIFICIAL INTELLIGENCE for interstellar Travel other agent is refined and the next action taken with respect to more intelligent versions of itself. Here we argue that the same the model. Regarding the model for the ETI’s actions, we face line of reasoning on Vingean reflection can be applied to the a principal challenge of predicting the actions of an agent that case of an AI on an interstellar probe encountering a more in- is more powerful than the Ambassador probe AI, which will be telligent ETI. Fallenstein and Soares [48] use backward induc- addressed in more detail in Section 3.4.5. tion as an illustrative example that an agent which is capable of reasoning about improved versions of itself would need the Sending back the interaction history of an Ambassador reasoning capabilities of its improved versions to do so. LaV- probe’s encounter with an ETI could be very useful, as it could ictoire [104] uses Löb’s theorem to show that an AI’s reasoning form the basis for training future agents to interact with the about a more powerful version of itself is unreliable. Several ETI. Even the transmission of an updated AI to the Ambassa- remedies for this ”Löbstacle” have been proposed [47, 168, 58, dor probe could be imagined if the encounter duration might 59]. be stretched to decades and longer due to the duration of trans- mission. There are, though, differences to those settings in the litera- ture compared to the case of an encounter with an ETI. Firstly, 3.4.4 Mission architectures we have good reasons to believe that such an AI would be vastly superior to an AI we have sent to the stars, as it is likely that Mission architectures for the Ambassador probe are likely to such an AI has developed well before we would have devel- resemble those of the Explorer and Philosopher if they are part oped an advanced AI, therefore having had much more time to of an exploration mission. A possible setting is where the Am- evolve. Secondly, the problem is not to predict if modifications bassador AI is stored on an Explorer or Philosopher probe and made to an AI are potentially harmful but to a certain extent tries to identify cues for ETI in the incoming data. The world predict the actions of an ETI. Let’s suppose that we could have model developed by the Explorer/Philosopher AI could also access to the entire code of the ETI. Such a case would happen serve as a source for cues for how to communicate with an ETI. when we receive a signal from ETI, which might turn out to be a program [156]. A program that would be able to prove that 3.4.5 Safety of encounters with an alien AI such a program is safe would need to be at least as powerful as the program it checks. We simply refer to Löb’s theorem for a The basic tasks of an Ambassador probe would be to commu- proof: nicate with an extraterrestrial intelligence, which means first, that it is able to understand signals from such an intelligence, Let X be any logical statement and L(X) be the statement “if and second, it is able to send signals that can be understood ProofSeeker(X) halts, then X”, where ProofSeeker is a program by the intelligence. Such an interaction sequence can be inter- that searches all possible proofs and halts if and only if one of preted in the previously introduced agent-based framework. them is a valid proof of the statement X. Löb’s theorem states However, a particular challenge is to avoid actions that could that for all statements X, if L(X) is provable, then X is provable. be interpreted as hostile or could otherwise have negative con- sequences. Furthermore, recent SETI / SETA publications have It is straightforward to apply Löb’s theorem to the case of conjectured that an extraterrestrial intelligence might not be checking whether an alien AI program is safe. In such a case, biological but an advanced AI itself [25, 39, 1]. Hence, we have we take X as the logical statement “alien program is safe”. L(X) reason to believe that if the Ambassador probe makes contact then translates to “if ProofSeeker(“alien program is safe”) halts, with an extraterrestrial intelligence, such an intelligence might then “alien program is safe””. The ProofSeeker would be the AI not be biological in nature but a kind of machine. More impor- on the Ambassador probe. However, this is in contradiction tantly, it might have more advanced AI capabilities than the to the inferior deductive capabilities of ProofSeeker compared Ambassador probe itself. to the alien program and therefore, ProofSeeker cannot prove that statement L(”alien program is safe”). Intuitively, we would expect that communication with such an advanced AI would be challenging for the Ambassador. We It follows that whatever action the Ambassador AI takes, it will argue that it is possible, at least for a limited formal case to cannot prove that the ETI would react safely, as it cannot pre- prove that it is in general impossible to fully interpret the ac- dict that the alien AI’s actions would be safe in general. The tions of such an AI via an Ambassador probe. One might argue problem of an encounter with alien AI is, therefore, an extreme that the formal case, where we refer to theorem proving, is not case of Vingean reflection, where approaches from the liter- applicable to a real encounter with an alien AI. However, given ature, such as from Everitt et al. [46] (Section 5) that aim at that we use formal theorem proving techniques for verifying containing potentially harmful self-modifications do not apply. computer programs that have to adhere to strict safety stand- For example, the correct specification of the reward function ards, we still think that such an approach would be suitable for for avoiding harmful AI does not apply to alien AI, as if there is shedding light on some fundamental issues regarding the en- a reward function, it has already been specified. counter with alien AI. If AI alignment is already a challenge for AI created by hu- Furthermore, we argue that it is impossible to generally mans, it is very likely that ETI is not aligned, neither with hu- prove that actions the Ambassador would take in interacting man values nor with values of an AI created by humans, e.g. the with the alien AI are “safe”. Ambassador probe’s AI, simply due to the vast space of possible AI designs. Unless there is some form of universal convergence The first case is analogous to the difficulty of “ensuring that of AI designs, it is unlikely that the ETI is similar to the Am- the initial agent’s reasoning about its future versions is reliable, bassador’s AI. even if these future versions are far more intelligent than the current reasoner” [48]. This type of reasoning has been called Although the values of the Ambassador probe’s AI and the Vingean reflection in the literature and may apply to human ETI are likely misaligned, it might still be possible that the en- reasoning about super-human AI as well as AI reasoning about counter does not result in a harmful result for either side. For

JBIS Vol 72 No.4 April 2019 135 ANDREAS HEIN & STEPHEN BAXTER example, empathy might be a characteristic that would allow TABLE 2 AI probe types and capabilities for mutual understanding, where empathy means “the capacity Explorer Philosopher Founder Ambassador to relate another’s emotional state” [166] and not the general Image X X X X prediction of an agent’s actions. Empathy might be linked to an recognition understanding of the other agent’s values and adopting these Hypothesis X X X values [131]. testing Signal To summarize, we have argued that it is in general not pos- pattern X X X X sible for an inferior AI to predict all actions of the superior recognition AI and it is likely that the Ambassador probe AI is inferior to Devise an alien AI. Hence, there is no guarantee that we can predict scientific X X X whether or not such an encounter will be safe. It is even more hypotheses unlikely that the values of these AIs will be aligned, given the Universal X vast space of possible AI designs. Nevertheless, characteristics translation such as empathy, which could still be present, if the AI is able Conversation X to engage in social interactions, could be a key to mutual un- Identify derstanding. X resources 4 ARTIFICIAL INTELLIGENCE CAPABILITIES Conceive design X (synthesis / 4.1 Task-oriented capability evaluation analysis) Resource X In the following, we use the task-oriented approach [78, 79] processing of comparing the performance of the AI of interstellar probes, looking at the set of tasks the probes need to accomplish. This Construction X qualitative approach is in contrast to the more formal approach Verification, we have previously taken. The result can be seen in 2. validation, X testing The Explorer in its basic form has capabilities similar to ex- isting spacecraft for interplanetary exploration. Data collection receiving and decoding signals from extraterrestrials, translat- and processing is performed with large degrees of autonomy. ing them into a language we are familiar with, composing a By contrast, the Philosopher is a probe that is able to conduct message the extraterrestrials are likely to understand, and its science autonomously, including devising hypotheses, experi- transmission. The most critical function is the translation of mental setups or identifying data collection procedures, and hy- the signal and the composition of proper responses. Such con- pothesis testing. We can imagine prototypical forms of such an versations are imagined by Bracewell [22] who assumes a hu- AI that are based on current machine learning algorithms and man-level intelligence for the probe. Of course, we can imagine a library of scientific hypotheses from which new hypotheses basic conversational capabilities of today’s chatbots. However, can be derived by recombination and mutation. This includes for a low-probability but high-risk event such as contact with the identification and analysis of alien life via remote sensing, an extraterrestrial civilization, we expect that much more so- in-situ analysis of celestial bodies, and analysis of signals [10]. phisticated forms of AI are required that are able to handle the subtlety and ambiguity of language. Another important aspect The Founder is expected to undertake extensive construc- is social intelligence, including empathy, as we already men- tion works within the target star system. These could include tioned in Section 3.4.5. self-replication, large communication infrastructures, space colonies, and even terraforming [49, 75]. The required AI It can be seen in Table 2 that the Philosopher’s set of tasks is probe capabilities differ. For example, mining resources in-situ, a superset of the Explorer’s tasks and the Founder’s set of tasks processing, and constructing truss structures are capabilities of is a superset of the Philosopher’s. It can also be seen that the technologies for asteroid mining. However, doing so in an en- Ambassador’s set of tasks is a superset of the Philosopher but vironment that is to large extents unknown seems to be much not a superset of the Founder’s set of tasks. more difficult. Furthermore, the complexity of the systems that are produced influence how sophisticated the AI needs to be. 4.2 Do we need an AGI? This is due to the emergence of unexpected phenomena in complex systems that require improvisation and creativity to We will also briefly touch on the question of whether or not resolve. When it comes to terraforming, the complexity of the AGI is a precondition for certain AI types. For most applica- expoplanet or system is enormous with limited pre- tions, no AGI is required, as their mission objectives are re- dictability. Furthermore, there is probably only a limited failure lated to specific tasks and capabilities for accomplishing them. tolerance for such a system. Therefore, complex systems such Hence, narrow but high-performance AI could, in principle, as space colonies, and very complex systems such as terraform- accomplish these tasks. With reference to Hernández-Orallo able planets require a broad range of capabilities at similar or [78], general abilities that underlie these specific tasks would superior levels to humans. When it comes to embryo space col- be required if the individual tasks would require them, such as onization, capabilities that allow for sophisticated social inter- those for the Founder probe. actions between the AI and the colonists are required. 4.3 Testing AI capabilities Finally, the Ambassador has the capability to initiate com- munication with an extraterrestrial intelligence in addition to Testing AI capabilities prior to an interstellar mission will be the capabilities of the Philosopher. Communication requires mandatory to reduce risks. In the following, different test cases

136 Vol 72 No.4 April 2019 JBIS ARTIFICIAL INTELLIGENCE for interstellar Travel for the tasks introduced in the previous section are presented, TABLE 3 AI probe capabilities and test cases as shown in Table 3. For most of the AI capabilities and tasks, AI capabilities and tasks Test cases test cases in our solar system or simulated environments can Image recognition Simulated images, solar system be imagined. environment AI for interstellar probes are likely to first be tested for so- Hypothesis testing Capturing and analysing data for hypothesis testing in virtual lar system exploration (e.g. objects, interstellar ob- environment, solar system jects [76]), economic development (e.g. asteroid mining [74, environment 143, 116, 117], space manufacturing [141, 128, 127, 81]), and Signal pattern recognition Various simulated signals of colonization (e.g. lunar/Martian base [43, 8, 171], free-floating increased sophistication, test in colonies [126, 125, 124, 91, 6]). solar system environment Devise scientific hypotheses Extensive tests for research within Simulated environments are in widespread use today for the solar system environment testing AI and robotics. We can also imagine the use of ad- versarial machine learning and generative adversarial networks Universal translation Decoding language of various forms (GAN), where agents are pitted against each other, in order to and of various organisms, including artificially generated languages (e.g. self-train [86, 63, 132]. These approaches have recently been using adversarial machine learning used for self-training games [138] and creating art [44]. and generative adversarial networks (GAN) where one agent tries to Although the solar system provides an environment for generate new languages that the testing various AI capabilities, representative conditions under other cannot translate [86, 63, 132]) which an AI for an interstellar probe could be tested are more Conversation Training with various living likely to be found in the outer solar system and deep space, due organisms; Training with artificial to the signal latency, which makes human intervention diffi- agents, e.g. agents that are cult. Nevertheless, the solar system environment, supplement- generated to ”beat” the AI ed by virtual environments, is likely to be the context in which Identify resources Testing in a solar system AI systems are matured before they are sent to the stars. environment (e.g. asteroid mining, planetary surface exploration, 5 DESIGN OF A GENERIC ARTIFICIAL INTELLIGENCE planetary surface habitat design, PROBE space colony construction) and simulated virtual environments In the following, we present a concept for an artificial intelli- Conceive design (synthesis / Testing in a solar system gence probe, based on the assumption that any sophisticated analysis) environment (e.g. asteroid mining, planetary surface exploration, AI will still likely use substantial computing resources, thereby planetary surface habitat design, consuming substantial amounts of energy. The probe concept space colony construction) and has already been presented in Hein [67]. In the following, we simulated virtual environments make the additional assumption that a future human-level or Resource processing Testing in a solar system super-human level AI would consume as much energy for its environment (e.g. asteroid mining, operation as the equivalent energy for simulating a human planetary surface exploration, brain. We think that this assumption is reasonable, given the planetary surface habitat design, large uncertainty regarding which path will lead to AGI and as space colony construction [81]) and simulating a human brain is considered as one possible path- simulated virtual environments way towards AGI [21]. Construction Testing in a solar system environment (e.g. asteroid mining, Today’s supercomputers use power in the MW range, the planetary surface exploration, human brain, by contrast, uses only about 25 W [94], for a planetary surface habitat design, computing power that has been estimated at 1020 flops. The space colony construction [81]) and required power consumption for equivalent computing power simulated virtual environments using today’s or near-future computing hardware can be esti- Verification, validation, testing Testing in a solar system mated to be between 1 MW and 100 GW, depending on how environment (e.g. asteroid mining, far current levels of increase in computing power can be ex- planetary surface exploration, planetary surface habitat design, trapolated into the future. Nevertheless, these figures provide space colony construction [81]) and lower and upper bounds for the power requirements to simu- simulated virtual environments late a human brain, i.e. between 106 to 1011 W.

The large difference between the lower and upper bounds AI probes, first, due to the large amounts of power consumed, translates into a large difference in the scale of the correspond- and second, as waste heat generation due to computing is likely ing power generation system in space. For generating pow- to take place within a small volume. We assume that computa- er on the order of 10 W, a power generation subsystems for tion is running within a rather small volume, in order to min- a 3U-CubeSat would be sufficient. For generating 1011 W, a imize data transfer latencies, as is the case for today’s comput- hundred solar power satellites would be required [112]. Fur- ers. As a consequence, we can expect that large amounts of heat thermore, power generation in interstellar space is more chal- are generated in a small volume. Heat rejection is currently a lenging, due to the absence of stellar radiation. Once the probe major issue for supercomputers [123] and the predominant ap- arrives in the target star system, we assume that power genera- proach for heat rejection is the use of heat pipes, which trans- tion using photovoltaic cells is feasible. port a cooling liquid to the processors and the heated liquid We argue that heat rejection is likely to be a major issue for away from them. The current heat density in supercomputers

JBIS Vol 72 No.4 April 2019 137 ANDREAS HEIN & STEPHEN BAXTER is as high as 10 kW/cm2, about one order of magnitude higher TABLE 4 Mass estimate for AI probe in the 2050-2060 time than inside a rocket engine nozzle [123]. frame Spacecraft subsystem Specific mass Subsystem mass [t] Heat rejection has already been an issue for terrestrial super- computers, the issue is aggravated in space, where heat can only Computing payload 0.025 1017 flops/kg 40 be rejected without mass loss via radiation, requiring large sur- Solar cells (current 1kW/kg 100 face areas facing towards free space. Advanced radiators could technology) reject about 1 kWt/kg thermal power per kilogram in the near future [2, 88, 93]. Consequently, about 100 tons of radiator mass Radiators 1kWt/kg 100 would be required for rejecting 100 MW and 1 ton for 1 MW Other subsystems (50% 20 respectively. Due to the uncertainty of several orders of magni- of computing payload) tude, we use an approach from Weinstein and Adam [163] for Total mass 260 making estimations under large uncertainties. We take the ge- ometric mean of 106 and 1011 W and get 108.5. We round off and get 108, which leads to a radiator mass of 100 tons. initially operated close to the star. In order to maximize One could argue that existing computer architectures are power input from the star and to minimize solar power very inefficient in replicating the function of a human brain, generator mass, the probe should be located as close to resulting in a huge difference in power consumption. Future the star as possible. The minimum distance is constrained computer architectures or working principles of computers by the maximum acceptable temperature for the space- such as quantum computing could have a disruptive effect on craft subsystems and an eventual heat shield that protects power consumption [113]. In order not to exclude this possi- against the heat and radiation from the star. A trade-off bility, we keep the power consumption of a few dozens to hun- between the heat shield mass and the mass savings from dreds of Watts as a lower boundary, in case revolutionary new lower solar power generator mass needs to be made. ways are found to reproduce the function of the human brain. However, note that the AI payload itself is a source of in- However, a more conservative estimate would put the required tense heat and more sensitive spacecraft subsystems such power at dozens to hundreds of MW for Philosopher, Founder, as sensors need to be located distant to the AI payload, and Ambassador type probes and lower values for AI, such as e.g. on a boom. for an Explorer type probe. • AI payload switched off outside star system: AI is switched Regarding the mass of the computing unit, current on- off outside the target star system, as there is no sufficient board data handling systems (OBDH) have a computing pow- power source available for its operation. This may also er on the order of 100 DMIPS per kg. The spacecraft OBDH protect against some forms of radiation damage to the AI from the literature have DMIPS values that are about two or- payload. Nevertheless, proper radiation protection is an ders of magnitude below the values for terrestrial processors. issue, as galactic cosmic rays could still destroy circuits If we extrapolate these values to the 2050 timeframe, we can via impact and the resulting particle shower; expect spacecraft OBDH with a processing power of 15 mil- lion DMIPS per kg. As DMIPS and flops are different perfor- • Large radiators: A large radiator is needed for rejecting mance measures, we use a value for flops per kg from an exist- the heat generated by the AI payload; ing supercomputer (MareNostrum) and extrapolate this value (0.025 1012 flops/kg) into the future (2050). • Compact computing unit: The computer is either su- per-compact or distributed. However, with a distributed By 2050, we assume an improvement of computational pow- system, communication speed becomes an issue and it is er by a factor 105, which yields 0.025 1017 flops/kg. In order to therefore likely that the architecture will be as compact as achieve 1020 flops, a mass of dozens to a hundred tons is need- possible to minimize the time for signals to travel within ed. We assume an additional 100 tons for radiator mass and the payload. with 1 kW/kg for solar cells, about 100 tons for the solar cells. This yields a total mass for an AI probe on the order of hun- Fig. 10 and Fig. 11 show an artist’s impression of an AI probe dreds of tons, which is roughly equivalent to the payload mass with its main subsystems. The design is similar to spacecraft of the Daedalus spacecraft of 450 tons [19]. with nuclear reactors but with important differences. Where the design of spacecraft with nuclear reactors is dominated Table 4 shows the mass estimates for the main spacecraft by large radiators and placing the reactor far away from oth- subsystems and its total mass in a 2050 to 2060 time frame. The er spacecraft subsystems, the AI probe has solar cells and the mass estimate is only valid for the part of the spacecraft that computing payload does not need to be placed as far away from actually arrives at the target star system. other subsystems for radiation concerns.

Due to the large power consumption and heat rejection re- The AI payload is likely to have a cylindrical shape, as it is quirements, the following characteristics for an AI probe can easier for the heat rejection system to have one backbone heat be deduced: channel and then smaller, radial pipes that reject heat from the processing units. The heat is rejected via large radiators. The • L arge solar panels: Unless other power sources such as radiator size may decreases with distance from the payload, as nuclear power is used, the probe will depend on large so- less and less fluid is available for rejection (not shown in im- lar panels / solar concentrators for generating power for age). It is furthermore better to reject the heat quickly. Hence, the AI payload. the larger size of the radiators close to the payload. The radiator is perpendicular to the payload in order to avoid heat radiation • O peration close to target star: The spacecraft is at least from the payload being absorbed by the radiators and letting

138 Vol 72 No.4 April 2019 JBIS ARTIFICIAL INTELLIGENCE for interstellar Travel IMAGES: ADRIAN MANN Fig.10 AI probe subsystems. Fig.11 View from the back of AI probe. the payload face as much free space as possible.

Fig. 11 shows an additional heat shield between the payload section and the radiators, in order to prevent radiative heat transfer from the payload. In order to maximize energy intake from the star, the spacecraft may be located as close as possible to the target star. There is likely to be a trade between distance to the star and other probe objectives. One can also imagine that sub-probes would do the majority of exploration and the AI probe would remain close to the star and perform the ma- jority of the computation-heavy tasks while communicating with the sub-probes.

If the probe operates close to the target star, strong thermal Fig.12 Spacecraft payload mass vs. year of development. radiation and particles from the star impact the spacecraft. In order to avoid heat and particle influx from the star, a heat and radiation shield is needed for protection against these particles. decreases due to the lower payload mass. The shield is located in the direction of the star and shields the payload. Furthermore, under the assumption that an advanced AI payload has equal capabilities for exploration as a human and The spacecraft needs to be constantly maintained and parts the mass required for transporting a human over interstellar replaced or repaired. This is similar to existing terrestrial su- distances is estimated to be about 100 tons [75, 115], the break- percomputers. A system of this complexity very likely needs even point for an AI probe with similar computing power to repair. If the computer is modular, these modules are replaced a human would be somewhere between 2050 and 2060. How- on a regular basis and parts replaced within these modules. We ever, it is clear that the similarities end here, as a human crew can imagine a storage depot of parts and robots that replace would have colonization as an objective and would also require these parts. With more advanced technology available, robots a large number of crew members to survive [142]. that reproduce even very complex replacement parts can be imagined. 7 CONCLUSIONS

The AI payload needs to either be protected against galactic We presented four types of artificial intelligence interstellar cosmic rays during its interstellar cruise or needs to have ap- probes along with their required capabilities and mission ar- propriate counter-measures in place such as self-healing [66, chitectures. Furthermore, a generic design for an artificial 118] and radiation-hardened electronics. intelligence interstellar probe was presented. Based on the ex- trapolation of existing technologies and trends, we estimated 6 WHEN WILL WE BE READY? that the payload of such an interstellar probe that has a sim- ilar computing power as the human brain is likely to have a Under the assumption that during the 2050 to 2090 timeframe, mass of hundreds of tons in the 2050 time frame and a mass computing power per mass is still increasing by a factor of of dozens of tons in the 2060 time frame. Furthermore, esti- 20.5 per decade, it can be seen in Fig. 12 that the payload mass mates for the advent of artificial general intelligence and first decreases to levels that can be transported by an interstellar interstellar missions coincide and are both estimated to be in spacecraft of the size of the Daedalus probe or smaller from the middle of the 21st century. We therefore conclude that a 2050 onwards. If the trend continues until 2090, even modest more in-depth exploration of the relationship between the two payload sizes of about 1 kg can be imagined. Such a mission should be attempted, looking into currently neglected areas might be subject to the “waiting paradox”, in which case the such as protecting the artificial intelligence payload from radi- development of the payload is postponed successively, as long ation in interstellar space and the role of artificial intelligence as computing power increases and consequently launch cost in self-replication.

JBIS Vol 72 No.4 April 2019 139 ANDREAS HEIN & STEPHEN BAXTER

REFERENCES 1. “Real-world Mining Feasibility Studies Applied to Asteroids, the Science Investigation System for Opportunistic Rover Science”. Journal and Mars”. AIAA SPACE 2011, 2011. of Field Robotics, 24(5):379–397, may 2007. 2. R. Adams, G. Statham, S. White, and B. Patton. “Crewed Mission to 28. A. Chakrabarti and K. Shea. “Computer-based Design Synthesis Using Advanced Plasma Propulsion Systems”. In 39th AIAA/ Research: an Overview”. Journal of Computing and Information Science ASME/SAE, 2003. in Engineering, 11(2):021003, 2011. 3. J. R. Anderson and C. Lebiere. “The Newell Test for a Theory of 29. S. Chien, R. Doyle, A. G. Davies, A. Jonsson, and R. Lorenz. “The Future Cognition. Behavioral and Brain Sciences”, 26(5):587–601, 2003. of AI in Space”. IEEE Intelligent Systems, 21(4):64–69, 2006. 4. M. Arbib. “From Universal Turing Machines to Self-reproduction” in A 30. S. Chien and K. L. Wagstaff. “Robotic Space Exploration Agents”.Science Half-century Survey on The Universal Turing Machine. 1988. Robotics, 2(7):eaan4831, 2017. 5. S. Armstrong and A. Sandberg. “Eternity in Six Hours: Intergalactic 31. G. S. Chirikjian. “An Architecture for Self-Replicating Lunar Factories”. Spreading of Intelligent Life and Sharpening the Fermi paradox”. Acta Technical report, NASA NIAC Phase 1 report, 2004. Astronautica, 89:1–13, 2013. 32. A. C. Clarke. 2001: A Space Odyssey. New American Library, page 6. N. Arora, A. Bajoria, and A. L. Globus. “Kalpana One: Analysis and number edition, 1968. Design of Space Colony”. In 14th AIAA/ASME/ASCE/AHS/ASC 33. A. C. Clarke. The Fountains of Paradise. Victor Gollancz, book club Structures, Structural Dynamics, and Materials Conference 7th AIAA/ edition, 1979. ASME/AHS Adaptive Structures Conference, page 2183, 2006. 34. A. C. Clarke. The Songs of Distant Earth. Del Rey, Grafton edition, 1986. 7. V. Badescu. “On the Radius of Dyson’s Sphere”. Acta Astronautica, 36(2):135–138, 1995. 35. R. J. Clinton. “NASA’s In Space Manufacturing Initiatives: Conquering the Challenges of In-Space Manufacturing”. In Design in Plastics 2017, 8. V. Badescu. Mars: Prospective Energy and Material Resources. Springer Detroit, MI; United States, 2017. Science & Business Media, 2009. 36. A. Crowl, J. Hunt, and A. Hein. “Embryo Space Colonisation to 9. S. Bartsch, F. Cordes, S. Haase, S. Planthaber, T. M. Roehr, and F. Overcome the Interstellar Time Distance Bottleneck”. Journal of the Kirchner. “Performance Evaluation of an Heterogeneous Multi- British Interplanetary Society, 65:283–285, 2012. robot System for Lunar Crater Exploration”. In Proceedings of the 10th International Symposium on Artificial Intelligence, Robotics and 37. P. Davies. “Afterword” in G. Benford & J. Benford, editors, Starship Automation in Space (iSAIRAS-10), 2010. century: Toward the Grandest Horizon, pages 301–310. Lucky Bat Press, 2013. 10. S. Baxter. “Project Icarus: Interstellar Spaceprobes and Encounters with Extraterrestrial Intelligence”. Journal of the British Interplanetary Society, 38. H. De Garis, C. Shuo, B. Goertzel, and L. Ruiting. “A World Survey 66(1/2), 2013. of Artificial Brain Projects, Part I: Large-scale Brain Simulations”. Neurocomputing, 74(1-3):3–29, 2010. 11. S. Baxter. “Project Icarus: Exploring Alpha Centauri: Trajectories and Strategies for Subprobe Deployment”. Journal of the British 39. S. Dick. “Cultural Evolution, the Postbiological Universe and SETI”. Interplanetary Society, 69:11–19, 2016. International Journal of Astrobiology, 2(1):65–74, 2003. 12. S. Baxter. “StarCall”. In Obelisk. Gollancz, 2016. 40. K. E. Drexler. “Exploring future technologies”. Doing Science: The Reality Club, pages 129–150, 1991. 13. G. Bear. Queen of Angels. Warner Books, 1990. 41. K. E. Drexler. Radical Abundance. PublicAffairs books, 2013. 14. K. Becker and J. Gottschlich. “AI Programmer: Autonomously Creating Software Programs Using Genetic Algorithms”. arXiv preprint arXiv:, 42. J. Dunn, M. Fagin, M. Snyder, and E. Joyce. Project RAMA: 1709.05703, sep 2017. Reconstructing Asteroids Into Mechanical Automata. 2017. 15. C. A. Beichman, G. Bryden, T. N. Gautier, K. R. Stapelfeldt, M. W. 43. P. Eckart. The Lunar Base Handbook: an Introduction to Lunar Base Werner, K. Misselt, and D. Trilling. “An excess due to small grains Design, Development, and Operations. McGraw-Hill, 1999. around the nearby K0 V star HD 69830: asteroid or cometary debris?” 44. A. Elgammal, M. Papazoglou, and B. Krämer. “Design for The Astrophysical Journal, 626(2):1061, 2005. Customization: A New Paradigm for Product-Service System 16. J. Blish. Surface Tension. Galaxy Science Fiction, 1952. Development”. In Procedia CIRP, 2017. 17. J. Blish. The Seedling Stars. Gnome Press, page number edition, 1957. 45. J. K. Erickson. “Living the Dream – an Overview of the Mars Exploration Project”. IEEE Robotics & Automation magazine, 13(2):12–18, 2006. 18. K. S. Boles, K. Kannan, J. Gill, M. Felderman, H. Gouvis, B. Hubby, K. I. Kamrud, J. C. Venter, and D. G. Gibson. “Digital-to-biological 46. T. Everitt, G. Lea, and M. Hutter. “AGI Safety Literature Review”. arXiv Converter for On-demand Production of Biologics”. Nature preprint, 1805.01109, 2018. biotechnology, 35(7):672, 2017. 47. B. Fallenstein and N. Soares. “Problems of Self-reference in Self- 19. A. Bond. Project Daedalus – The Final Report on the BIS Starship Study. improving Space-Time Embedded Intelligence”. In International Technical report, British Interplanetary Society, 1978. Conference on Artificial General Intelligence 2014, pages 21–32. Springer, 2014. 20. P. J. Boston, P. Todd, and K. R. McMillen. “Robotic Lunar Ecopoiesis Test Bed: Bringing the Experimental Method to Terraforming”. AIP 48. B. Fallenstein and N. Soares. “Vingean Reflection: Reliable Reasoning Conference Proceedings, 699(1):975–983, 2004. for Self-improving Agents”. Technical report, Machine Intelligence Research Institute, Berkeley, CA, 2015. 21. N. Bostrom. Superintelligence: Paths, dangers, strategies. Oxford University Press, 2014. 49. M. J. Fogg. “Terraforming, as part of a Strategy for Interstellar Colonisation”. Journal of the British Interplanetary Society, 44:183–192, 22. R. Bracewell. “Communications from Superior Galactic Communities”. 1991. Nature, 186:670– 671, 1960. 50. S. Franklin and A. Graesser. “Is it an Agent, or just a Program?: a 23. R. Bracewell. The Galactic Club: Intelligent Life in Outer Space. San Taxonomy for Autonomous Agents”. In International Workshop on Agent Francisco Books, page number edition, 1975. Theories, Architectures, and Languages, pages 21–35. Springer, Berlin, 24. R. Bradbury. Matrioshka brains, 2001. Heidelberg, 1996. 25. R. J. Bradbury, M. M. Cirkoivc, and G. Dvorsky. “Dysonian Approach to 51. A. Freeman and L. Alkalai. “First Interstellar Explorer: What Should it SETI: a Fruitful Middle Ground?”. Journal of the British Interplanetary do When it Arrives at its Destination?”. In American Geophysical Union Society, 64(5), 156, 64(5):156, 2011. Fall Meeting, 2017. 26. L. Busoniu, R. BabuŠka, and B. De Schutter. “Multi-agent Reinforcement 52. R. Freitas. “A self-reproducing interstellar probe”. Journal of the British Learning: an Overview”. Innovations in multi-agent systems and Interplanetary Society, 33(7):251–64, 1980. applications-1, 310:183–221, 2010. 53. R. Freitas. “The Search for Extraterrestrial Artifacts (SETA)”. Journal of 27. R. Castano, T. Estlin, R. C. Anderson, D. M. Gaines, A. Castano, B. the British Interplanetary Society, 36:501–506, 1983. Bornstein, C. Chouinard, and M. Judd. “Oasis: Onboard Autonomous 54. R. Freitas and W. Gilbreath. “Advanced Automation for Space Missions”.

140 Vol 72 No.4 April 2019 JBIS ARTIFICIAL INTELLIGENCE for interstellar Travel

Journal of the Astronautical Sciences, 30(1):221, 1982. oriented to Ability-oriented measurement”. Artificial Intelligence Review, 48(3):397–447, 2017. 55. R. Freitas and W. Zachary. “A Self-replicating, Growing Lunar Factory”. Princeton/AIAA/SSI Conference on Space Manufacturing, 35:18–21, 79. J. Hernández-Orallo. The measure of all minds: evaluating natural and 1981. artificial intelligence. Cambridge University Press, 2017. 56. R. A. Freitas. “Interstellar probes – A New Approach to SETI”. Journal of 80. A. Hintze. Understanding 4 AI Types, 2016. the British Interplanetary Society, 33:95–100, 1980. 81. M. Hirai, A. Hein and C. Welch. “Autonomous Space Colony 57. R. A. Freitas. The Case for Interstellar Probes. Journal of the British Construction”. In 65th International Astronautical Congress, Tortonto, Interplanetary Society, 36: 490–495, 1983. Canada, 2014. 58. S. Garrabrant, T. Benson-Tilsen, A. Critch, N. Soares, and J. Taylor. “A 82. D. R. Hofstadter. Gödel, Escher, Bach. Basic Books, 1979. Formal Approach to the Problem of Logical Non-omniscience”. arXiv 83. J. P. Hogan. Voyage from Yesteryear. Del Rey, 1982. preprint, 1707.08747, 2017. 84. R. Hoyt, J. Slosad, T. Moser and J. Cushing. “MakerSat: In-Space 59. S. Garrabrant, T. Benson-Tilsen, N. Critch, A., Soares, and J. Taylor. Additive Manufacturing of ConstructableTM Long-Baseline Sensors “Logical Induction”. arXiv preprint, 1609.03543, 2016. using the TrusselatorTM Technology”. In AIAA SPACE 2016, 2016. 60. B. Goertzel. The Hidden Pattern. Brown Walker, 2006. 85. J. Hu and M. P. Wellman. “Multiagent Reinforcement Learning: 61. B. Goertzel. “Toward a Formal Characterization of Real-World General Theoretical Framework and an Algorithm”. ICML, 98:242–250, 1998. Intelligence”. In Artificial General Intelligence: Proceedings of the 86. L. Huang, A. D. Joseph, B. Nelson, B. I. Rubinstein and J. D. Tygar. Third Conference on Artificial General Intelligence, AGI 2010, Lugano, “Adversarial machine Learning”. In Proceedings of the 4th ACM Switzerland, March 5-8, 2010, pages 19–24, 2010. workshop on Security and artificial intelligence, pages 43–58. ACM, 2011. 62. B. Goertzel, R. Lian, I. Arel, H. De Garis, and S. Chen. “A world survey 87. M. Hutter. Universal Artificial Intelligence: Sequential Decisions Based on of artificial brain projects, Part II: Biologically Inspired Cognitive Algorithmic Probability. Springer, 2004. Architectures”. Neurocomputing, 74(1-3):30–49, 2010. 88. R. Hyers, B. Tomboulian, P. Crave, and J. Rogers. Lightweight, High- 63. I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, Temperature Radiator for Space Propulsion. 2012. S. Ozair, Courville, and Y. Bengio. “Generative adversarial nets”. In Advances in Neural Information Processing Systems, pages 2672–2680, 89. F. Ingrand, S. Lacroix, S. Lemai-Chenevier and F. Py. “Decisional 2014. Autonomy of Planetary Rovers”. Journal of Field Robotics, 24(7):559– 580, jul 2007. 64. S. Griffith, D. Goldwater, and J. M. Jacobson. “Robotics: Self-replication from Random Parts”. Nature, 437(7059):636, 2005. 90. K. Jakab, C. Norotte, F. Marga, K. Murphy and G. Vunjak-Novakovic, G. Forgacs. “Tissue Engineering by Self-assembly and Bio-printing of 65. D. Ha and J. Schmidhuber. “World Models”. arXiv preprint, arXiv:1803, Living Cells. Biofabrication, 2(2):022001, 2010. 2018. 91. R. Johnson and C. Holbrow. “Space Settlements: A Design Study”. 66. J. W. Han, M. Kebaili, and M. Meyyappan. “System on Microheater for Technical report, NASA SP-413, NASA, 1977. On-chip Annealing of Defects Generated by Hot-carrier Injection, Bias Temperature Instability, and Ionizing Radiation”. IEEE Electron Device 92. M. Jones. “Practical Von Neumann Machines and the Fermi Paradox”. Letters, 37(12):1543–1547, 2016. In International Astronautical Congress 2017 – 46th IAA Symposium on the Search for Extraterrestrial Intelligence (SETI) – The Next Steps, 2017. 67. A. Hein. “Artificial Intelligence Probes for Interstellar Exploration and Colonization”. arXiv preprint, 1612.08733, 2016. 93. A. Juhasz and G. Peterson. Review of Advanced Radiator Technologies for Spacecraft Power Systems and Space Thermal Control. 1994. 68. A. Hein. “Heritage Technologies in Space Programs – Assessment Methodology and Statistical Analysis”. PhD thesis, Technical University 94. E. R. Kandel, J. H. Schwartz, T. M. Jessell, and S. A. Siegelbaum. of Munich, 2016. Principles of Neural Science. McGraw-Hill, 2000. 69. A. Hein, K. Long, G. Matloff, R. Swinney, R. Osborne, A. Mann, and M. 95. C. Kemp. “Back to the Wild”. New Scientist, 2015. Ciupa. “Project Dragonfly: Small, Sail-Based Spacecraft for Interstellar 96. G. M. Kennedy, L. Matrà, M. Marmier, J. S. Greaves, M. C. Wyatt, G. Missions” submitted to JBIS, 2016. Bryden and B. Sibthorpe. “Kuiper Belt Structure Around Nearby Super- 70. A. M. Hein. “The Greatest Challenge: Manned Interstellar Travel”. In Earth Host Stars”. Monthly Notices of the Royal Astronomical Society, Beyond the Boundary: Exploring the Science and Culture of Interstellar 449(3):3121–3136, 2015. Spaceflight, pages 349–376. Lulu, 2014. 97. E. Kirchner. “Embedded Brain Reading”. Phd thesis, Universität 71. A. M. Hein. Transcendence Going Interstellar: How the Singularity Might Bremen, 2014. Revolutionize Interstellar Travel, 2014. 98. E. A. Kirchner, J. de Gea Fernandez, P. Kampmann, M. Schröer, J. H. 72. A. M. Hein and H. Condat. “Can Machines Design? An Artificial Metzen and F. Kirchner. “Intuitive Interaction with Robots – Technical General Intelligence Approach”. In B. Iklé, M. Franz, A. Rzepka & Approaches and Challenges”. In Formal Modeling and Verification R.Goertzel, editor, Artificial General Intelligence: 11th International of Cyber-Physical Systems, pages 224–248. Springer Fachmedien Conference, AGI 2018, pages 87–99. Springer, Prague, Czech Republic, Wiesbaden, Wiesbaden, 2015. 2018. 99. O. Kroemer and G. Sukhatme. “Learning Relevant Features for 73. A. M. Hein, K. F. Long, D. Fries, N. Perakis, A. Genovese, S. Zeidler, Manipulation Skills using Meta-level Priors”. arXiv preprint, M. Langer, R. Os- borne, R. Swinney, J. Davies, B. Cress, M. Casson, A. 1605.04439, 2016. Mann, and R. Armstrong. “The Andromeda Study: A Femto-Spacecraft 100. O. Kroemer and G. Sukhatme. “Meta-level Priors for Learning Mission to Alpha Centauri”. arXiv preprint, 1708.03556, 2017. Manipulation Skills with Sparse Features”. In Springer, pages 211–222, 74. A. M. Hein and R. Matheson. “A Techno-Economic Analysis of Asteroid 2016. Mining”. In 69th International Astronautical Congress (IAC), Bremen, 101. O. Kroemer and G. Sukhatme. “Feature Selection for Learning Versatile Germany, 2018. Manipulation Skills Based on Observed and Desired Trajectories”. 75. A. M. Hein, M. Pak, D. Pütz, C. Bühler, and P. Reiss. “World In 2017 IEEE International Conference on Robotics and Automation Ships –Architectures & Feasibility Revisited”. Journal of the British (ICRA), pages 4713–4720, 2017. Interplanetary Society, 65(4):119–133, 2012. 102. T. Kuiper and M. Morris. “Searching for Extraterrestrial Civilisations”. 76. A. M. Hein, N. Perakis, K. F. Long, and A. Crowl. “Project Lyra: Sending Science, 196:616–621, 1977. a Spacecraft to 1I/’Oumuamua (former A/2017 U1), the Interstellar 103. R. Kurzweil. The Singularity is Near: When Humans Transcend Biology. Asteroid”. arxiv.org, 2017. Penguin Books, 2005. 77. A. M. Hein, A. C. Tziolas, and R. Osborne. “Project Icarus: Stakeholder 104. P. LaVictoire. “An Introduction to Löbs Theorem in MIRI Research”. Scenarios for an Interstellar Exploration Program”. Journal of the British Technical report, Machine Intelligence Research Institute, 2015. Interplanetary Society, 64(6/7):224–233, 2011. 105. Y. LeCun, Y. Bengio, and G. Hinton. “Deep learning”. Nature, 521(7553), 78. J. Hernández-Orallo. Evaluation in Artificial Intelligence: from Task-

JBIS Vol 72 No.4 April 2019 141 ANDREAS HEIN & STEPHEN BAXTER

436-444, 521(7553):436–444, 2015. 132. A. Radford, L. Metz, and S. Chintala. “Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks”. 106. S. Legg and M. Hutter. “Universal Intelligence: a Definition of Machine arXiv preprint, 1511.06434, 2015. Intelligence”. In Minds and Machines. 2007. 133. S. Saeedi, M. Trentini, M. Seto and H. Li. “Multiple-Robot Simultaneous 107. H. Lipson and E. Malone. “Autonomous Self-Extending Machines for Localization and Mapping: A Review”. Journal of Field Robotics, Accelerating Space Exploration”. Technical report, NASA NIAC Phase 1 33(1):3–46, jan 2016. study, 2002. 134. C. Scharf. The Copernicus Complex. Farrar, Strauss and Giroux, 2014. 108. M. L. Littman. “Markov Games as a Framework for Multi-agent Reinforcement Learning”. In Machine Learning Proceedings 1994, pages 135. J. Schmidhuber. “Ultimate Cognition à la Gödel”. Cognitive 157–163, 1994. Computation, 1(2):177–193, jun 2009. 109. K. F. Long, M. J. Fogg, R. Obousy, A. Tzioloas, A. Mann, R. Osborne, 136. J. Schmidhuber. “Deep Learning in Neural Networks: an Overview”. and A. Presby. “Project Icarus: Son of Daedalus – Flying Closer to Neural networks, 61:85–117, 2015. Another Star”. Journal of the British Interplanetary Society, 62:403–414, 137. SFE. Pantropy, 2015. 2009. 138. D. Silver, I. Schrittwieser, J., Simonyan, K., Antonoglou, A. Huang, A. 110. P. Lubin. “A Roadmap to Interstellar Flight”. Journal of the British Guez, T. Hubert, L. Baker, M. Lai, A. Bolton and Y. Chen. “Mastering Interplanetary Society, 69(2-3), 2016. the Game of Go Without Human Knowledge”. Nature, 550(7676):354, 111. R. Lutz. “Software engineering for space exploration”. Computer, 2017. 44(10):41–46, 2011. 139. R. L. Simpson Jr and C. R. Twardy. “Refining the Cognitive Decathlon”. 112. J. Mankins. SPS-ALPHA: “The First Practical Solar Power Satellite via In Proceedings of the 8th Workshop on Performance Metrics for Intelligent Arbitrarily Large Phased Array” (a 2011-2012 NASA NIAC phase 1 Systems, pages 124–131. ACM, 2008. project). Artemis Innovation Management Solutions LLC, 2012. 140. M. Sipper. “Fifty Years of Research on Self-replication: an Overview”. 113. I. Markov. “Limits on Fundamental Limits to Computation”. Nature, Artificial Life, 4(3):237–257, 1998. 512(7513):147–154, 2014. 141. R. Skomorohov, A. Hein and C. Welch. “In-orbit Spacecraft 114. R. G. Martin and M. Livio. “On the Formation and Evolution of Manufacturing: Near-term Business Cases”. In International Asteroid Belts and their Potential Significance for Life”. Monthly Notices Astronautical Congress 2016, IAC-2016, Guadalajara, Mexico, 2016. of the Royal Astronomical Society: Letters, 428(1):L11–L15, 2012. 142. C. M. Smith. “Estimation of a Genetically Viable Population for 115. G. L. Matloff. Deep space probes: To the outer solar system and beyond. Multigenerational Interstellar Voyaging: Review and Data for Project Springer Science & Business Media, 2006. ”. Acta Astronautica, 97:16– 29, 2014. 116. P. T. Metzger. “Space Development and Space Science Together, an 143. M. Sonter. “The Technical and Economic Feasibility of Mining the Historic Opportunity”. Space Policy, 37(2):77–91, 2016. Near-Earth Asteroids”. Acta Astronautica, 1997. 117. P. T. Metzger, A. Muscatello, R. P. Mueller and J. Mantovani. “Affordable, 144. O. Stapledon. Last and First Men. Methuen, page number edition, 1930. Rapid Bootstrapping of the Space Industry and Solar System 145. B. Steunebrink, J Schmidhuber. “Towards an Actual Gödel Machine Civilization”. Journal of Aerospace Engineering, 26(1):18–29, 2012. Implementation: a Lesson in Self-reflective Systems”. In Theoretical 118. D. Moon, J. Park, J. Han, G. Jeon, J. Kim, J. Moon, M. Seol, C. Kim, Foundations of Artificial General Intelligence, pages 173–195, 2012. H. Lee, M. Meyyappan and Y. Choi. “Sustainable Electronics for 146. B. Steunebrink and J. Schmidhuber. “A Family of Gödel Machine Nano-spacecraft in Deep Space Missions”. In 2016 IEEE International Implementations”. In Artificial General Intelligence, pages 275–280, 2011. InElectron Devices Meeting (IEDM), pages 31–8, 2016. 147. M. Tan. “Multi-agent Reinforcement Learning: Independent vs. 119. S. T. Mueller. Is the Turing Test Still Relevant? a Plan for Developing the Cooperative Agents”. In Proceedings of the Tenth International Cognitive Decathlon to Test Intelligent Embodied Behavior”. In 19th Conference on Machine Learning, pages 330–337, 1993. Midwest artificial intelligence and cognitive science conference, MAICS, pages Vol. 1, p.3, 2008. 148. J. Tarter. “Alternative Models for Detecting Very Advanced Civilisations”. Journal of the British Interplanetary Society, 49:291–295, 1996. 120. S. T. Mueller, M. Jones, B. S. Minnery and J. M. Hiland. “The BICA Cognitive Decathlon: a Test Suite for Biologically-inspired Cognitive 149. D. R. Thompson, D. S. Wettergreen and F. J. C. Peralta. “Autonomous Agents”. In Proceedings of behavior representation in modeling and Science During Large-scale Robotic Survey”. Journal of Field Robotics, simulation conference, pages Vol. 1, p. 3, Norfolk, UK, 2007. 28(4), 542-564, 28(4):542–564, 2011. 121. S. V. Murphy and A. Atala. “3D Bioprinting of Tissues and Organs”. 150. F. Tipler and J. Barrow. The anthropic cosmological principle. Oxford Nature biotechnology, 32(8):773, 2014. University Press, 1986. 122. J. Myhill. “The Abstract Theory of Self-reproduction”. In Views on 151. F. J. Tipler. “Extraterrestrial Intelligent Beings Do Not Exist”. Quarterly general systems theory, pages 106–118. 1964. Journal of the Royal Astronomical Society, 21:267–281, 1980. 123. W. Nakayama. “Heat in Computers: Applied Heat Transfer in 152. F. J. Tipler. The Physics of Immortality: Modern cosmology, God, and the Information Technology”. Journal of Heat Transfer, 136(1):013001, 2014. Resurrection of the Dead. Doubleday, 1994. 124. G. K. O’Neill. “The Colonization of Space”. Physics Today, 27:32–40, 1974. 153. T. Toth-Fejel, R. Freitas and M. Moses. “Modeling Kinematic Cellular Automata Final Report”. Technical report, NASA NIAC Phase 1 report, 125. G. K. O’Neill. The High Frontier: Human Colonies in Space. William 2004. Morrow, New York, USA, 1977. 154. A. Tough. “Small Smart Interstellar Probes”. Journal of the British 126. G. K. O’Neill. 2081: A Hopeful View of the Human Future. Simon and Interplanetary Society, 51:167–174, 1998. Schuster, New York, 1981. 155. A. E. Trujillo, M. T. Moraguez, A. Owens, S. I. Wald and O. De 127. A. Owens and O. De Weck. “Systems Analysis of In-Space Weck. “Feasibility Analysis of Commercial In-Space Manufacturing Manufacturing Applications for the International Space Station and the Applications”. In AIAA SPACE and Astronautics Forum and Exposition, Evolvable Mars Campaign”. In AIAA SPACE 2016, Reston, Virginia, sep Reston, Virginia, sep 2017. American Institute of Aeronautics and 2016. American Institute of Aeronautics and Astronautics. Astronautics. 128. A. Owens, S. Do, A. Kurtz and O. Weck. “Benefits of Additive 156. A. Turchin. “The Global Catastrophic Risks Connected with Possibility Manufacturing for Human Exploration of Mars”. In 45th International of Finding Alien AI During SETI”. Journal of British Interpanetary Conference on Environmental Systems, 2015. Society, 71(2):71–79, 2018. 129. J. Pearson. Craig Venter’s ‘Digital-to-Biological Converter’ Is Real, 2017. 157. J. Veness, K. S. Ng, M. Hutter and D. Silver. “Reinforcement Learning 130. N. Perakis and A. M. Hein. “Combining Magnetic and Electric Sails for via AIXI Approximation”. In AAAI 2010, 2010. Interstellar Deceleration”. Acta Astronautica, 128:13–20, 2016. 158. J. Veness, K. S. Ng, M. Hutter, W. Uther, and D. Silver. “A Monte- 131. A. Potapov and S. Rodionov. “Universal Empathy and Ethical Bias for Carlo Aixi Approximation. Journal of Artificial Intelligence Research, Artificial General Intelligence”. Journal of Experimental & Theoretical 40(1):95–142, 2011. Artificial Intelligence, 26(3):405– 416, 2014. 159. S. A. Vere. “A Cognitive Process Shell”. Behavioral and Brain Sciences,

142 Vol 72 No.4 April 2019 JBIS ARTIFICIAL INTELLIGENCE for interstellar Travel

15(3):460–461, 1992. 166. O. N. Yalcin and S. DiPaola. A computational model of empathy for interactive agents. Biologically Inspired Cognitive Architectures, in print, 160. V. Vinge. Long Shot. Analog Science Fiction and Fact, 1972. 2018. 161. N. Vlassis. “A Concise Introduction to Multiagent Systems and 167. M. Yim, W. Shen, B. Salemi, D. Rus, M. Moll, H. Lipson, E. Klavins and Distributed Artificial Intelligence”. Lectures on Artificial Intelligence and G. Chirikjian. “Modular Self-reconfigurable Robot Systems [Grand Machine Learning, 1(1):1–71, 2007. Challenges of Robotics]. IEEE Robotics & Automation Magazine, 162. J. Von Neumann and A. W. Burks. Theory of Self-reproducing Automata. 14(1):43–52, 2007. University of Illinois Press, 1966. 168. E. Yudkowsky and M. Herreshoff. “Tiling Agents for Self-modifying 163. L. Weinstein and J.A. Adam. “Guesstimation: Solving the World's AI and the Löbian Obstacle”. Technical report, Machine Intelligence Problems on the Back of a Cocktail Napkin. Princeton University Press Research Institute, Berkeley, California, USA, 2013. 2019 169. R. Zubrin. The case for Mars. Simon and Schuster, 2012. 164. M. Wirkus. “Towards Robot-independent Manipulation Behavior 170. V. Zykov, E. Mytilinaios, B. Adams and H. Lipson. “Robotics: Self- Description”. arxiv preprint, 1412.3247, dec 2014. Reproducing Machines”. Nature, 435(7039):163, 2005. 165. M. Woods, A. Shaw, D. Barnes, D. Price, D. Long and D. Pullan. 171. V. Zykov, E. Mytilinaios, M. Desnoyer and H. Lipson. “Evolved and “Autonomous Science for an ExoMars Rover-like Mission”. Journal of Designed Self-reproducing Modular Robotics”. IEEE Transactions on Field Robotics, 26(4):358–390, apr 2009. robotics, 23(2):308–319, 2007.

Received 20 November 2018 Approved 5 March 2019

JBIS Vol 72 No.4 April 2019 143 Foundations of Interstellar Studies II A Workshop on Interstellar Flight

27–30 June 2019, Gloucestershire, United Kingdom

At the start of the new millennium we face one of the greatest challenges of our age: how can we cross the vast distances of space to visit worlds around other stars? It is a challenge that presents many difficult technical problems. Who better to address them than the global scientific community? www.fisw.space The first Foundations of Interstellar Studies workshop took place in New York City in June 2017; the second will take place in the county of Gloucestershire, United Kingdom, at the HQ of the Initiative for Interstellar Studies (i4is), who are the host organisation in partnership with the British Interplanetary Society. The workshop will consist of three days of scientific discussions on interstellar flight, focused on the themes below. Submissions are invited for presentations. For further details, visit www.fisw.space.

LIVING IN DEEP SPACE This theme includes space habitats on or planets. It The event is organised by the Co- Chairmen for the meeting: may also include existing on small exploration vessels, living Kelvin F Long, Initiative for Interstellar within medium slow boats or large world ships that travel over Studies, Stellar Engines, UK interstellar distances over many lifetimes. Rob Swinney, Initiative for Interstellar Studies, UK ADVANCED PROPULSION TECHNOLOGY & MISSIONS Harold ‘Sonny’ White, NASA Lyndon B This includes technologies that will take our probes outside Johnson Space Center, USA of the Solar System to interstellar or intergalactic distances. Propulsion concepts which include an application of known An invitation will be made to submit physics will be considered. Not considered here would be papers from selected authors post- concepts which only enable orbital or inner planetary travel. conference. For further information, contact the Workshop secretary: BUILDING ARCHITECTURAL MEGASTRUCTURES Samar AbdelFattah, University of This include constructions the size of moons or planets, Cairo, Aerospace Department, Egypt, such as planetary/stellar engineering initiatives like Dyson- Email: [email protected] Stapledon spheres, Stellar Engines, Matrioska brains, Ring Worlds and other innovative inventions. This may also include This event has been sponsored by the the possibility of constructing gravity-based engines from Interstellar Research Centre, Stellar space-time geometry such as worm holes. Engines Ltd CALL FOR PAPERS

Putting Astronauts in Impossible Locations A one day technical symposium

9:00 a.m - Wednesday 27 November 2019

BIS HQ, 27/29 South Lambeth Road

While the human exploration of the Moon and Mars has been extensively examined, serious technical consideration of the rest of the solar system has been largely ignored. This symposium is designed to explore the limits of where human exploration can go in the solar system and how to overcome the challenges involved. The symposium is open to papers on the transportation requirements, the practicalities of habitation in extreme environments and any other aspects of a solar system wide civilisation. Submissions should be on the basis that there will be a completed paper delivered before the symposium as well as giving a presentation on the day. All papers will be considered for publication in the Journal of the British Interplanetary Society.

Proposed papers should be described in an abstract of no more than 400 words, and submitted to the Society at [email protected]

Submission deadline 31 July 2019 Journal of the British Interplanetary Society

VOLUME 72 NO.4 APRIL 2019

www.bis-space.com

ISSN 0007-084X PUBLICATION DATE: 30 MAY 2019