DOCSLIB.ORG

A Review of Robot: Mere Machine to Transcendent Mind

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
A Review of Robot: Mere Machine to Transcendent Mind AI Magazine Volume 20 Number 3 (1999) (© AAAI) Book Review in this graph is just a foretaste of the Robot: Mere Machine degree to which Moravec will go in predicting the future. This book traces robotic history, projects the field’s to Transcendent Mind immediate future, and travels well beyond. Without giving away the A Review ending, suffice it to say that in this book, the reader will be treated to such ideas as, “What is reality any- way?” (p. 195). Another idea is that Bonnie Holte Bennett perhaps the most unsettling implication of this train of thought is that anything can be interpreted as possessing any ans Moravec’s new book, lar escalation of the estimates of the abstract property, including con- Robot: Mere Machine to Tran- complexity of human intelligence. sciousness and intelligence. Giv- Hscendent Mind, will delight With regard to the first caveat, en the right playbook, the ther- readers familiar with this work. Moravec’s estimates of animal equiva- mal jostling of the atoms in a Moravec’s strengths—his insightful lence are based solely on hardware rock can be seen as the operation data analysis, extrapolation of tech- complexity. It is often the case that of a complete, self–aware mind nology to extreme conclusions, and hardware alone cannot deliver perfor- (after Evert) (Everett, H., Many– provocative predictions—are all here mance, but it also requires software Worlds of Interpretation/ Quantum and will probably gain him some new sufficient to the task. Thus, projecting Mechanics, Princeton University fans. “human equivalence” in computer Press, 1973, p. 199). His insightful data analysis, for hardware does not mean that “human Yes, it is all here—quantum physics, example, includes plots of computing equivalent software” will be ready to chaos, cosmology, and more. trends of the last century, comparing run on the hardware. With regard to Moravec’s panoramic grasp of science mechanical, electromechanical, and the second caveat, neuroscientists reg- and technology is not unlike a good electronic computing capabilities in ularly adjust upward the estimate of carnival ride, exhilarating and some- terms of computational power per human brain complexity. This upward times exhausting. Parts of the book unit cost (MIPS per $1000 in 1998 U.S. estimation has the effect of extending need to be savored and digested slow- dollars) across time. This summarizes the estimated time at which comput- ly, if only because of their density and the evolution of computer power per magnitude. As for provocative predic- unit of cost. Readers of his previous tions, this book will not disappoint. book, Mind Children (Harvard Univer- Topics include virtual immortality, sity Press, 1988) will recognize the ultimate transcendence, and time current graph as a refinement of one travel, to name just a few. Hans Moravec, Robot: Mere that appeared in the earlier book. The chapters run approximately Machine to Transcendent Mind. Three things become clear from this chronologically. Chapter 1, “Escape Oxford.: Oxford University refined analysis: (1) computing Velocity,” introduces some of the Press, 1999. 227 pp. $25.00. progress has been tightly coupled themes of the book. It also includes ISBN 0–19–511630–5. around a trend line; (2) this makes some interesting graphs showing a predictions of future technologies rea- correlation between U.S. air pollution sonable by extrapolation; and (3) and wealth that suggests that we’ve there’s been an interesting accelera- survived the dirty industrial revolu- tion over the last 30 years, even on tion and are living richer and cleaner this logarithmic scale. The last point is ers will reach human equivalence. lives, having mended our evil ways. a new addition to this. These data Indeed, Moravec’s estimates have We’re using our wealth to live more allow predictions about when com- increased too. In 1988, he projected pristinely, and the future looks bright. puters will cross the boundaries of computers will reach human equiva- Chapter 2, “Caution Robot Vehicle,” complexity equivalent to a lizard lence in a supercomputer in the years traces the last 30 years of robotics (approximately the years 2002–2008), 2000 to 2005 and in a personal com- research, with particular detail on the a mouse (approximately the years puter by 2028 to 2030+. The 1999 Stanford University and Carnegie Mel- 2008–2017), a monkey (approximate- estimate is 2020 to 2030+. However, lon University labs with which ly the years 2013–2028), and a human these details are minor. This graph is Moravec has been affiliated. It is an (approximately the years 2020– fascinating in its scope and breadth excellent history and summary of the 2030+). There are, however a few and particularly notable in its predic- progress made. Chapter 3, “Power and caveats to consider: (1) these are hard- tive capabilities. It is vintage Moravec. Presence,” sets up the paradigms used ware estimates and (2) there is a regu- The insightful extrapolation of data throughout the rest of the book. These 92 AI MAGAZINE Copyright © 1999, American Association for Artificial Intelligence. All rights reserved. 0738-4602-1999 / $2.00 Book Review Call for Papers The Fifth International Conference on Artificial Intelligence Planning and Scheduling Systems (AIPS 2000) April 14-17, 2000, Breckenridge, Colorado, USA Co-loca ted with KR 2000, April 12-15, 2000 The International Conference on AI Planning and Scheduling Systems is being held in cooper ation with AAAI and wil l bring together researchers working on all aspects of planning, scheduling, planning and learning, and plan execution. The format of the conference will include paper presentations, invited speakers, panel discussions, workshops, and competitions among implemented planning, scheduling, and execution systems. Topics of interest include: - Planning algorithms - Planning and reasoning about action and change - Planning and learning - Planning and temporal reasoning - Planning under uncertainty - Decision-theoretic planning - Scheduling al gorithms - Co nstraint management in p lanning and scheduling - Resource management, Mixed-initiative planning - Multi -agent planning - Plan execution, monitoring and replanning - Agent architectures for planning and control - Active perception and sensor-based planning - Reactive systems - Applications and domain-specific techniques - Plan and schedule visualization - Issues in fielding planning and scheduling systems - Connections between planning and CSP/OR The paper Submissio n Deadl ine is November 1, 1999. The Workshop Submission Deadline is December 15, 1999. We plan to provide special recognition to papers of exceptional quality. For more information on workshops, the main conference, and the competitions see http://www.isi.edu/aips. paradigms include the basis for com- vides a cheap, clean access to space, cussion of a personally uninteresting putational complexity; storage and and time travel. discipline so that the real reward is processing speeds that are delineated This summary of the chapters leads never reached. here; and the graph predicting com- to a caution I would make in recom- Moravec’s work has epic propor- puter growth, along with others docu- mending this book to colleagues. tions. He engenders provocative dis- menting chess successes and computa- Some people, for example, my cussion and provides daring insight. tional power. Chapter 4, “Universal philosopher friends, might be less His new book is destined to be the Robots,” delineates four generations of interested in the technical detail. To classic his previous book has become. universal robots that will appear over them I would say, “Skim chapters 2 the next half–century. The final gener- and 5; don’t let them bog you down. ation includes superrational robots You can skip them for now and come with emotional displays. Chapter 5, back to them if you want.” For analyt- Bonnie Holte Bennett is director of the Artificial Intelligence Laboratory and an “The Age of Robots,” chapter 6, “The ic readers, I might encourage them to associate professor in the Graduate Pro- Age of Mind,” and chapter 7, “Mind start with chapter 3 as a “hook.” The grams in Software at the University of St. Fire,” look at the future, beyond the point is there is something very Thomas. Her e-mail address is point where robots surpass human rewarding in this book for many peo- [email protected]. capabilities. In these chapters are con- ple, but the vastness of some sections cepts as intriguing as the “syn- might overwhelm them, and it would chronous orbital bridge,” which pro- be a shame to be lost in the rich dis- FALL 1999 93 New Proceedings from AAAI Press Proceedings of the Proceedings of the Seventh International Conference Twelfth International Florida on Intelligent Systems for Artificial Intelligence Research Molecular Biology Society Conference Edited by Thomas Lengauer, Reinhard Schneider, Edited by Amruth Kumar and Ingrid Russell Peer Bork, Douglas Brutlag, Janice Glasgow, he Florida AI Research Society Conference was founded in Hans-Werner Mewes, and Ralf Zimmer T1987 to promote and advance AI research within the state he ISMB conference series provides a general forum for of Florida, fostering interaction between researchers at col- Tdisseminating the latest developments in bioinformatics. leges, universities, and industry. Since 1990, Florida AI ISMB is a multidisciplinary conference that brings together Research Society conferences have been broadened to include scientists from computer science, mathematics, molecular
Load more

Recommended publications
  • Technological Singularity
    TECHNOLOGICAL SINGULARITY BY VERNOR VINGE The original version of this article was presented at the VISION-21 Symposium sponsored by NASA Lewis Research Center and the Ohio Aerospace Institute, March 30-31, 1993. It appeared in the winter 1993 Whole Earth Review --Vernor Vinge Except for these annotations (and the correction of a few typographical errors), I have tried not to make any changes in this reprinting of the 1993 WER essay. --Vernor Vinge, January 2003 1. What Is The Singularity? The acceleration of technological progress has been the central feature of this century. We are on the edge of change comparable to the rise of human life on Earth. The precise cause of this change is the imminent creation by technology of entities with greater-than-human intelligence. Science may achieve this breakthrough by several means (and this is another reason for having confidence that the event will occur): * Computers that are awake and superhumanly intelligent may be developed. (To date, there has been much controversy as to whether we can create human equivalence in a machine. But if the answer is yes, then there is little doubt that more intelligent beings can be constructed shortly thereafter.) * Large computer networks (and their associated users) may wake up as superhumanly intelligent entities. * Computer/human interfaces may become so intimate that users may reasonably be considered superhumanly intelligent. * Biological science may provide means to improve natural human intellect. The first three possibilities depend on improvements in computer hardware. Actually, the fourth possibility also depends on improvements in computer hardware, although in an indirect way.
    [Show full text]
  • Prepare for the Singularity
    Prepare for the Singularity The definition of man's uniqueness has always formed the kernel of his cosmological and ethical systems. With Copernicus and Galileo, he ceased to be the species located at the center of the universe, attended by sun and stars. With Darwin, he ceased to be the species created and specially endowed by God with soul and reason. With Freud, he ceased to be the species whose behavior was—potentially—governable by rational mind. As we begin to produce mechanisms that think and learn, he has ceased to be the species uniquely capable of complex, intelligent manipulation of his environment. I am confident that man will, as he has in the past, find a new way of describing his place in the universe—a way that will satisfy his needs for dignity and for purpose. But it will be a way as different from the present one as was the Copernican from the Ptolemaic. —Herbert Simon Whether you think Herb is the successor to Copernicus, Galileo, Darwin, and Freud, and whatever you think about the Singularity, you must admit that technology is moving at a fast, accelerating pace, perhaps challenging humans’ pre-eminence as the vessels of intelligence. As Computer Science drives artificial intelligence forward, we should understand its impact. Figure 1.Kurzweil’s Projection Hans Moravec first suggested that intelligent robots would inherit humanity’s intelligence and carry it forward. Many people now talk about it seriously, e.g. the Singularity Summits. Academics like theoretical physicist David Deutsch and philosopher David Chalmers, have written about it.
    [Show full text]
  • Transhumanism Between Human Enhancement and Technological Innovation*
    Transhumanism Between Human Enhancement and Technological Innovation* Ion Iuga Abstract: Transhumanism introduces from its very beginning a paradigm shift about concepts like human nature, progress and human future. An overview of its ideology reveals a strong belief in the idea of human enhancement through technologically means. The theory of technological singularity, which is more or less a radicalisation of the transhumanist discourse, foresees a radical evolutionary change through artificial intelligence. The boundaries between intelligent machines and human beings will be blurred. The consequence is the upcoming of a post-biological and posthuman future when intelligent technology becomes autonomous and constantly self-improving. Considering these predictions, I will investigate here the way in which the idea of human enhancement modifies our understanding of technological innovation. I will argue that such change goes in at least two directions. On the one hand, innovation is seen as something that will inevitably lead towards intelligent machines and human enhancement. On the other hand, there is a direction such as “Singularity University,” where innovation is called to pragmatically solving human challenges. Yet there is a unifying spirit which holds together the two directions and I think it is the same transhumanist idea. Keywords: transhumanism, technological innovation, human enhancement, singularity Each of your smartphones is more powerful than the fastest supercomputer in the world of 20 years ago. (Kathryn Myronuk) If you understand the potential of these exponential technologies to transform everything from energy to education, you have different perspective on how we can solve the grand challenges of humanity. (Ray Kurzweil) We seek to connect a humanitarian community of forward-thinking people in a global movement toward an abundant future (Singularity University, Impact report 2014).
    [Show full text]
  • Roboethics: Fundamental Concepts and Future Prospects
    information Review Roboethics: Fundamental Concepts and Future Prospects Spyros G. Tzafestas ID School of Electrical and Computer Engineering, National Technical University of Athens, Zographou, GR 15773 Athens, Greece; [email protected] Received: 31 May 2018; Accepted: 13 June 2018; Published: 20 June 2018 Abstract: Many recent studies (e.g., IFR: International Federation of Robotics, 2016) predict that the number of robots (industrial, service/social, intelligent/autonomous) will increase enormously in the future. Robots are directly involved in human life. Industrial robots, household robots, medical robots, assistive robots, sociable/entertainment robots, and war robots all play important roles in human life and raise crucial ethical problems for our society. The purpose of this paper is to provide an overview of the fundamental concepts of robot ethics (roboethics) and some future prospects of robots and roboethics, as an introduction to the present Special Issue of the journal Information on “Roboethics”. We start with the question of what roboethics is, as well as a discussion of the methodologies of roboethics, including a brief look at the branches and theories of ethics in general. Then, we outline the major branches of roboethics, namely: medical roboethics, assistive roboethics, sociorobot ethics, war roboethics, autonomous car ethics, and cyborg ethics. Finally, we present the prospects for the future of robotics and roboethics. Keywords: ethics; roboethics; technoethics; robot morality; sociotechnical system; ethical liability; assistive roboethics; medical roboethics; sociorobot ethics; war roboethics; cyborg ethics 1. Introduction All of us should think about the ethics of the work/actions we select to do or the work/actions we choose not to do.
    [Show full text]
  • Ray Kurzweil Reader Pdf 6-20-03
    Acknowledgements The essays in this collection were published on KurzweilAI.net during 2001-2003, and have benefited from the devoted efforts of the KurzweilAI.net editorial team. Our team includes Amara D. Angelica, editor; Nanda Barker-Hook, editorial projects manager; Sarah Black, associate editor; Emily Brown, editorial assistant; and Celia Black-Brooks, graphics design manager and vice president of business development. Also providing technical and administrative support to KurzweilAI.net are Ken Linde, systems manager; Matt Bridges, lead software developer; Aaron Kleiner, chief operating and financial officer; Zoux, sound engineer and music consultant; Toshi Hoo, video engineering and videography consultant; Denise Scutellaro, accounting manager; Joan Walsh, accounting supervisor; Maria Ellis, accounting assistant; and Don Gonson, strategic advisor. —Ray Kurzweil, Editor-in-Chief TABLE OF CONTENTS LIVING FOREVER 1 Is immortality coming in your lifetime? Medical Advances, genetic engineering, cell and tissue engineering, rational drug design and other advances offer tantalizing promises. This section will look at the possibilities. Human Body Version 2.0 3 In the coming decades, a radical upgrading of our body's physical and mental systems, already underway, will use nanobots to augment and ultimately replace our organs. We already know how to prevent most degenerative disease through nutrition and supplementation; this will be a bridge to the emerging biotechnology revolution, which in turn will be a bridge to the nanotechnology revolution. By 2030, reverse-engineering of the human brain will have been completed and nonbiological intelligence will merge with our biological brains. Human Cloning is the Least Interesting Application of Cloning Technology 14 Cloning is an extremely important technology—not for cloning humans but for life extension: therapeutic cloning of one's own organs, creating new tissues to replace defective tissues or organs, or replacing one's organs and tissues with their "young" telomere-extended replacements without surgery.
    [Show full text]
  • Transhumanism
    T ranshumanism - Wikipedia, the free encyclopedia http://en.wikipedia.org/w/index.php?title=T ranshum... Transhumanism From Wikipedia, the free encyclopedia See also: Outline of transhumanism Transhumanism is an international Part of Ideology series on intellectual and cultural movement supporting Transhumanism the use of science and technology to improve human mental and physical characteristics Ideologies and capacities. The movement regards aspects Abolitionism of the human condition, such as disability, Democratic transhumanism suffering, disease, aging, and involuntary Extropianism death as unnecessary and undesirable. Immortalism Transhumanists look to biotechnologies and Libertarian transhumanism other emerging technologies for these Postgenderism purposes. Dangers, as well as benefits, are Singularitarianism also of concern to the transhumanist Technogaianism [1] movement. Related articles The term "transhumanism" is symbolized by Transhumanism in fiction H+ or h+ and is often used as a synonym for Transhumanist art "human enhancement".[2] Although the first known use of the term dates from 1957, the Organizations contemporary meaning is a product of the 1980s when futurists in the United States Applied Foresight Network Alcor Life Extension Foundation began to organize what has since grown into American Cryonics Society the transhumanist movement. Transhumanist Cryonics Institute thinkers predict that human beings may Foresight Institute eventually be able to transform themselves Humanity+ into beings with such greatly expanded Immortality Institute abilities as to merit the label "posthuman".[1] Singularity Institute for Artificial Intelligence Transhumanism is therefore sometimes Transhumanism Portal · referred to as "posthumanism" or a form of transformational activism influenced by posthumanist ideals.[3] The transhumanist vision of a transformed future humanity has attracted many supporters and detractors from a wide range of perspectives.
    [Show full text]
  • Necropolitics and the Dark Comedy of the Posthuman
    North American Bison 1 Necropolitics and the Dark Comedy of the Posthuman Twenty years ago, CAE wrote its first essay on the posthuman (“Posthuman Development in the Age of Pancapitalism”), and now we are back grap- pling with this topic once again. The upside is that very little has changed over the past two decades. The posthuman fantasy is still just that, but the fact that this fantasy remains so entrenched in the collective imagination of technocrats, engineers, solutionists, and digital enthusiasts speaks to the power of nihilistic desire in humans—a desire for the end of humanity. This desire is what truly distinguishes those who champion the posthuman. The posthumanists are not pretenders or reformers. They have no interest in panhumanism like the so-called postmodern antihumanists (a very poor choice of words), nor are they promoting some newly revised philosophy of humanism. They are revolutionaries dedicated to creating an explosive movement in evolution in which a new creature specializing in intelligence is formed, and/or to eliminating humans from much or all of the earth. Humans Unfortunately, for the purposes of this essay, a model of what a human is is necessary, so that it might function as a point of contrast to the posthuman. 22 Aesthetics, Necropolitics, and Environmental Struggle What CAE is about to offer is meant neither as a universal nor as a phil- osophically robust definition. We are only offering an imperfect, opera- tional model that consists of eight points of tremendous elasticity. We are speaking of tendencies that, when bundled in various configurations, could represent many human variations.
    [Show full text]
  • A Humanless World: the Progressive Artificial Intelligence and Its Impact on Intellectual Property
    A Humanless World: The Progressive Artificial Intelligence and its Impact on Intellectual Property Kaloyan Dinkov Dinev Abstract With recent leaps in the field of robotics and artificial intelligence, the boundary between science fiction and reality continues to blur, especially as computers performing actions previously thought to be unique only to man (often times, more efficiently) becomes more common. Though challenged by the traversal of the physical realm, their capabilities of mimicking human problem solving and creativity is growing exponentially. In discovering this potential lies the possibility of a cultural and scientific enrichment, the likes of which will dwarf both the Renaissance and the Industrial Revolution, respectively. Naturally this is accompanied by countless inventions and artistic expressions from which society will inevitably benefit. However, due to their originators likely being wholly, or at least in part, computers, the question will arise as to who their true owners will be from a legal standpoint. This article will examine the extent to which human and machine intelligence can be compared and will attempt to ascertain the ways in which the latter could pose a threat to the system of Intellectual Property laws in place to protect the rights of creators. Where possible, the piece will also attempt to propose potential remedies and pre-emptive actions to alleviate the dangers such a synthetic revolution could cause to IP’s legal framework. 1 KENT STUDENT LAW REVIEW Volume 4 2018 Introduction Throughout the ages, science fiction writers and futurists alike have dreamt of a world in which technological advancements have created machines capable of mimicking and reproducing intelligence indistinguishable from that of unaltered humans.
    [Show full text]
  • Emergence of the Posthuman Subject a Two-Day International Conference Hosted by the Department of English, University of Surrey
    Emergence of the Posthuman Subject A two-day international conference hosted by the Department of English, University of Surrey Sponsored by the Institute for Advanced Studies Welcome Programme. Speakers. Abstracts. Over the past two decades theories and critical practices associated with posthumanism have become an increasingly significant presence in the Arts and Sciences. Inspired by the radical innovation that period has seen in information and communication technology, philosophers and writers have hailed what amounts to a break with the humanist tradition that has underpinned western civilisation for over five-hundred years. This conference sets out to trace the Emergence of the Posthuman Subject, in fields as diverse as literature, psychology, philosophy, anthropology, film, palaeontology, zoology, theatre and physics. The event takes place at the University of Surrey, the institution at the forefront of space exploration technology in the UK – situated in that territory west of London, criss-crossed by flight-paths and motorways, celebrated in key novels by JG Ballard. 2 www.surrey.ac.uk www.surrey.ac.uk 3 Programme Programme Friday 2 July 2010 Friday 2 July 2010 Programme Programme Friday 2 July 2010 Friday 2 July 2010 9.00-9.30 Registration and Coffee 13.15-14.45 Panel 3 – Cloning 9.30-10.00 Welcome Address Adrienne Banner (San Francisco State University) Isaac Asimov’s “Evidence” 9.45-10.45 Keynote 1 Josie Gill (University of Cambridge) Dr Andy Mousley, Senior Lecturer in English at the University of De Montfort, Posthumanism
    [Show full text]
  • Intelligence Explosion FAQ
    MIRI MACHINE INTELLIGENCE RESEARCH INSTITUTE Intelligence Explosion FAQ Luke Muehlhauser Machine Intelligence Research Institute Abstract The Machine Intelligence Research Institute is one of the leading research institutes on intelligence explosion. Below are short answers to common questions we receive. Muehlhauser, Luke. 2013. “Intelligence Explosion FAQ.” First published 2011 as “Singularity FAQ.” Machine Intelligence Research Institute, Berkeley, CA Contents 1 Basics 1 1.1 What is an intelligence explosion? . 1 2 How Likely Is an Intelligence Explosion? 2 2.1 How is “intelligence” defined? . 2 2.2 What is greater-than-human intelligence? . 2 2.3 What is whole-brain emulation? . 3 2.4 What is biological cognitive enhancement? . 3 2.5 What are brain-computer interfaces? . 4 2.6 How could general intelligence be programmed into a machine? . 4 2.7 What is superintelligence? . 4 2.8 When will the intelligence explosion happen? . 5 2.9 Might an intelligence explosion never occur? . 6 3 Consequences of an Intelligence Explosion 7 3.1 Why would great intelligence produce great power? . 7 3.2 How could an intelligence explosion be useful? . 7 3.3 How might an intelligence explosion be dangerous? . 8 4 Friendly AI 9 4.1 What is Friendly AI? . 9 4.2 What can we expect the motivations of a superintelligent machine to be? 10 4.3 Can’t we just keep the superintelligence in a box, with no access to the Internet? . 11 4.4 Can’t we just program the superintelligence not to harm us? . 11 4.5 Can we program the superintelligence to maximize human pleasure or desire satisfaction? .
    [Show full text]
  • Pigs in Cyberspace Hans Moravec Robotics Institute Carnegie Mellon University Pittsburgh, PA 15213
    N94- 27376 Pigs in Cyberspace Hans Moravec Robotics Institute Carnegie Mellon University Pittsburgh, PA 15213 Exploration and colonization of the universe awaits, but Earth-adapted biological humans are ill-equipped to respond to the challenge. Machines have gone farther and seen more, limited though they presently are by insect-like behavioral inflexibility. As they become smarter over the coming decades, space will be theirs. Organizations of robots of ever increasing intelligence and sensory and motor ability will expand and transform what they occupy, working with matter, space and time. As they grow, a smaller and smaller fraction of their territory will be undeveloped frontier. Competitive success will depend more and more on using already available matter and space in ever more refined and useful forms. The process, analogous to the miniaturization that makes today's computers a trillion times more powerful than the mechanical calculators of the past, will gradually transform all activity from grossly physical homesteading of raw nature, to minimum-energy quantum transactions of computation. The final frontier will be urbanized, ultimately into an arena where every bit of activity is a meaningful computation: the inhabited portion of the universe will transformed into a cyberspace. Because it will use resources more efficiently, a mature cyberspace of the distant future will be effectively much bigger than the present physical universe. While only an infinitesimal fraction of existing matter and space is doing interesting work, in a well developed cyberspace every bit will be part of a relevant computation or storing a useful datum. Over time, more compact and faster ways of using space and matter will be invented, and used to restructure the cyberspace, effectively increasing the amount of computational spacetime per unit of physical spacetime.
    [Show full text]
  • Occam's Razor Is Insufficient to Infer the Preferences of Irrational Agents
    Occam’s razor is insufficient to infer the preferences of irrational agents Stuart Armstrong *∗ † Soren¨ Mindermann* ‡ Future of Humanity Institute Vector Institute University of Oxford University of Toronto [email protected] [email protected] Abstract Inverse reinforcement learning (IRL) attempts to infer human rewards or pref- erences from observed behavior. Since human planning systematically deviates from rationality, several approaches have been tried to account for specific hu- man shortcomings. However, the general problem of inferring the reward func- tion of an agent of unknown rationality has received little attention. Unlike the well-known ambiguity problems in IRL, this one is practically relevant but cannot be resolved by observing the agent’s policy in enough environments. This paper shows (1) that a No Free Lunch result implies it is impossible to uniquely decom- pose a policy into a planning algorithm and reward function, and (2) that even with a reasonable simplicity prior/Occam’s razor on the set of decompositions, we cannot distinguish between the true decomposition and others that lead to high regret. To address this, we need simple ‘normative’ assumptions, which cannot be deduced exclusively from observations. 1 Introduction In today’s reinforcement learning systems, a simple reward function is often hand-crafted, and still sometimes leads to undesired behaviors on the part of RL agent, as the reward function is not well aligned with the operator’s true goals4. As AI systems become more powerful and autonomous, these failures will become more frequent and grave as RL agents exceed human performance, oper- ate at time-scales that forbid constant oversight, and are given increasingly complex tasks — from driving cars to planning cities to eventually evaluating policies or helping run companies.
    [Show full text]