Prediction is hard, especially if it is about the future Niels Bohr (1957), Nobel Prize winner Some remarkable bloops

● The Internet is like a Supernova and will completely collapse in 1996 Bob Metcalfe (1995), Ethernet LAN inventor ● 640 KB (RAM) should be sufficient for anyone "Urban Legend“ (1981), wrongly attributed to ● There is no reason why anyone would want a computer at home Ken Olson, DEC CEO (1977) ● Future computers could weigh less than 1.5 Ton Popular Mechanics (1949) ● There could be a global market for five computers T.J.Watson, IBM CEO (1943) ● Atomic energy is “moonshine” Lord Rutherford, Nobel Prize winner (11.09.1933, 24h before Leo Szilard published how to do it) ● Everything that could be invented, has been invented US Patent Office Director (1899) ● Heavier than air flying machines are impossible Lord Kelvin (1895)

● Predicting the future is like scratching yourself before it starts to itch Peter Sellers

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 0 / 50 The Limits of (Artificial) Intelligence AI – Salvation or damnation of mankind

Ph. Janson

21 July 2020

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 1 / 50 R. Kurzweil’s Law of Accelerating Returns captured by K. Kelly’s 12 technological trends (Kevin Kelly – The Inevitable (2016))

1. Becoming – constant change, development, beta-testing 2. Cognifying – AI everywhere 3. Flowing – information fluidity and streaming 4. Screening – pixels instead of paper 5. Accessing – instead of owning => customer lock-in! Kurzweil’s 6. Sharing – through “dot.communismus” cooperation accelerating 7. Filtering – selecting from massive info wealth returns 8. Remixing – “mashups” … [as am doing in this talk] … 9. Interacting – with “Internet-of-Things” devices 10. Tracking – through “Internet-of-Things” devices => privacy threat 11. Questioning – anybody can publish anything => democracy threat 12. Beginning – “innovation on steroïds”

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 2 / 50 Central question of this talk (R. Kurzweil, H. Moravec, E. Fredkin, M.Tegmark: Life 3.0 (2017), Y.N.Harari: Sapiens (2014), D.Christian: A Big History of Everything (2018)) Time

Life 1.0 Biology, genetics : Darwin’s evolution = natural selection

Life 2.0 Human Intelligence Sociology: cultural evolution = religions, politics

Life 3.0 Technology: AI takeover

Evolutionary Force: Intelligence?

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 3 / 50 C. Christensen’s Inventor’s Dilemna applied to human intelligence ?

Technology ages, is abandoned, replaced, and disappears Roy Amara: Technology is underestimated in the long term Does that hold for biological intelligence?

Roy Amara: Technology is overestimated In the short term

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 4 / 50 Agenda

●What is AI ?

●Computability limits => intelligence limits ?

●The Singularity – if AI > HI then AI >> HI ?

●AI Risks

●Metaphysical ruminations – the future of mankind and intelligence ?

●Outlook

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 5 / 50 The Universal Turing Machine

No computer (however fast) will ever be able to compute anything that Turing’s Universal Machine could not compute – albeit much slower • NB: RNA Polymerase has the exact same capabilities as a Turing Machine

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 6 / 50 Artificial Intelligence (John McCarthy (1955))

●L. Tessler’s Theorem: AI is «whatever computers still cannot compute»

●What should computer intelligence mean ?=> Turing Test

•●A computer counts as intelligent when it exhibits human cognitive abilities ●Perception, knowledge representation, argumentation, problem solving, learning, etc. => mimicking a human brain

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 7 / 50 Brain simulation research projects

●EU (EPFL, ETHZ, …): Human Brain Project ●DARPA (US): SyNAPSE Project ●IBM: SyNAPSE Chip: 4’000 processors, 63 mW, 256’000 synapses connecting 1 million neurons Target: SyNAPSE Computer 2 dm³ 4KW, 100T synapses connecting 10 billion neurons 3 15 ●NB: Human brain: 1.5 dm , 20W, 10 synapses connecting 100 billion neurons 1016 computations per second =>• Computers are are only a few orders of magnitude away from that ≈> 10-100 years ?

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 8 / 50 (ML): How do computers learn ? https://www.simplilearn.com/machine-learning-tutorial

1. Supervised Learning Learning by examples (as a child under guidance of parents / teachers) (Computers need million more examples than kids!) e.g. Deep Machine Learning = through deep neural networks Questions Learning AI Trained AI Answers

2. Reinforcement Learning Learning through positive feedback (as a child without guidance but with external feedback) Evolutionary Learning = through trial and error (Good for problems with clear and concise goals) Data (Autonomous) Trial Trained AI Goal Learning AI Feedback

3. Unsupervised Learning Learning through autonomous discovery (as a child exploring with neither guidance nor feedback) “Intelligent” learning = through observation and imitation Data Continuously Improving (Goal) Learning AI results

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 9 / 50 Supervised Learning: Deep Machine Learning (Geoff Hinton, Google & U. of Toronto) (Tensor) / Graphics Processing Unit (T/GPU) have the same capabilities as Universal Turing Machines a a a a Information: a b b a Every node x,y computes b b 푦=푧 ●Data wx,y = σ푦=푎 px,y wx−1,y b c c b ●Text c c ●Graphics c d d c Decision <= output node with highest vote ●Pictures d d d e e d ●Videos e e «Training» the many px,y parameters ●Audio e f f e is extremely intensive, expensive f f and training-data biased! g g Level: Inputs 1 2 3 4 Outputs Remarkable results in translation, medecine, etc.

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 10 / 50 Reinforcement learning: GANs (Generative Adversarial Networks)

●Two GPUs are learing from one another by trying to outsmart one another

a a a a a a a a a b b a a b b a b b b b b c c b b c c b c c c c c d d c c d d c d d d d d e e d d e e d e e e e e f f e e f f e f f f f g g g g

Both AIs are learning through a sort of Darwinian natural selection Learning does however stop in production, contrary to unsupervised – or human – learning The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 11 / 50 Reinforcement Learning example: walking

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 12 / 50 AI performance and potential AI already can & increasingly may solve problems that humans cannot solve

●Climate & environmental crises ●Geology, seismics, archeologie, weather ●Materials science ●Transport, logistics, autonomous vehicles … meat without animals & methane emissions ●Advertising, art, video games … construction without steel and concrete ● Language translation ●Energy production ●Finance, economy … planes without fossil fuel ●Health, drug development ●Resource recycling instead of waste disposal ●Administration, audit ●Economy without growth ● ●(Cyber)security, justice Fair financial world ●Medicine & education ●Politics & peace e.g. UN Policy Priority Inference (PPI) AI ●Space colonisation ●etc.

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 13 / 50 Agenda

●What is AI ?

●Computability limits => intelligence limits ?

●The Singularity – if AI > HI then AI >> HI ?

●AI Risks

●Metaphysical ruminations – the future of mankind and intelligence ?

●Outlook

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 14 / 50 Computation has limits: many things are not computable

●All possible programs are countable, just like integers

●All possible math functions are NOT countable, just like real numbers

=> Set of all functions >>> set of all programs

=> Only a small minority of all functions are programmable

=> An infinite majority of phenomena are not computable

=> Example: a true random number generator

•●Quantum physics uncertainty principle puts limits to measurability => computability

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 15 / 50 Intelligence has limits: Self-references make statements undecidable / meaningless

Examples: ●This paradoxical drawing by the Belgian cartoonist Ph. Geluck

It is forbidden to read this writing

●The paradoxical statement by the Creatan philosopher Epimenides «All Creatans are liars» (or «I am a liar»)

•●The paradox of Berry / Russel «N is the smallest number than cannot be defined in less than 20 words» (only 14 words!)

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 16 / 50 Intelligence has limits: Self-references make questions undecidable / meaningless Further examples:

●No program can ever decide, whether another program always terminates (or not)

●A formal system can never prove itself consistent & complete at the same time (e.g. mathematics, logic, any formal (e.g. programming) language, etc.) (K. Gödel’s Theorems, 1933)

●Reproduction is ‘self-referencing’ (D. Hofstadter’s strange loops, 1979) HI

Data in Program Data out Data in AI Data out => Resulting question: can AI, given Gödel’s theorems, be consistent and complete ?

•●Are genes meant to reproduce phenotypes OR are phenotypes meant to reproduce genes ?? (R. Dawkins, 1976)

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 17 / 50 Three worlds – Another strange loop ? (R. Penrose: Shadows of the Mind (1994)

Physical (real) world Is abstracted, Models modeled behavior

Platonic Mental (mathematical) (intellectual) world world

Discovers rules

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 18 / 50 What is in the end computable / decidable ? (Source: G. Brookshear) Problem complexity

Trivial Non-trivial problems problems

Computable problems Undecidable Polynomial Non-polynomial problems computable computable Simple tabulation P ? NP e.g. Theoretically table Non-deterministically computable of city computable but but practically time zones Polynomially provable inefficient

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 19 / 50 Example of an NP problem: the Traveling Salesman Problem (TSP)

●What is the shortest way thru all 26 canton capitals?

●Not efficiently computable

●But polynomially provable

●NB: an amoeba seems to be able to solve the problem in about linear time ! https://thenewstack.io/amoeba-based-computer-solves-traveling-salesman-puzzle/ The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 20 / 50 Example of an inefficiently computable problem

●A modern computer can execute 10 billion computations per second

●CERN generates about 1 PB (1’000’000’000’000’000!) of data per week

●A program that would compare ever data with every other data would need 1015 x 1015 = 1030 comparisons => 1030 comparisons/1010 computations/sec = 1020 seconds ≈ 300 million years!

15 ●Even a program that would read each data only once would need 10 operations => 1015 operations/1010 computations/sec = 105 seconds ≈ about 27 hours!

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 21 / 50 Agenda

●What is AI ?

●Computability limits => intelligence limits ?

●The Singularity – if AI > HI then AI >> HI ?

●AI Risks

●Metaphysical ruminations – the future of mankind and intelligence ?

●Outlook

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 22 / 50 The singularity: If? How? Why? When?

●Definition: AI breakout: AI learns, improves, reproduces faster than HI => KI >> MI => uncontrolable & irrevocable (self-preservation)?

●If, how? Is it possible at all?

●Why not? Evolution => particles => cells => organs => animals => humans Humans => particles => transistors => computers => AI => AI

●Why? Baldwin Effect: survival & reproduction benefit from learning & knowledge Sapiens eliminated Neanderthals – Colons decimated natives Experts survey: ●Intelligence as fundamental ingredient of the universe ? How long to the singularity? => AI as next step of Darwin’s evolution ?

•●When? Your opinion? => survey @ ageofai.org https://futureoflife.org/superintelligence-survey/

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 23 / 50 4 possible futures R. Penrose (1994)

AI capabilities Incl. conscience

A Full HI Yes

B Simulated HI No Full / simulated HI but not only with C additional scientific discoveries Depends (e.g. quantum physics effects ?) HI capabilities impossible D No (because of self-references ?)

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 24 / 50 Optimists – Skeptics – Pessimists: ½ full vs. ½ empty glass?

╋Ray Kurzweil (The Age of Spiritual Machines (1999)) ●AI will unavoidably develop faster than HI (AI breakout impied by the Law of Accelerating Returns) ● Biology "in vitro„ => Biology "in silicio" ● Cyborgs => man-machine fusion, brain-computer interfaces ● Human brains uploaded to computers

╋Yuval Harari (Sapiens (2011-2014)) ●Intelligent design => DNA reprogramming (today’s CRISPR), microbots (today’s Xenobots)

( (2014)) ●Brain simulation is unavoidable ▬✝, Bill Gates, , ✝, etc. ●AI is like nuclear power => bridle it before developing it or interrupt & abandon development !

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 25 / 50 Skeptic opinions: AI will not supersede HI that easily ! (Rodney Brooks: The Seven Deadly Sins of AI Predictions (2017))

●Exponentialism & Extrapolation ≠ infinite (Gartner’s missing phase 3)

●Computers People miss framework of reference ≠ have framework of reference learn tricks learn to think know how know why are not interested make introspective self-references pursue given goals have ethics, motivations, intuition

●‼ When a scientist feels that something is possible, s/he is mostly right When s/he thinks it is impossible, s/he is possibly wrong (Arthur C. Clarke’s 1. Law)

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 26 / 50 Skepticism: What is AI lacking ? Computers Animals, children (Y. Harari (2015), S. Dehaene (2014), B. Smith (2019)) Processing ●Does intelligence require the biology of a body ?

Communication, memory, processing (learning, abstraction, generalization, logical thinking, prediction, Creativity Perception problem solving, risk evaluation, decision making, etc.) Emergent properties ? (G. Tononi’s Information Integration Potential Φ = Judgement how much parts of a system know about one another) ?

Goals, mission, role, meaning, convictions, values, ethics, ?  CAUSALITY: PAST FUTURE Conscience Unconsciousness Insight, intuition, wishes, responsibility, Ego, self-reflection, subjectivity, etc. Sensibility ●How should such missing human capabilities be (1) defined and (2) programmed ?

●Intelligence X ontological (world-)models Instinct, inclinations, etc.

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 27 / 50 Orthogonal opinion: HI (& AI?) as Bayesian inference machines? (Karl Friston: A Free Energy Principle for Biological Systems (2012))

●Free energy = mathematically computable energy difference between perceived and expected state

●Free energy principle => minimization of own free energy, i.e. of the gap between perceived reality and expectations

= Life organizing principle above thermodynamics & Darwinian evolution principles

=> HI (& AI ?) would be «simply» reactive => unpredictable ! ●Partly causal: they follow the laws of physics, chemistry, biology, etc. ●Partly final: they aim to minimize free energy ●But also partly random: they adapt to their environment => They are chaotic2 = reactive to chaotic forecasts

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 28 / 50 Agenda

●What is AI ?

●Computability limits => intelligence limits ?

●The Singularity – if AI > HI then AI >> HI ?

●AI Risks

●Metaphysical ruminations – the future of mankind and intelligence ?

●Outlook

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 29 / 50 Risk no. 1: From end of privacy to manipulation of opinions

●The hope was a free, open, democratic Internet against overly powerful states

●The reality is the invasion of privacy by the oligarchy of the unelected GAFAM and totalitarian states

●Peoples‘ opinions are analysed => algorithmically influenced => manipulated

•=> Societal regulation through surveillance & digital control

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 30 / 50 Risk no. 2: What is still credible and what is not ?

●Jordan Peele falsified, synthetized, synchronized Obama‘s face and speech https://www.youtube.com/watch?v=cQ54GDm1eL0

●Such deepfakes are increasingly easy to create with AI help but increasingly difficult to detect

●Anyone can create & spread them without the liability of professional editors

●They are more sensational than reality => they encourage polarization

=> They threaten democracy even more than textual fake news that can relatively simply be checked

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 31 / 50 Risk no. 3: AI weapons & cyberwar

●Development of autonomous robotic AI weapons (e.g. drones)

●Powerful lobbies always want to sell more weapons ●Mankind was never able to give up using such «toys» ●Much larger potential for mass destruction than with atomic/bio/chemical weapons

●Cyberwar races between enemy AI algorithms

●Become always easier, cheaper, stealth, less risky than air/sea/land combat https://www.nytimes.com/2019/06/15/us/politics/trump-cyber-russia-grid.html

●•Aim at «SCADA» infrastructure (e.g. RU vs. Georgia, RU vs. Eastland, US vs. Iran) Public utilities systems: energy, transport, telecomm, water, payment, health, etc.

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 32 / 50 Risk no. 4: Job destruction & digital chasm (1)

10 dominant countries according to labor force (ca. 60% of the global labor force) A = Agriculture, primary sector I = Industry, secondary sector S = Services, tertiary sector

% global % % % % delta S over 2020 USA 2020 Land labor force A I S 25 years

(A) Agriculture: China 21.0 49 22 29 191 Value creation from nature (I) Industry: India 17.0 60 17 23 28 Value creation from products USA 4.8 1 23 76 21 (S) Services: Indonesia 3.9 45 16 39 35 Value creation from intelligence Brasil 3.0 20 14 66 20 Russia 2.5 10 22 68 38 Japan 2.4 4 28 68 40 70% of all existing jobs could evaporate Nigeria 2.2 70 10 20 30 in industry, packing, transport, utilities, Bangladesh 2.2 63 11 26 30 finance, health, justice, administration, etc. Germany 1.4 3 33 64 44 >50% (S) Services Quelle: http://www.nationmaster.com

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 33 / 50 Risk no. 4: Job destruction & digital chasm (2) (Klaus Schwab – The Fourth Industrial Revolution (2016))

●Information storage, transport, replication and marginal costs => 0 Wealth is created with fewer workers => income goes to capital owners vs. workers

●Labor market and wealth distribution Milankovic‘s elephant curve (see also H.Rosling: Factfulness (2018)) Industrial countries => higher ●Universal Basic Income / Services ? Gini Index ●What is „basic“ ? Developing countries => Lower Gini Index ●Who will pay for that ?

•●The problem is less unemployment than irrelevance / uselessness (Human robots > human vegetables)

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 34 / 50 Risk no. 4: Job destruction & digital chasm (3) (Source : https://www.forbes.com/real-time-billionaires/)

60% Wealth distribution Jeff Bezos’ Salary = $81’840 5E+09 Net worth Net worth 50% Switzerland global Income / loss = 4,5E+09 10K in 4 sec. 4E+09 150K / minute 3,5E+09 30% 200M / day 3E+09

2,5E+09 25%

2E+09 15% 9% 2186 billionaires 1,5E+09 Bill Gates @ 100B together Jeff Bezos @ 140B 1E+09 2000 160 @ 10T Net worth global 500000000 Net worth Switzerland 0 < 10 K 10 - 100 K 100 K - 1M 1 - 10 M 10 - 100 M 100 M - 1 B 1 - 10 B 10 - 100 B 100 B - 1 T 10 T 10K 100K 1M 10M 100M 1B 10B 100B 1T USD / CHF (1mm) (1000km) annual income of world-class celebrities (art, show biz, sports, … "golden boys") annual income of top manager (5-10M) annual income of top professional (1-5M) annual salary of average Swiss / American = 64K annual salary of average Amazon warehouse worker

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 35 / 50 Risk no. 4: Job destruction & digital chasm (4)

●„Risks are socialized – Profits are privatized“ ●Growing inequality, segregation, fragmentation, tensions, unrest, extremism https://mkorostoff.github.io/1-pixel-wealth/?fbclid=IwAR1i2e7TnOyIiApsI4ar9sYWsv2UHhFJoj4DJYl-or2NmcpmiOreKK0Ne2w

●Same problem within as between countries („Reshoring“) (https://www.abebooks.com/Military-Conversion-Impact-Science-Technology-Digumarti/209697465/bd) (See also Prof. Ian Angell, LSE, 1995, “It was the best of times, it was the worst of times“)

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 36 / 50 Risk no. 5: AI Algorithms skid out of control

●Ashby’s Law ●A control system needs to be as complex as what it aims to control

●AI can solve problems that HI cannot (https://www.theguardian.com/te7chnology/2018/aug/29/coding-algorithms-frankenalgos-program-danger)

●BUT: AIs can neither explain nor verify their programs

●AND: Our justice system requires proof of intent or negligence (https://www.technologyreview.com/s/613578/ai-algorithms-liability-human-blame/)

•●AI could be misused by irresponsible individuals

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 37 / 50 OECD Principles for AI 1. AI should benefit people and the planet by driving inclusive growth, sustainable development and well-being. 2. AI systems should be designed in a way that respects the rule of law, human rights, democratic values and diversity, and they should include appropriate safeguards —for example, enabling human intervention where necessary— to ensure a fair and just society. 3. There should be transparency and responsible disclosure around AI systems to ensure that people understand AI-based outcomes and can challenge them. 4. AI systems must function in a robust, secure and safe way throughout their life cycles and potential risks should be continually assessed and managed. 5. Organizations and individuals developing, deploying or operating AI systems should be held accountable for their proper functioning in line with the above principles.

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 38 / 50 Agenda

●What is AI ?

●Computability limits => intelligence limits ?

●The Singularity – if AI > HI then AI >> HI ?

●AI Risks

●Metaphysical ruminations – the future of mankind and intelligence ?

●Outlook

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 39 / 50 Whereto with AI? (M.Tegmark (2019), S.Russel (2019))

●AI could be biggest achievement … or damnation of mankind ? success … or total failure ?

●The issue is not whether AI becomes conscious or not … but whether it becomes excessively competent

•●7 questions 1. Do we want superintelligence? 2. Do we want people to survive, be uploaded, simulated, eliminated? 3. Do we want people to remain in control? 4. Do we want AIs to become conscious or not? 5. Do we want to maximize the positive, minimize the negative, or leave the choice to a benevolent AI? 6. Do we want to colonize space? 7. Do we want a civilization with purpose or not? Already in 1532 Rabelais wrote in his famous Pantagruel: „Science without conscience is but destruction of the soul“

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 40 / 50 AI goal definition and control – A negative proof! (M.Tegmark (2019), S.Russel (2019)) ●Goal specification: Explicit / objective goals vs. implicit / subjective values Caveat: instrumental goals, e.g. AI self-preservation (like any Corp.!)

●Goal bias: Whose goals (individual, collective)? 1 almighty AI / few ‘coopeting’ powerful AIs / many competing AIs All AIs are inevitably biased – by their programmers –> a super-HI ?

●Goal evolution: Neither HI nor AI can know everything => Tolerance for fuzziness, randomness, shortsightedness, change => Need to adapt to changing times and values

●•Goal preservation: HI control vs. AI take-over => Make HI preservation over AI preservation an explicit goal otherwise flipping the power switch is no longer an option! The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 41 / 50 Examples of control rules

●Asimov’s Laws (1942!): (AI) robots must … ●… not hurt humans ●… follow humans‘ instructions ●… protect their own existence

●EU GDPR (2018): an algorithm must not make decisions about humans

•●Stuart Russel proposal (2019): 3 goals 1. Aim for human preferences

2. Discover human preferences

3. … through observation of human behavior = unsupervised learning !

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 42 / 50 Political observations

●Globalization & social networks: cultural boundaries replace national borders But global electronic networks do not replace local human networks

●Technology is global: there is only one common world civilization “STEM”, climate, resources, economy, sport, art, etc. are global – follow the same rules everywhere

●Global problems => global solutions Nationalism is no global solution: village -> city -> region -> state-> federation -> world perspective

•●From war to peace? Genetic law of the jungle replaced by humanistic law of peace? Terrorism is not a severe risk – the real risk is overplaying it (mosquitos cannot create havoc but bitten buffalos can) Atomic war -> Biological war -> Cyberwar are real risks … but Waring costs increase while benefits decrease: information / know-how are the new oil => peace benefits increase

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 43 / 50 Philosophical questions on ethics

●“STEM” & AI provide no cultural / ethical guidelines

●Culture, ethics, mission, meaning, identity always came from religions, philosophies, ideologies, …

●Philosophy & ideologies always dealt with tensions between two extremes ● Individualism (Darwinian evolution) Collectivism (sometimes totalitarian) Natural selection of the fittest Equal rights and duties for all Total liberty of the individual Collectivity has priority over individuals No central coordination Central control

●Social-democratic humanism seems to be the only viable compromise between them, though it is threatened

●My favorite motto on that: Someone‘s freedom stops where someone else‘s starts

●My associated guideline: Universally shared reality: suffering => Reciprocity: don‘t do to others what you wouldn‘t want them to do to you

•●Can AI follow the path of philosophy: From logic -> to probability -> to causality ? The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 44 / 50 Philosophical questions on technology, society, biology

●Should AIs get the same rights as humans ? Legal identity, human rights, property rights, patents rights, etc.

●Quantum physics uncertainty principle + Gödel’s incompleteness theorems => fundamental limits to understanding and predicting things ? => fundamental limits to (artificial) intelligence ?

●Given uncertainty / unpredictability, can the universe have any directed purpose ?

●Does feeling of free will exist only because we cannot detect directed purpose ?

•●J. Hersch Can an information message such as DNA have neither origin (sender) nor destination (receiver)? The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 45 / 50 Agenda

●What is AI ?

●Computability limits => intelligence limits ?

●The Singularity – if AI > HI then AI >> HI ?

●AI Risks

●Metaphysical ruminations – the future of mankind and intelligence ?

●Outlook

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 46 / 50 To wrap up

●“This is the future we are sleepwalking into and public administrations remain in autopilot mode”

●“To err is human but to really mess things up takes a computer“ 

●AI = existential potential vs. existential risk Existential risk = probability of failure x potential damage => very hard to estimate

●The nuclear war risk is immediately clear – The AI risk is underestimated => Urgent requirement: a “Geneva Convention for AI” => global AI guidelines China & USA err in opposite directions: totalitarianism vs. unbridled liberalism (The Big Nine: How the Tech Titans and Their Thinking Machines Could Warp Humanity)

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 47 / 50 AI risks need to be balanced against AI potential (Hans Rosling, et al. (2018), S. Pinker (2019)) The media report what sells: scary / negative news To be fair there are still major existential risks But the world actually fares better than 100 years ago Life expectation Financial collapse following …

Health 1. A global     Nutrition https://www.foreignaffairs.com/articles/united-states/2020-05- Wealth 21/coronavirus-chronicle-pandemic-foretold Security 2. Atomic- / Bio- / Cyberwar Peace 3. Climate & environmental crisis https://newrepublic.com/article/135684/declare-war-climate- Democracy change-mobilize-wwii Equal rights https://www.technologyreview.com/s/613343/lessons-from-a- Knowledge genocide-can-prepare-humanity-for-climate-/ Quality of life http://nymag.com/intelligencer/2017/07/climate-change-earth-too- hot-for-humans.html Happiness (?) 4. Increasing wealth chasm

5. Corruption Immediate AI Risks remain minor in comparison and AI potential against above risks is huge

Closing thought: Max Frisch question to TU Berlin, 1987: Dinosaurs lived 250M years; how do you envision an economy growing for 250M years?

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 48 / 50 Bibliography

● (2016) Calum Chace: The Economic Singularity: Artificial intelligence and the death of capitalism ● (2016) Kevin Kelly: The Inevitable – ● (1910) Karl Jaspers: Einführung in die Philosophie Understanding the 12 Technological Forces That Will Shape Our Future ● (1944) Erwin Schrödinger: Was ist Leben? ● (2016) Klaus Schwab: The Fourth Industrial Revolution ● (1970) Jacques Monod: Le Hasard et la Nécessité ● (2016) Yuval Noah Harari: Homo Deus ● (1976) Richard Dawkins: The Selfish Gene ● (2017) : Life 3.0 ● (1979) Douglas Hofstadter: Gödel, Escher, Bach: An Eternal Golden Braid ● (2018) BVBA OpenMind: The Age of Perplexity: ● (1988) Stephen Hawking: A Brief History of Time: Rethinking the World We Knew From the Big Bang to Black Holes ● (2018) David Christian: A Big History of Everything ● (1989) Roger Penrose: The Emperor's New Mind: ● (2018) Hans Rosling, et al.: Factfulness Concerning Computers, Minds and The Laws of Physics ● (2018) Steven Pinker: Enlightenment Now: ● (1993) Jeanne Hersch: L’Etonnement Philosophique The Case for Reason, Science, Humanism, and Progress ● (1994) Roger Penrose: Shadows of the Mind: ● (2018) Yuval Noah Harari: 21 Lessons for the 21st Century A Search for the Missing Science of Consciousness ● (2019) BBVA Openmind: Towards a New Enlightenment? ● (1997) Clayton Christensen: The innovator's dilemma: A Transcendent Decade when new technologies cause great firms to fail ● (2019) Brian Cantwell Smith: The Promise of Artificial Intelligence: ● (1999) Ray Kurzweil: The Age of Spiritual Machines Reckoning and Judgment ● (2008) Leonard Susskind: The Black Hole War ● (2019) Augustin Rayo: On the Brink of Paradox: ● (2011) James Gleick: The Information Highlights from the Intersection of Philosophy and Mathematics ● (2014) Nick Bostrom: Superintelligence ● (2019) Stuart Russel: : ● (2014) Stanislas Dehaene: Consciousness and the Brain: Artificial Intelligence and the Problem of Control Deciphering How the Brain Codes Our Thoughts ● (2019) Harald Welzer: Alles könnte anders sein: ● (2014) Yuval Noah Harari: Sapiens Eine Gesellschaftsutopie für freie Menschen ● (2019) Lê Nguyên Hoang & El El Mhamdi: Le fabuleux chantier Rendre l'intelligence artificielle robustement bénéfique

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 49 / 50 Other sources

Articles MIT AI Interviews (Lex Fridman, YouTube) ● David Kirsch: Today the earwig, man tomorrow (1991) ●Christoph Koch ● Karl Friston: A free energy principle for biological systems (2012) ●Eric Weinstein ● Oren Etzioni: Are the experts worried about ●Ian Goodfellow the existential risk of artificial intelligence? (2016) ●Lisa Feldman ● Kieran Healy: The moral economy of tech (2016) ●Max Tegmark ● Rodney Brooks: The seven deadly sins of ai predictions (2017) ●Sean Carroll ● Eve Smith: Silicon Valley, we have a problem (2018) ●Stephen Wolfram ●Steven Pinker Magazines ●Stuart Russel ● MIT Technology Review ●Tomaso Poggio ● Wired ●…. ● American Scientist ● The Atlantic Videos ● NYT ●David Autor ● WSJ ●Andrea Glorioso ● The Economist ●ARTE iHuman ● The Guardian ●TED Talks about AI by the above Authors ● BBC News ● …

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 50 / 50 Acknowledgements

●C. Andres ●Prof. G. Dandy ●B. Gaetani ●B. Gilliand ●Dr. P. Guéret ●C. Rudow for their patient reviews, ●U. Höhn ●C. Janson friendly comments, and ●N. Janson constructive criticisms ●P. Janson ●Dr. H. Rudin ●Dr. L. Svobodova ●C. Trachsel ●M. Waldesbühl ●Dr. D. WebbThe Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 51 / 50 Thank you!

Questions?

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 52 / 50 BACKUP

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 53 / 50 Fangen wir mit Bits an – Warum? Woher?

●In China: 214 graphische Motive (= Schlüssel) => ~ 100.000 Ideogramme

●Im Westen: ±26 Buchstaben im Alphabet => ~ 1.000.000 Worte in ? Sprachen

●In Mathematik: 10 Indo-Arabische Ziffer => unendlich viele Zahlen

●•In Computern: 2 Bits => alles: Zahlen, Buchstaben, Bilder, Ton, Video, usw.

= Binary digits: 0 und 1 (elektrisch: Spannung & keine Spannung)

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 54 / 50 Information «Komplexität» wird an Entropie gemessen

●Thermodynamik

●1. Gesetz: Energie Menge bleibt im Universum erhalten

●2. Gesetz: Energie Qualität wird aber immer schlechter («verlorene chaotische» Wärme)

●Entropie = Messung der Verschlechterung der Energie (wird immer grösser)

●Gravitation & Leben schaffen gegen Entropie => schaffen Struktur aus Chaos !

●•Informatik

●Entropie = Messung der Komplexität einer Informationsmeldung

“ABCD” Göthe Werke The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 55 / 50 => Information ist physisch 2-5 Billionen Galaxien

●Information  Zustand irgendeiner physischen Entität

●Inhalt des Universums / eines schwarzes Lochs 46 Milliarden LJ ist als holografische Information beschrieben 1081 Atome die an deren äusseren Flächen gespeichert ist

●Information löschen / vernichten braucht Energie sonst würde die Entropie des Universums abnehmen x Milliarden Sternen / Galaxie =>• Masse  Energie  Entropie ≈ Entropie  Information ist physisch (!)

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 56 / 50 Gödel’s Unvollständigkeit Theoreme in Kürze

●Stellen wir uns eine Maschine vor, die alle wahre und nur wahre Aussagen drucken kann Mögliche Deklarationen sind W*x bedeutet ich kann x drucken F*x bedeutet ich kann x nicht drucken => z.B. W*xyzxyz = WW*xyz WW*x bedeutet ich kann xx drucken FF*x bedeutet ich kann xx nicht drucken ●Gödel’s Theoreme sagen, dass es keine solche Maschine geben kann, die gleichzeitig ● Konsistent ist (nur Wahrheiten druckt) und ● Komplett ist (alle Wahrheiten drucken kann) ●Intuitiver Beweis: Fragen wir die Maschine, ob sie FF*FF* drucken kann oder nicht ●Wenn die Maschine FF*FF* druckt, dann ist sie inkonsistent, denn FF*FF* bedeutet dass sie das nicht drucken kann ●Wenn die Maschine FF*FF* nicht druckt, dann ist sie inkomplett, denn FF*FF* ist eine Wahrheit, die sie erkennen soll ●Gödel’s geniale Beobachtung ●Wenn eine Maschine FF*x drucken kann, kann sie auch das Gegenteil FF*FF* drucken ●FF*FF* mathematisch auszudrucken ist wahnsinning schwierig … aber das hat Gödel’s Genie geschaffen => Es gibt kein formelles System das gleichzeitig konsistent und komplett sein kann!

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 57 / 50 Meinungen zur Singularität

Auswirkung

Pessimisten: W.Joy, W.Gates, E.Musk, S.Hawking

Vorsichtigen: Skeptischen: N.Bostrom, S.Russel R.Brooks, H.Moravec, M.Tegmark, B.Smith D. Kirsch

Optimisten: L.Page, R.Kurzweil, Y.Harari, S.Pinker, Y.LeCunn Zeit Irgendwann Nie

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 58 / 50 Risiko 1: Voreingenommene oder gefälschte Informationen

¡«Deepfakes»!

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 59 / 50 Biologie -> Soziologie -> Technologie: Strukturen

Biologie Soziologie Technologie Zellen Religionen Soziale Netzwerke ? Organen Philosophien ? Individuen Ideologien Liberalismus ? Stämme Politik Geld / GAFAM Hierarchische Strukturen Autonomische Strukturen Dezentralisierte Strukturen    Dezentralisierte Steuerung Zentralisierte Steuerung Zentralisierte Steuerung ?

=>• Google & Facebook kennen uns besser als wir selbst => Zivilisation als gigantischer Computer

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 60 / 50 Selbstreferenzen  Seltsame Schleifen  Hierarchische Durchkreuzungen (Douglas Hofstadter – Gödel, Escher, Bach – Ein ewiger goldener Zopf (1979) Domain Reich ● Vieles ist im Universum hierarchisch Galaxien, Stamm Philosophie (Aristoteles, Leibnitz, usw.) Organisationen Wissenschaften Nebel Klasse Politische Ordnung «Gott» ???? Familie Militärische Gattung Seele, Geist, Unbewusstsein Wirtschaftliche Sterne Art / Spezies Verstand, Urteil usw. Bewusstsein Asteroïden, Wesen … Planeten Organe Menschen Seltsame Schleifen: Kristallen Zellen nur aus höheren Ebenen Molekülen Tiere sichtbar Atome Pflanzen ∞ ∞ Elektronen, Protonen, Neutronen Teilchen: Quarks, Photonen∞, usw. Materie ●Möbius Streifen ● Kleinsche Flaschen •● Software ∞

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 61 / 50 Selbstreferenzen  Seltsame Schleifen  Hierarchische Durchkreuzungen (Douglas Hofstadter – Gödel, Escher, Bach – Ein ewiger goldener Zopf (1979)

●Kurt Gödel: Unvollständichkeit der Mathematik: konsistent oder komplett, nicht beides ●Maurits Escher:

●J.-S. Bach: Krebskanon • Contrapunctus XIV im letzten Stück der Kunst der Fuge The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 62 / 50 Weitere seltsame Schleifen ?

Werner Kurt Alan Claude Andrei Greg Heisenberg Gödel Turing Shannon Kolmogorov Chaitin 1901-1976 1906-1978 1912-1954 1916-2001 1903-1987 1947- Bemessung Beweisbarkeit Berechenbarkeit Information Struktur

Unschärferelation Unvollständigkeit Unentscheidbarkeit Entropie Komplexität • Wenn so viele Dinge nicht messbar, beweisbar, berechenbar sind, wie soll eine KI Fragestellungen lösen, die MI grundsätzlich nicht lösen kann? => Die Singularität scheint weit entfernt wenn überhaupt erreichbar

The Limits of AI – Salvation or Damnation of Mankind © P. Janson 2020 63 / 50