The Myth of Artificial Intelligence: Why Computers Can't Think the Way We Do
by Erik J. Larson · 5 Apr 2021
-25993-5 (PDF) The Library of Congress has cataloged the printed edition as follows: Names: Larson, Erik J. (Erik John), author. Title: The myth of artificial intelligence : why computers can't think the way we do / Erik J. Larson. Description: Cambridge, Massachusetts : The Belknap Press of Harvard University Press, 2021. | Includes
…
bibliographical references and index. Identifiers: LCCN 2020050249 Subjects: LCSH: Artificial intelligence. | Intellect. | Inference. | Logic. | Natural language processing (Computer science) | Neurosciences. Classification: LCC Q 335 .L37 2021 | DDC 006.3—dc23 LC record
…
But as Turing no doubt knew, if this were true, then so too was at least the possibility of artificial intelligence. Thus, between 1938 and 1950, Turing had a change of heart about ingenuity and intuition. In 1938, intuition was the mysterious “power of selection” that helped mathematicians decide which systems to work with
…
vast mathematical possibilities. To Turing, this suggested that intuition could be embodied in machines. In other words, the success at Bletchley implied that perhaps an artificial intelligence could be built. To make sense of his line of thought, however, some particular idea about “intelligence” had to be settled on. Intelligence as
…
conveniently absolves individual scientists from the responsibility of needing to make scientific breakthroughs or develop revolutionary ideas. Artificial intelligence just evolves, like we did. We can call the futurists and AI believers in this camp evolutionary technologists, or ETs. The ET view is popular among new age technologists like Wired cofounder Kevin
…
of times. A couple decades later, Good thought he’d found the mechanism: the digital computer. UCLA computer scientist and Hugo award-winner Vernor Vinge introduced “Singularity” into computation, and specifically into artificial intelligence, in 1986, in his science-fiction book Marooned in Realtime.3 In a later technical paper for NASA, Vinge channeled
…
The field was hyped from the get-go. The conference proceedings themselves said it all: We propose that a 2 month, 10 man study of artificial intelligence be carried out during the summer of 1956 at Dartmouth College in Hanover, New Hampshire. The study is to proceed on the basis of the
…
Comte’s nineteenth-century idea of a utopian technoscience but with the twentieth-century obsession with building more and more powerful technologies, culminating in the grand project of, in effect, building ourselves—artificial intelligence. This project would not make sense if the traditional notions of the meaning of humanity had remained intact
…
ever more advanced technologies. Once embarked on this route, it is a short journey to artificial intelligence. And here is the obvious tie-in with the intelligence errors first made by Turing and then extended by Jack Good and others up to the present day: the ultimate triumph of Homo faber as a species is
…
equating a mind with a computer is not scientific, it’s philosophical. THE FOLLY OF PREDICTION As Stuart Russell points out, in the quest for artificial intelligence, we shouldn’t bet against “human ingenuity.”2 But in a similar vein we shouldn’t make hopeful (or dire) predictions without a sound scientific
…
in a premise, to varying degrees acknowledged, that successes on narrow AI systems like playing games will scale up to general intelligence, and so the predictive line from artificial intelligence to artificial general intelligence can be drawn with some confidence. This is a bad assumption, both for encouraging progress in the field toward
…
robotics company Robust.AI, who coauthored with computer scientist Ernest Davis the 2019 Rebooting AI: Building Artificial Intelligence We Can Trust.9 Marcus and Davis make a compelling argument that the field is yet again overhyped, and that deep learning has its limits; some fundamental advance will be required to achieve generally intelligent
…
modern guise, as well. Stuart Russell, who coauthored the definitive textbook introduction to AI with Google’s Peter Norvig, argues in his 2019 Human Compatible: Artificial Intelligence and the Problem of Control that intelligence means nothing more than achieving objectives—providing a definition that includes not only humans
…
get us into much trouble with reliance on data-driven induction, but driverless cars and other critical technologies certainly can. A growing number of AI scientists understand the issue. Oren Etzioni, head of the Allen Institute for Artificial Intelligence, calls machine learning and big data “high-capacity statistical models.”9 That’s impressive computer science, but
…
discussion of automating deductive syllogisms. At the end of the paper, tying up loose ends, Peirce comments on the possibility of what we now call Artificial Intelligence. “Every reasoning machine, that is to say, every machine, has two inherent impotencies. In the first place, it is destitute of all originality, of
…
Ideas have consequences. In the following chapters, I hope to convince you that the consequences of the myth of artificial intelligence pose a significant and even grave threat to the future of scientific discovery and innovation—and, ironically, to progress in the field of AI itself. This final section is about our future, but we
…
of the first exa-scale supercomputer.5 This blurring of the line between computing and information technology and neuroscience research is typical of both projects. Given the stated goals of the Big Brain initiatives, the focus on artificial intelligence concepts and techniques is, of course, necessary. Both projects confront what is known as the
…
animate much of the vision of Data Brain projects reproducing the human mind in silica are hopelessly general and unusable. The theories themselves are of very little use (ironically) to computer science or Artificial Intelligence engineering efforts, as they don’t tell us enough about what the brain is actually doing when
…
INTELLIGENCE A popular theory of intelligence has been put forth by computer scientist, entrepreneur, and neuroscience advocate Jeff Hawkins. Famous for developing the Palm Pilot and as an all-around luminary in Silicon Valley, Hawkins dipped his toe into the neuroscience (and artificial intelligence) waters in 2004 with the publication of On Intelligence, a bold
…
consequence of inductive systems masquerading as a path to intelligence. Russell points to the “alignment” problem, an issue in AI of suddenly central importance, concerned with aligning current and future AI systems with our own interests and purposes. But the problem arises not, as Russell suggests, because AI systems are getting so smart
…
intelligence and value. Considering the alignment problem might give rise to considerations of augmentation—how we can best use increasingly powerful idiots savants to further our own objectives, including in the pursuit of scientific progress. IN CONCLUSION The inference framework I’ve presented in this book clarifies the project of expanding current artificial intelligence into
…
Automata, ed. Arthur W. Banks (Urbana: University of Illinois Press, 1966), fifth lecture, 78. 4. Daniel Kahneman, Thinking, Fast and Slow (New York: Farrar, Straus and Giroux, 2013). 5. Stuart Russell, Human Compatible: Artificial Intelligence and the Problem of Control (New York: Viking, 2019), 37. 6. Kevin Kelly, What Technology Wants (New York: Penguin, 2010
…
1992), ix. Chapter 5: Natural Language Understanding 1. John McCarthy, M. Minsky, N. Rochester, and C. E. Shannon, “A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence,” August 1955. 2. Gary Marcus and Ernest Davis, Rebooting AI: Building Artificial Intelligence We Can Trust (New York: Pantheon Books, 2019), 1. 3. Massimo Negrotti, ed., Understanding
…
the Artificial: On the Future Shape of Artificial Intelligence (Berlin Heidelberg: Springer-Verlag, 1991), 37. 4. See
…
as philosophically essential. 5. Hannah Arendt, The Human Condition (Chicago: Chicago University Press, 1958). Chapter 7: Simplifications and Mysteries 1. B. F. Skinner, Walden Two [1948] (Indianapolis: Hackett, 2005). 2. Stuart Russell, Human Compatible: Artificial Intelligence and the Problem of Control (New York: Viking, 2019), 8. 3. Dan Gardner, Future Babble: Why Expert Predictions
…
Quest for Real AI (Cambridge, MA: MIT Press, 2017). 11. Erik J. Larson, “Questioning the Hype About Artificial Intelligence,” The Atlantic, May 14, 2015. 12. Stuart Russell, Human Compatible: Artificial Intelligence and the Problem of Control (New York: Viking, 2019), 9. 13. Russell, Human Compatible, 41. 14. Russell, Human Compatible, 16–17. 15. Ford,
…
did indeed learn something new. Presumably this also proves that she has a conscious mind. 17. Eliezer Yudkowsky, “Artificial Intelligence as a Positive and Negative Factor in Global Risk,” in Global Catastrophic Risks, eds. Nick Bostrom and Milan M. Ćirković (New York: Oxford University Press), 308–345. 18. Jaron Lanier, You Are Not a
…
of Queensland Press, 1982), 41–42. 6. Stuart Russell, Human Compatible: Artificial Intelligence and the Problem of Control (New York: Viking, 2019), 48. 7. Gary Marcus and Ernest Davis, Rebooting AI: Building Artificial Intelligence We Can Trust (New York: Pantheon Books, 2019). 8. Marcus and Davis, Rebooting AI, 62. 9. Martin Ford, Architects of Intelligence: The
…
Rebooting AI. 15. Russell, Human Compatible. 16. Pearl, The Book of Why, 36. Chapter 11: Machine Learning and Big Data 1. Stuart Russell, Human Compatible: Artificial Intelligence and the Problem of Control (New York: Viking, 2019). 2. Tom Mitchell, Machine Learning (New York: McGraw-Hill Education, 1997), 2. 3. We trust the filters in large part because they
…
1931–5), 5.189. 8. Charles Sanders Peirce. Charles Sanders Peirce Papers. Houghton Library, Harvard University, ms 692. 9. Melanie Mitchell, Artificial Intelligence: A Guide for Thinking Humans (New York: Farrar, Straus, and Giroux, 2019). 10. Ibid. 11. Peirce Papers, ms 692. 12. Collected Papers, 5.171. 13. Work on abduction experienced a
…
degree that it actually understands human language, which is one reason why the Turing test remains unsolved, and perhaps also why AI scientists seem so keen to dismiss it. 6. Stuart Russell, Human Compatible: Artificial Intelligence and the Problem of Control (New York: Viking, 2019). 7. Martin Ford, Architects of Intelligence: The Truth
…
about AI from the People Building It (Birmingham, UK: Packt Publishing, 2018). 8. Gary Marcus and Ernest Davis, Rebooting AI: Building Artificial Intelligence We Can Trust (New York: Pantheon Books, 2019), 6–7. 9. See Gary Marcus, “Why Can’t My Computer Understand Me?” New
…
Paul M. Matthews, Rafael Yuste, and Christof Koch, “Neuroscience Thinks Big (and Collaboratively),” Nature Reviews Neuroscience 14, no. 9 (2013): 659. 11. Lee Gomes, “Machine-Learning Maestro Michael Jordan on the Delusions of Big Data and Other Huge Engineering Efforts,” IEEE Spectrum, October 20, 2014, https://spectrum.ieee.org/robotics/artificial-intelligence/machinelearning-maestro-michael-jordan-on
…
in, 155–156 AdaptWatson (Jeopardy! playing system), 223–225 affirming the consequent, 170 airplane crashes, 113–114 Alexander, Hugh, 40, 284n4 AlexNet (computer program), 165 alignment problem, 279 AlphaGo (computer program), 125, 161–162 altruism, 82 Amazon (firm), 144 anaphora, 210 Anderson, Chris, 145, 243 anomaly detection, 150–151 anti-intellectual policies
…
–216 Copernicus, Nicolaus, 104 counterfactuals, 174 creative abduction, 187–189 Cukier, Kenneth, 143, 144, 257 Czechoslovakia, 60–61 Dartmouth Conference (Dartmouth Summer Research Project on Artificial Intelligence; 1956), 50–51 data: big data, 142–146; observations turned into, 291n12 Data Brain projects, 251–254, 261, 266, 268, 269 data science, 144
…
John, 275–278 Hottois, Gilbert, 287n1 Human Brain Project, 245, 247–254, 256, 267–268 Human Genome Project, 252 human intelligence: artificial intelligence versus, 1–2; behaviorism on, 69; Data Brain projects and, 251; Good on, 33–35; infinite amount of knowledge in, 54; neocortical theories of, 263–268; as problem solving, 23
…
game, 9, 51 The Imitation Game (film), 21 incompleteness theorems, 12–15 induction, 115–121, 171–172; abduction and, 161; in artificial intelligence, 273–274; in life situations, 125–126; limits to, 278–279; machine learning as, 133; not strategy for artificial general intelligence, 173; problems of, 122–124; regularity in, 126–129 inductive inference
…
, 189 inference, 4, 104, 280–281, 283n1; abductive inference, 99–102, 162–163; in artificial intelligence, 103; combining types of, 218–219, 231; guesses
Artificial Intelligence: A Modern Approach
by Stuart Russell and Peter Norvig · 14 Jul 2019 · 2,466pp · 668,761 words
–robot interaction, of which the self-driving car is one example. The problem of achieving agreement between our true preferences and the objective we put into the machine is called the value alignment problem: the values or objectives put into the machine must be aligned with those of the human. If we are
…
et al. (2019) present a 60-page catalog of ways in which machine learning can be used to tackle climate change. These are just a few examples of artificial intelligence systems that exist today. Not magic or science fiction—but rather science, engineering, and mathematics, to which this book provides an introduction. 1.5 Risks
…
give overviews of machine learning for a general audience, and Kai-Fu Lee (2018) describes the race for international leadership in AI. Martin Ford (2018) interviews 23 leading AI researchers. The main professional societies for AI are the Association for the Advancement of Artificial Intelligence (AAAI), the ACM Special Interest Group in Artificial Intelligence (SIGAI, formerly SIGART
…
European Conference on AI (ECAI), and the AAAI Conference. Machine learning is covered by the International Conference on Machine Learning and the Neural Information Processing Systems (NeurIPS) meeting. The major journals for general AI are Artificial Intelligence, Computational Intelligence, the IEEE Transactions on Pattern Analysis and Machine Intelligence, IEEE Intelligent Systems, and the Journal of Artificial Intelligence Research. There are also many
…
in the appropriate chapters. 1In the public eye, there is sometimes confusion between the terms “artificial intelligence” and “machine learning.” Machine learning is a subfield of AI that studies
…
the ability to improve performance based on experience. Some AI systems use machine learning methods to achieve competence, but some do not. 2We are not suggesting that humans are “irrational
…
: ruled during; Properties: evil, king. The language of first-order logic, whose syntax and semantics we define in the next section, is built around objects and relations. It has been important to mathematics, philosophy, and artificial intelligence precisely because those fields—and indeed, much of everyday human existence—can be usefully thought of as dealing with
…
Reasoning. Many papers on graphical models, which include Bayesian networks, appear in statistical journals. The proceedings of the conferences on Uncertainty in Artificial Intelligence (UAI), Neural Information Processing Systems (NeurIPS), and Artificial Intelligence and Statistics (AISTATS) are good sources for current research. 1Bayesian networks, often abbreviated to “Bayes net,” were called belief networks in the 1980s
…
LeCun et al. (1995) for a comparison of approaches. Starting in the late 1990s, accompanying a much greater role of probabilistic modeling and statistical machine learning in the field of artificial intelligence in general, there was a rapprochement between these two traditions. Two lines of work contributed significantly. One was research on face detection (Rowley
…
that we need to be very careful in specifying what we want, because with utility maximizers we get what we actually asked for. The value alignment problem is the problem of making sure that what we ask for is what we really want; it is also known as the King Midas problem
…
a machine learning system, and Tygar (2011) surveys adversarial machine learning. Researchers at IBM have a proposal for gaining trust in AI systems through declarations of conformity (Hind et al., 2018). DARPA requires explainable decisions for its battlefield systems, and has issued a call for research in the area (Gunning, 2016). AI safety: The book Artificial Intelligence Safety and
…
Journal of the Association for Computing Machinery JAIR Journal of Artificial Intelligence Research JAR Journal of Automated Reasoning JASA Journal of the American Statistical Association JMLR Journal of Machine Learning Research JSL Journal of Symbolic Logic KDD Proceedings of the International Conference on Knowledge Discovery and Data Mining KR Proceedings of the International Conference on
…
Association. Academic Press. Bar-Shalom, Y., Li, X.-R., and Kirubarajan, T. (2001). Estimation, Tracking and Navigation: Theory, Algorithms and Software. Wiley. Barber, D. (2012). Bayesian Reasoning and Machine Learning. Cambridge University Press. Barr, A. and Feigenbaum, E. A. (Eds.). (1981). The Handbook of Artificial Intelligence, Vol. 1. HeurisTech Press and William Kaufmann. Barreiro, J., Boyce, M., Do, M., Frank
…
., Marcedone, A., McMahan, H. B., Patel, S., Ramage, D., Segal, A., and Seth, K. (2017). Practical secure aggregation for privacy-preserving machine learning. In Proc. ACM SIGSAC Conference on Computer and Communications Security. Bond, A. H. and Gasser, L. (Eds.). (1988). Readings in Distributed Artificial Intelligence. Morgan Kaufmann. Bonet, B. (2002). An epsilon-optimal grid-based algorithm
…
, R. V. (2004). RDF vocabulary description language 1.0: RDF schema. Tech. rep., W3C. Briggs, R. (1985). Knowledge representation in Sanskrit and artificial intelligence. AIMag, 6, 32–39. Brill, E. (1992). A simple rule-based part of speech tagger. In Proc. Third Conference on Applied Natural Language Processing. Brin,
…
on the Allais example. American Economic Review, 79, 392–407. Connell, J. (1989). A Colony Architecture for an Artificial Creature. Ph.D. thesis, Artificial Intelligence Laboratory, MIT. Conway, D. and White, J. (2012). Machine Learning for Hackers. O’Reilly. Cook, S. A. (1971). The complexity of theorem-proving procedures. In STOC-71. Cook, S. A
…
. (1992). A Bayesian method for the induction of probabilistic networks from data. Machine Learning, 9, 309–347. Copeland, J. (1993). Artificial Intelligence: A Philosophical Introduction. Blackwell. Corbett-Davies, S. and Goel, S. (2018). The measure and mismeasure of fairness: A critical review of fair machine learning. arXiv:1808.00023. Corbett-Davies, S., Pierson, E., Feller, A., Goel, S
…
. (1988a). Neural nets. Quarterly Reviews of Biophysics, 21, 365–427. Cowan, J. D. and Sharp, D. H. (1988b). Neural nets and artificial intelligence. Daedalus, 117, 85–121. Cowell, R., Dawid, A. P, Lauritzen, S., and Spiegelhalter, D. J. (2002). Probabilistic Networks and Expert Systems. Springer. Cox, I. (1993). A review of statistical data association techniques for
…
, 46, 185–209. Edelkamp, S. (2009). Scaling search with symbolic pattern databases. In Model Checking and Artificial Intelligence (MOCHART). Edelkamp, S. and Schrödl, S. (2012). Heuristic Search. Morgan Kaufmann. Edmonds, J. (1965). Paths, trees, and flowers. Canadian J. of Mathematics, 17, 449–467. Edwards, P. (Ed.). (1967). The Encyclopedia of Philosophy. Macmillan. Eiter, T., Leone
…
, 121–131. Feigenbaum, E. A., Buchanan, B. G., and Lederberg, J. (1971). On generality and problem solving: A case study using the DENDRAL program. In Meltzer, B. and Michie, D. (Eds.), Machine Intelligence 6. Edinburgh University Press. Feldman, J. and Sproull, R. F. (1977). Decision theory and artificial intelligence II: The hungry monkey. Technical report, Computer Science
…
Department, University of Rochester. Feldman, J. and Yakimovsky, Y. (1974). Decision theory and artificial intelligence I: Semantics-based region analyzer. AIJ, 5, 349–371. Feldman, M. (2017). Oak Ridge readies
…
photographs. Jama, 316, 2402–2410. Gunkel, D. J. (2018). Robot Rights. MIT Press. Gunning, D. (2016). Explainable artificial intelligence (xai). Tech. rep., DARPA. Guo, C., Goldstein, T., Hannun, A., and van der Maaten, L. (2019). Certified data removal from machine learning models. arXiv:1911.03030. Gururangan, S., Swayamdipta, S., Levy, O., Schwartz, R., Bowman, S
…
. (1998). The Lumiere project: Bayesian user modeling for inferring the goals and needs of software users. In UAI-98. Horvitz, E. J., Breese, J. S., and Henrion, M. (1988). Decision theory in expert systems and artificial intelligence. IJAR, 2, 247–302. Horvitz, E. J. and Breese, J. S. (1996). Ideal partition of resources for metareasoning. In
…
. Liang, P., Jordan, M. I., and Klein, D. (2011). Learning dependency–based compositional semantics. arXiv:1109.6841. Liang, P. and Potts, C. (2015). Bringing machine learning and compositional semantics together. Annual Review of Linguistics, 1, 355–376. Liberatore, P. (1997). The complexity of the language A. Electronic Transactions on Artificial Intelligence, 1, 13–38. Lifschitz, V. (2001
…
, J. (1973). Artificial intelligence: A general survey. In Lighthill, J., Sutherland
…
). An Introduction to Genetic Algorithms. MIT Press. Mitchell, M. (2019). Artificial Intelligence: A Guide for Thinking Humans. Farrar, Straus and Giroux. Mitchell, M., Holland, J. H., and Forrest, S. (1996). When will a genetic algorithm outperform hill climbing? In NeurIPS 6. Mitchell, T. M. (1997). Machine Learning. McGrawHill. Mitchell, T. M. (2005). Reading the web: A
…
. S. (1997). Cached sufficient statistics for efficient machine learning with large datasets. JAIR, 8, 67–91. Moore, E. F. (1959). The shortestpaththroughamaze. In Proc. International Symposium on the Theory of Switching, Part II. Harvard University Press. Moore, R. C. (1980). Reasoning about knowledge and action. Artificial intelligence center technical note, SRI International. Moore, R.
…
C. (1985). A formal theory of knowledge and action. In Hobbs, J. R. and Moore, R. C. (Eds.), Formal Theories of the Commonsense World. Ablex. Moore, R. C. and DeNero, J. (2011). L1 and L2 regularization for multiclass hinge loss models. In Symposium on Machine Learning in Speech and
…
Natural Language Processing. Moravčík, M., Schmid, M., Burch, N., Lisy, V., Morrill, D., Bard, N., Davis, T., Waugh, K., Johanson, M., and Bowling, M. (2017). Deepstack: Expert-level artificial intelligence in no-limit poker. arXiv:1701.01724. Moravec, H. P. (1983). The
…
Recipes: The Art of Scientific Computing (3rd edition). Cambridge University Press. Preston, J. and Bishop, M. (2002). Views into the Chinese Room: New Essays on Searle and Artificial Intelligence. Oxford University Press. Prieditis, A. E. (1993). Machine discovery of effective admissible heuristics. Machine Learning, 12, 117–141. Prosser, P. (1993). Hybrid algorithms for constraint satisfaction problems
…
. (2013). Intriguing properties of neural networks. arXiv:1312.6199. Szeliski, R. (2011). Computer Vision: Algorithms and Applications. Springer-Verlag. Szepesvari, C. (2010). Algorithms for reinforcement learning. Synthesis Lectures on Artificial Intelligence and Machine Learning, 4, 1–103. Tadepalli, P., Givan, R., and Driessens, K. (2004). Relational reinforcement learning: An overview. In ICML-04. Tait, P G. (1880
…
., 735, 1115 validation set, 684 validity, 240, 264 Vallati, M., 399, 1115 value (of a variable), 77 in a CSP, 164 VALUE-ITERATION, 563 value alignment problem, 23, 1054 value function, 522 additive, 533 value iteration, 562, 562–566, 585 point-based, 588 value node, see utility node value of computation, 1070
The Alignment Problem: Machine Learning and Human Values
by Brian Christian · 5 Oct 2020 · 625pp · 167,349 words
The Alignment Problem MACHINE LEARNING AND HUMAN VALUES BRIAN CHRISTIAN A Norton Professional Book This e-book contains some places that ask the reader to fill in questions or comments. Please keep pen and paper handy as you read this e-book so that you can complete the exercises within. For Peter who convinced me And for
…
the physical world. The past decade has seen what is inarguably the most exhilarating, abrupt, and worrying progress in the history of machine learning—and, indeed, in the history of artificial intelligence. There is a consensus that a kind of taboo has been broken: it is no longer forbidden for AI researchers to discuss concerns
…
the most central and most urgent scientific questions in the field of computer science. It has a name: the alignment problem. In reaction to this alarm—both that the bleeding edge of research is getting ever closer to developing so-called “general” intelligence and that real-world machine-learning systems are touching more and more ethically fraught
…
up, some of them for the first time, to sound notes of caution—and redirecting their research funding accordingly. The first generation of graduate students is matriculating who are focused explicitly on the ethics and safety of machine learning. The alignment problem’s first responders have arrived at the scene. This book is the product of
…
learn may eventually produce a system which is a true analogue of some form of biological learning.” The history of artificial intelligence is famously one of cycles of alternating hope and gloom, and the Jetsonian future that the perceptron seemed to herald is slow to arrive. Rosenblatt, with a few years of hindsight
…
science alike, was huge. An idea developed in a pure machine-learning context, inspired by models of classical and operant conditioning from psychology, had suddenly come full circle. It wasn’t just a model of how artificial intelligence might be structured. It appeared to offer a description of one of the universal principles for
…
blows his mind. “The revelation came,” he says, “when I realized that the TD model goes into what we now see, the Go programming, and this artificial intelligence, machine-learning thing. That was a revelation, where I said, My God, what have I done? You know, understanding my data came from the Rescorla-Wagner model
…
the simplest “closed-loop” control systems there is—the canonical cybernetics example, in fact—is the lowly mechanical thermostat. There is no “machine learning” involved as such. But here was the alignment problem, in full, sweat-soaked force. First: You don’t measure what you think you measure. I wanted to regulate the temperature
…
we ought rather to simply train a system with examples of things that humans believe are “fair” and “unfair,” and have machine learning construct the formal, operational definition itself.17 This is, itself, likely to be an alignment problem as subtle as any other. TRANSPARENCY In Chapter 3, we discussed an encouraging frontier of work on
…
between hundreds of dog breeds, I believe it could make a great contribution to dermatology.” This spurred the collaboration with Ko and others. See Justin Ko, “Mountains out of Moles: Artificial Intelligence and Imaging” (lecture), Big Data in Biomedicine Conference, Stanford, CA, May 24, 2017, https://www.youtube.com/watch?v=kClvKNl0Wfc. 56.
…
up to the same as the true, eventually. Balance the books” (Stuart Russell, personal interview, May 13, 2018). 45. Andrew Ng, “The Future of Robotics and Artificial Intelligence” (lecture), May 21, 2011, https://www.youtube.com/watch?v=AY4ajbu_G3k. 46. See Ng et al., “Autonomous Helicopter Flight via Reinforcement Learning,” as well
…
back to the same state, the total, the integral v . ds, is zero” (Stuart Russell, personal interview, May 13, 2018). 50. Russell and Norvig, Artificial Intelligence. 51. Ng, Harada, and Russell, “Policy Invariance Under Reward Transformations.” 52. Spignesi, The Woody Allen Companion. 53. For an evolutionary psychology perspective, see, e.g., Al-Shawaf
…
Neural Networks.” For a more detailed history of these ideas, see Gal, “Uncertainty in Deep Learning.” For an overview of probabilistic methods in machine learning more generally, see Ghahramani, “Probabilistic Machine Learning and Artificial Intelligence.” 17. Yarin Gal, personal interview, July 11, 2019. 18. Yarin Gal, “Modern Deep Learning Through Bayesian Eyes” (lecture), Microsoft Research, December
…
to cure cancer, make sure the patient still dies!” See https://intelligence.org/2016/12/28/ai-alignment-why-its-hard-and-where-to-start/. See also Armstrong and Levinstein, “Low Impact Artificial Intelligences,” which uses the example of an asteroid headed for earth. A system constrained to only take “low-impact” actions
…
Processing Systems, 9505–15, 2018. Aha, David W., and Alexandra Coman. “The AI Rebellion: Changing the Narrative.” In Thirty-First AAAI Conference on Artificial Intelligence, 2017. Akrour, Riad, Marc Schoenauer, and Michèle Sebag. “APRIL: Active Preference-Learning Based Reinforcement Learning.” In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, 116–31. Springer, 2012. Akrour
…
Needed,” LessWrong, February 16, 2012. https://www.lesswrong.com/posts/8Nwg7kqAfCM46tuHq/the-mathematics-of-reduced-impact-help-needed. ———. “Motivated Value Selection for Artificial Agents.” In Artificial Intelligence and Ethics: Papers from the 2015 AAAI Workshop. AAAI Press, 2015. ———. “Reduced Impact AI: No Back Channels,” LessWrong, November 11, 2013. https://www.lesswrong.com
…
/posts/gzQT5AAw8oQdzuwBG/reduced-impact-ai-no-back-channels. Armstrong, Stuart, and Benjamin Levinstein. “Low Impact Artificial Intelligences.” arXiv Preprint arXiv:1705.10720, 2017. Arrow, Kenneth J. “A Difficulty in the Concept of Social Welfare.” Journal of Political Economy 58,
…
Conference on Machine Learning, 144–51. ACM, 2008. Colombetti, Marco, and Marco Dorigo. “Robot Shaping: Developing Situated Agents Through Learning.” Berkeley, CA: International Computer Science Institute, 1992. Coman, Alexandra, Benjamin Johnson, Gordon Briggs, and David W. Aha. “Social Attitudes of AI Rebellion: A Framework.” In Workshops at the Thirty-First AAAI Conference on Artificial Intelligence, 2017
…
Richard Ambrosino, John Aronis, Bruce G. Buchanan, Richard Caruana, Michael J. Fine, et al. “An Evaluation of Machine-Learning Methods for Predicting Pneumonia Mortality.” Artificial Intelligence in Medicine 9, no. 2 (1997): 107–38. Cooper, Lori, and Paul Aronowitz. “DNR Tattoos: A Cautionary Tale.” Journal of General Internal Medicine 27, no. 10 (2012): 1383. Cooper
…
Department, July 8, 2016. Dietterich, Thomas G. “Steps Toward Robust Artificial Intelligence.” AI Magazine 38, no. 3 (2017): 3–24. Diuk, Carlos, Andre Cohen, and Michael L. Littman. “An Object-Oriented Representation for Efficient Reinforcement Learning.” In Proceedings of the 25th International Conference on Machine Learning, 240–47. ACM, 2008. Doctorow, Cory. “Two Years Later
…
thesis, University of Alberta, 2013. Gergely, György, Harold Bekkering, and Ildikó Király. “Rational Imitation in Preverbal Infants.” Nature 415, no. 6873 (2002): 755. Gershgorn, Dave. “Companies Are on the Hook If Their Hiring Algorithms Are Biased.” Quartz, October 22, 2018. Ghahramani, Zoubin. “Probabilistic Machine Learning and Artificial Intelligence.” Nature 521, no. 7553 (2015): 452–59. Ghorbani
…
Artificial Intelligence 33 (2019): 3681–88. Gielniak, Michael J., and Andrea L. Thomaz. “Generating Anticipation in Robot Motion.” In 2011 RO-MAN, 449–54. IEEE, 2011. Giusti, Alessandro, Jérôme Guzzi, Dan C. Cireşan, Fang-Lin He, Juan P. Rodríguez, Flavio Fontana, Matthias Faessler, et al. “A Machine Learning Approach to Visual
…
Conference on Machine Learning, 1861–69. 2015. Hersher, Rebecca. “When a Tattoo Means Life or Death. Literally.” Weekend Edition Sunday, NPR, January 21, 2018. Hester, Todd, Matej Večerík, Olivier Pietquin, Marc Lanctot, Tom Schaul, Bilal Piot, Dan Horgan, et al. “Deep Q-Learning from Demonstrations.” In Thirty-Second AAAI Conference on Artificial Intelligence, 2018
…
G. Dietterich, Alan Fern, and Dan Hendrycks. “Open Category Detection with PAC Guarantees.” In Proceedings of the 35th International Conference on Machine Learning, 2018. Liu, Si, Risheek Garrepalli, Alan Fern, and Thomas G. Dietterich. “Can We Achieve Open Category Detection with Guarantees?” In Workshops at the Thirty-Second AAAI Conference on Artificial Intelligence, 2018. Lockhart, Ted
…
-Controlled Cars,” 1969. http://www-formal.stanford.edu/jmc/progress/cars/cars.html. ———. “What Is Artificial Intelligence?” 1998. http://www-formal.stanford.edu/jmc/whatisai.pdf. McCarthy, John, and Edward A. Feigenbaum. “In Memoriam: Arthur Samuel: Pioneer in Machine Learning.” AI Magazine 11, no. 3 (1990): 10. McCulloch, Warren S. The Collected Works of
…
for Machine Learning. San Mateo, CA: Morgan Kaufmann Publishers, 1993. Radford, Alec, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. “Language Models Are Unsupervised Multitask Learners.” OpenAI Blog, 2019. https://openai.com/blog/better-language-models/. Ramakrishnan, Ramya, Chongjie Zhang, and Julie Shah. “Perturbation Training for Human-Robot Teams.” Journal of Artificial Intelligence
…
), 101–03. Madison, WI: ACM Press, 1998. ———. “Should We Fear Supersmart Robots?” Scientific American 314, no. 6 (2016): 58–59. Russell, Stuart J., and Peter Norvig. Artificial Intelligence: A Modern Approach. 3rd ed. Upper Saddle River, NJ: Pearson, 2010. Rust, John. “Do People Behave According to Bellman’s Principle of Optimality?” Working
…
. Taylor, Jessica. “Quantilizers: A Safer Alternative to Maximizers for Limited Optimization.” In Workshops at the Thirtieth AAAI Conference on Artificial Intelligence, 2016. Taylor, Jessica, Eliezer Yudkowsky, Patrick LaVictoire, and Andrew Critch. “Alignment for Advanced Machine Learning Systems.” Machine Intelligence Research Institute, July 27, 2016. Tesauro, Gerald. “Practical Issues in Temporal Difference Learning.” In Advances
…
Artificial Intelligence Research 19 (2003): 205–08. Wilson, Aaron, Alan Fern, and Prasad Tadepalli. “A Bayesian Approach for Policy Learning from Trajectory Preference Queries.” In Advances in Neural Information Processing Systems 25, 1133–41. 2012. Wirth, Christian, Riad Akrour, Gerhard Neumann, and Johannes Fürnkranz. “A Survey of Preference-Based Reinforcement Learning Methods.” Journal of Machine Learning
…
on Side Effects for Safe Optimality in Factored Markov Decision Processes.” In Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, 4867–73, IJCAI, 2018. Zheng, Zeyu, Junhyuk Oh, and Satinder Singh. “On Learning Intrinsic Rewards for Policy Gradient Methods.” In Advances in Neural Information Processing Systems, 4644–54. 2018.
…
Maximum Causal Entropy.” In Proceedings of the 27th International Conference on Machine Learning, 1255–62. 2010. Ziebart, Brian D., Andrew L. Maas, J. Andrew Bagnell, and Anind K. Dey. “Maximum Entropy Inverse Reinforcement Learning.” In Proceedings of the Twenty-Third AAAI Conference on Artificial Intelligence, 1433–38. AAAI Press, 2008. Ziegler, Daniel M., Nisan
…
method, 6–7, 37–38, 316, 342–43n68, 397n13 parameters, 18 parentese (motherese), 269, 385–86n45 parenting alignment problem and, 166 human-machine cooperation and, 269, 385–86n45 incentives and, 165–66, 170 reinforcement learning and, 155, 158, 364n15 as transformative experience, 321 See also child development parole system Burgess study, 51–54, 76 predictive
…
functions, 24 Reddy, Raj, 223 redundant encodings, 40, 64, 343n74 reinforcement learning (RL) actor-critic architecture, 138, 362n51 actualism vs. possibilism and, 238–40, 379n69 addiction and, 205–08, 374n65 alignment problem and, 151 Arcade Learning Environment, 181–83, 209, 369n4 assumptions on, 320–22, 397n23, 398n25 backward learning, 162, 365n27, 369n9 boat race
…
follows: Names: Christian, Brian, 1984– author. Title: The alignment problem : machine learning and human values / Brian Christian. Description: First edition. | New York, NY : W.W. Norton & Company, [2020] | Includes bibliographical references and index. Identifiers: LCCN 2020029036 | ISBN 9780393635829 (hardcover) | ISBN 9780393635836 (epub) Subjects: LCSH: Artificial intelligence—Moral and ethical aspects. | Artificial intelligence—Social aspects. | Machine learning—Safety measures. | Software failures. | Social values. Classification
These Strange New Minds: How AI Learned to Talk and What It Means
by Christopher Summerfield · 11 Mar 2025 · 412pp · 122,298 words
in any form without permission. You are supporting writers and allowing Penguin Random House to continue to publish books for every reader. Please note that no part of this book may be used or reproduced in any manner for the purpose of training artificial intelligence technologies or systems. VIKING is a registered trademark
…
make Aristotle look like Homer Simpson? By the 1950s, a new academic discipline had been founded with the goal of turning these dreams into reality, and the field of Artificial Intelligence (AI) research was born. In tandem, the vision of an artificial mind has inspired hundreds of novels, plays
…
running on newly powerful computers could be trained to perform next-word prediction (‘the postman delivered the _____’). One landmark paper,[*1] published in 2003 by machine-learning grandee Yoshua Bengio, set the direction of travel by training deep networks to learn semantic information from patterns among words alone. Neural networks take numbers
…
to minimize perplexity. The architects of these systems are aware of this, and have taken steps to suppress this sort of language. Attempts to trick GPT-4 into making claims of intentionality elicit a rather prim denial: As an artificial intelligence, I don’t have personal beliefs, opinions, or predictions. But I
…
physical or chemical processes in the body or brain. This idea was known as vitalism. Today, some exceptionalist arguments – including the categorical denial that a machine-learning system could ever learn to ‘think’ – invoke a sort of modern-day version of this vitalist view, in which ‘true’ humanlike cognition is animated
…
Common Crawl that was used to train GPT-3 was first screened to remove as much of the hateful or erotic content as possible, using machine-learning tools that automatically detect tell-tale words and phrases. But the main approach that is used to make models less harmful is called ‘fine-tuning
…
model thus acts like an automated human judge, telling the LLM whether its replies are acceptable or not. It allows the researchers to use a machine-learning method called reinforcement learning, where the model weights are adjusted to maximize a target numerical value (such as the score in a video game, or
…
blur the fragile line between human–human and human–computer interaction. Skip Notes *1 Elkins and Chun, 2020. *2 www.mcafee.com/blogs/privacy-identity-protection/artificial-imposters-cybercriminals-turn-to-ai-voice-cloning-for-a-new-breed-of-scam/. *3 www.theverge.com/a/luka-artificial-intelligence-memorial-roman-mazurenko-bot. *4 Skjuve
…
al., 2021. *5 www.vice.com/en/article/z34d43/my-ai-is-sexually-harassing-me-replika-chatbot-nudes. *6 Depounti, Saukko, and Natale, 2023. *7 www.thecut.com/article/ai-artificial-intelligence-chatbot-replika-boyfriend.html. *8 https://woebothealth.com/img/2023/02/Woebot-Health-Research-Bibliography.pdf. *9 www.reuters.com/article
…
about what constitutes good advice – in other words, to ‘cheat’ by twisting the problem you are trying to solve. Auto-induced distribution shift occurs when machine-learning systems use the latter approach – for example, when a recommender system manipulates what the user thinks they want, in order to game the advice-giving
…
it believes to be most worthy. A hungry monkey might value the state ‘eating fruit’ more than ‘not eating fruit’ and so decide to climb a tree to pick mangos. In machine learning, the subfield that studies how to build instrumental agents is called reinforcement learning (RL). In RL, the researcher operationalizes the
…
their arch-enemies, cybersecurity experts). AI systems that code could even be used for autonomous machine-learning research, potentially kicking off a recursive loop of AI self-improvement, in which stronger and stronger AI systems write more and more powerful versions of themselves. These are the sorts of ideas that tend to get bandied
…
AI Conference in Shanghai, China.[*1] In the field, Sutton is a legend – something like a cross between a Godfather of AI and a Grand High Wizard of Machine Learning. His groundbreaking work from the 1980s laid the foundations for the entire field of reinforcement learning, when he discovered the seminal algorithms that
…
seek to regulate it. Technology, they argue, is ‘the glory of human ambition and achievement, the spearhead of progress, and the realization of our potential’. And AI specifically is touted as a sort of panacea: We believe Artificial Intelligence can save lives – if we let it. Medicine, among many other fields, is in the
…
will seek to achieve that task by any means possible. This is the power of machine-learning methods like RL. It’s what enables the (theoretical) robotic dog discussed in Part 5 to learn, by purely trial and error, to walk by itself without ever being taught. It’s what allows ChatGPT to
…
worry that as we build more powerful agents, they will find increasingly inventive and potentially unwanted ways to solve even trivial tasks, such as seeking positions of authority or robbing banks to fund their endeavours. This is called the ‘alignment problem’, and it has been written about recently in excellent books by Stuart Russell
…
Risorgimento or the Secret Life of Butterflies, then this wouldn’t matter a jot. But compared to Italian History and Lepidopterology, the fields of machine learning and AI research are moving at close to warp speed, and quite a lot has happened over the past ten months – with more surely to come before book-launch
…
Simulate Human Samples’, Political Analysis, 31(3), pp. 337–51. Available at https://doi.org/10.1017/pan.2023.2. Bai, H. et al. (2023), ‘Artificial Intelligence Can Persuade Humans on Political Issues’. Preprint. Open Science Framework. Available at https://doi.org/10.31219/osf.io/stakv. Bai, Y. et al. (2022
…
Oriented Deep Net Analysis in Linguistic Theorizing’. Available at https://doi.org/10.48550/arXiv.2106.08694. Belkin, M. et al. (2019), ‘Reconciling Modern Machine-Learning Practice and the Classical Bias–Variance Trade-Off’, Proceedings of the National Academy of Sciences, 116(32), pp. 15849–54. Available at https://doi.org/10.1073
…
5185–98. Available at https://doi.org/10.18653/v1/2020.acl-main.463. Bengio, Yoshua, Ducharme, Réjean, Vincent, Pascal, and Jauvin, Christian (2003), ‘A Neural Probabilistic Language Model’, Journal of Machine Learning Research 3, pp. 1137–55, www.jmlr.org/papers/volume3/bengio03a/bengio03a.pdf. Binz, M. et al. (2023), ‘Meta-Learned
…
‘Sparks of Artificial General Intelligence: Early Experiments with GPT-4’. arXiv. Available at http://arxiv.org/abs/2303.12712 (accessed 18 February 2024). Cerina, R. and Duch, R. (2023), ‘Artificially Intelligent Opinion Polling’. arXiv. Available at http://arxiv.org/abs/2309.06029 (accessed 20 October 2023). Chater, N
…
http://arxiv.org/abs/2309.13788 (accessed 6 October 2023). Chomsky, Noam (1957), Syntactic Structures, The Hague: Mouton. Christian, B. (2020), The Alignment Problem: Machine Learning and Human Values. New York: W. W. Norton & Co. Cobb, M. (2021), The Idea of the Brain: A History. London: Profile. Cristia, A. et al. (
…
Available at https://doi.org/10.1162/tacl_a_00115. Liu, T. and Low, B. K. H. (2023), ‘Goat: Fine-Tuned LLaMA Outperforms GPT-4 on Arithmetic Tasks’. Available at https://doi.org/10.48550/arXiv.2305.14201. Lobina, D. (2023), ‘Artificial Intelligence [sic: Machine Learning] and the Best Game in Town; or How Some Philosophers
…
, and the Bbs, Missed a Step’, 3 Quarks Daily, 13 February. Available at https://3quarksdaily.com/3quarksdaily/2023/02/artificial-intelligence-sic-machine-learning-and-the-best-game-in-town-or-how-some-philosophers
…
07128. Mahowald, K. et al. (2023), ‘Dissociating Language and Thought in Large Language Models: A Cognitive Perspective’. arXiv. Available at http://arxiv.org/abs/2301.06627 (accessed 16 September 2023). Marcus, G. (2020), ‘The Next Decade in AI: Four Steps Towards Robust Artificial Intelligence’. Preprint. arXiv. Available at http://arxiv.org/abs/2002
…
–33. Available at https://doi.org/10.1007/BF02478259. Metzinger, T. (2021), ‘Artificial Suffering: An Argument for a Global Moratorium on Synthetic Phenomenology’, Journal of Artificial Intelligence and Consciousness, 08(01), pp. 43–66. Available at https://doi.org/10.1142/S270507852150003X. Mialon, G. et al. (2023), ‘Augmented Language Models: A Survey’.
…
.informatik.uni-stuttgart.de/pdf/rand/ipl/P-1584_Report_On_A_General_Problem-Solving_Program_Feb59.pdf. Noy, S. and Zhang, W. (2023), ‘Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence’, Science, 381(6654), pp. 187–92. Available at https://doi.org/10.1126/science.adh2586. OpenAI (2023), ‘GPT-
…
1038/s41586-021-03854-z. Runciman, D. (2019), How Democracy Ends. London: Profile. Russell, S. (2019), Human Compatible: AI and the Problem of Control. New York: Viking. Russell, S. and Norvig, P. (2020), Artificial Intelligence: A Modern Approach, 4th edn. Hoboken, Nj: Pearson. Ryle, G. (2009), The Concept of Mind. London: Routledge. Sahlgren, M
…
. and Carlsson, F. (2021), ‘The Singleton Fallacy: Why Current Critiques of Language Models Miss the Point’, Frontiers in Artificial Intelligence, 4, 682578. Available at https://doi.org/10.3389/frai.2021.682578. Santurkar, S. et al. (2023
…
March 2024). Searle, J. (1999), ‘The Chinese Room’, in R. A. Wilson and F. C. Keil (eds.), The MIT Encyclopedia of the Cognitive Sciences, Cambridge, Ma: MIT Press. Sejnowski, T. J. (2020), ‘The Unreasonable Effectiveness of Deep Learning in Artificial Intelligence’, Proceedings of the National Academy of Sciences, 117(48), pp. 30033–8
…
A. (1961), ‘Confabulation in the Wernicke-Korsakoff Syndrome’, Journal of Nervous and Mental Disease, 132(5), pp. 361–81. Available at https://doi.org/10.1097/00005053-196105000-00001. Tegmark, M. (2017), Life 3.0: Being Human in the Age of Artificial Intelligence. London: Penguin. Terrace, H. et al. (1979), ‘Can an Ape
…
–9, 311 AI Safety Institute, 311n, 346 algorithms, 21, 38, 59, 75, 76, 99, 249–50, 263, 278, 279, 326–31 alignment, LLM, 179–238 alignment problem, 322 AlphaCode, 287 AlphaFold, 3–4, 347 AlphaGo, 4, 267 Altman, Sam, 1, 2, 5, 162, 222–3 alt-right, 181–2 American Sign
…
(API), 283–5, 290–91, 292, 295, 300, 301 Aristotle, 13, 16, 32; De Interpretatione (‘On Interpretation’), 73 Artificial General Intelligence (AGI), 1, 136, 140 Artificial Intelligence (AI) Artificial General Intelligence (AGI), 1, 136, 140 assistants, personal, 7, 227–8, 243, 246–7, 251, 263, 272, 292, 295, 297, 300, 301, 332
…
2 (goal-based learning system), 156–8, 160, 167, 268–9 brain and, 36–8 continual learning, 253 deep learning, see deep learning in-context learning, 159, 163, 164, 268, 291, 308 knowledge and, 18, 19 language, see language machine learning, 49, 90–91, 112, 152, 188, 190, 262, 267, 287, 305, 322 meta
…
75, 77, 168–9 logical positivism, 25, 35 logos, 16 London Tube map, 278–80 loneliness, chatbots and, 229 Long Short-Term Memory network (LSTM), 99, 116 longtermism, 312, 313 Luria, Alexander, 176 M machine learning, 49, 90–91, 112, 152, 188, 190, 262, 267, 287, 305, 322 Making of a Fly, The
Artificial Intelligence: A Guide for Thinking Humans
by Melanie Mitchell · 14 Oct 2019 · 350pp · 98,077 words
appreciate irony. That’s what was on my mind a few years ago, when, on my way to a discussion about artificial intelligence (AI), I got lost in the capital of searching and finding—the Googleplex, Google’s world headquarters in Mountain View, California. What’s more, I was lost inside the Google
…
(or even his children’s) lifetime, so he didn’t worry much about it. Near the end of GEB, Hofstadter had listed “Ten Questions and Speculations” about artificial intelligence. Here’s one of them: “Will there be chess programs that can beat anyone?” Hofstadter’s speculation was “no.” “There may be programs which
…
was threatening to destroy what I most cherished about humanity. I think EMI was the most quintessential example of the fears that I have about artificial intelligence.” Google and the Singularity Hofstadter then spoke of his deep ambivalence about what Google itself was trying to accomplish in AI—self-driving cars, speech recognition
…
physicist Stephen Hawking proclaimed, “The development of full artificial intelligence could spell the end of the human race.”7 In the same year, the entrepreneur Elon Musk, founder of the Tesla and SpaceX companies, said that artificial intelligence is probably “our biggest existential threat” and that “with artificial intelligence we are summoning the demon.”8 Microsoft’s cofounder
…
“summoning the demon.” What This Book Is About This book arose from my attempt to understand the true state of affairs in artificial intelligence—what computers can do now, and what we can expect from them over the next decades. Hofstadter’s provocative comments at the Google meeting were something of a wake
…
much further there is to go before our machines can argue for their own humanity. Part I Background 1 The Roots of Artificial Intelligence Two Months and Ten Men at Dartmouth The dream of creating an intelligent machine—one that is as smart as or smarter than humans—is centuries old but
…
can do.”8 Soon after, Marvin Minsky, founder of the MIT AI Lab, forecasted that “within a generation … the problems of creating ‘artificial intelligence’ will be substantially solved.”9 Definitions, and Getting On with It None of these predicted events have yet come to pass. So how far do we remain from the
…
above the anarchy to become the dominant AI paradigm. In fact, in much of the popular media, the term artificial intelligence itself has come to mean “deep learning.” This is an unfortunate inaccuracy, and I need to clarify the distinction. AI is a field that includes a broad set of approaches, with the
…
well as more complicated networks of simulated neurons, have been dubbed “subsymbolic” in analogy to the brain. Their advocates believe that to achieve artificial intelligence, language-like symbols and the rules that govern symbol processing cannot be programmed directly, as was done in the General Problem Solver, but must emerge from neural-like
…
general, we’re least aware of what our minds do best.”27 The attempt to create artificial intelligence has, at the very least, helped elucidate how complex and subtle are our own minds. 2 Neural Networks and the Ascent of Machine Learning Spoiler alert: Multilayer neural networks—the extension of perceptrons that was dismissed by Minsky
…
and Papert as likely to be “sterile”—have instead turned out to form the foundation of much of modern artificial intelligence. Because they are the basis of several of
…
for AI, when a program called AlphaGo stunningly defeated one of the world’s best players in four out of five games. The buzz over artificial intelligence was quickly becoming deafening, and the commercial world took notice. All of the largest technology companies have poured billions of dollars into AI research
…
sure that the purpose put into the machine is the purpose which we really desire.”18 Wiener’s comment captures what is called the value alignment problem in AI: the challenge for AI programmers to ensure that their systems’ values align with those of humans. But what are the values of humans
…
cognitive neuroscience from University College London in order to further his goal of building brain-inspired AI. Hassabis and his colleagues founded DeepMind Technologies in order to “tackle [the] really fundamental questions” about artificial intelligence.1 Perhaps not surprisingly, the DeepMind group saw video games as the proper venue for tackling those questions
…
reinforcement learning matures, I’ll be eagerly awaiting a dishwasher-loading robot that learns on its own, and maybe plays both soccer and Go in its spare time. Part IV Artificial Intelligence Meets Natural Language 11 Words, and the Company They Keep It’s time for a story. The Restaurant A man went into a
…
author of that article, the journalist John Markoff, wrote a careful description: “Two groups of scientists, working independently, have created artificial intelligence software capable of recognizing and describing the content of photographs and videos with far greater accuracy than ever before, sometimes even mimicking human levels of understanding.”26 FIGURE 41: Four (accurate) automatically
…
director of the National Cancer Institute), was interviewed in the 60 Minutes broadcast. Charlie Rose asked him, “What did you know about artificial intelligence and Watson before IBM suggested it might make a contribution in medical care?” Sharpless replied, “Not much, actually. I had watched it play Jeopardy!” Sharpless went
…
missing links required for future progress in AI. In the next chapter, I describe some approaches to giving machines these capabilities. 15 Knowledge, Abstraction, and Analogy in Artificial Intelligence Since the 1950s, many people in the AI community have explored ways to make crucial aspects of human thought—such as core intuitive knowledge
…
. Bill Gates, on Reddit, Jan. 28, 2015, www.reddit.com/r/IAmA/comments/2tzjp7/hi_reddit_im_bill_gates_and_im_back_for_my_third/?. 10. Quoted in K. Anderson, “Enthusiasts and Skeptics Debate Artificial Intelligence,” Vanity Fair, Nov. 26, 2014. 11. R. A. Brooks, “Mistaking Performance for Competence,” in What to Think
…
, D.C.: National Academy of Sciences, 2012). 4. McCarthy et al., “Proposal for the Dartmouth Summer Research Project in Artificial Intelligence.” 5. Ibid. 6. G. Solomonoff, “Ray Solomonoff and the Dartmouth Summer Research Project in Artificial Intelligence, 1956,” accessed Dec. 4, 2018, www.raysolomonoff.com/dartmouth/dartray.pdf. 7. H. Moravic, Mind Children: The Future
…
Voltaire (New York: Penguin Books, 1977), 225. 11. M. L. Minsky, The Emotion Machine: Commonsense Thinking, Artificial Intelligence, and the Future of the Human Mind (New York: Simon & Schuster, 2006), 95. 12. One Hundred Year Study on Artificial Intelligence (AI100), 2016 Report, 13, ai100.stanford.edu/2016-report. 13. Ibid., 12. 14. J. Lehman, J
…
the Official History of the Perceptrons Controversy.” 23. G. Nagy, “Neural Networks—Then and Now,” IEEE Transactions on Neural Networks 2, no. 2 (1991): 316–18. 24. Minsky and Papert, Perceptrons, 231–32. 25. J. Lighthill, “Artificial Intelligence: A General Survey,” in Artificial Intelligence: A Paper Symposium (London: Science Research Council, 1973). 26. Quoted in C
…
. 27. M. L. Minsky, The Society of Mind (New York: Simon & Schuster, 1987), 29. 2: Neural Networks and the Ascent of Machine Learning
…
the International Conference on Machine Learning (2012), 507–14. 2. P. Hoffman, “Retooling Machine and Man for Next Big Chess Faceoff,” New York Times, Jan. 21, 2003. 3. D. L. McClain, “Chess Player Says Opponent Behaved Suspiciously,” New York Times, Sept. 28, 2006. 4. Quoted in M. Y. Vardi, “Artificial Intelligence: Past and Future,” Communications of
…
com/3052885/mark-zuckerberg-facebook. 8. V. C. Müller and N. Bostrom, “Future Progress in Artificial Intelligence: A Survey of Expert Opinion,” in Fundamental Issues of Artificial Intelligence, ed. V. C. Müller (Cham, Switzerland: Springer International, 2016), 555–72. 9. M. Loukides and B. Lorica, “What Is Artificial Intelligence?,” O’Reilly, June 20, 2016, www.oreilly.com/ideas
…
. presidential election will recognize the pun on Bernie Sanders’s supporters’ tagline, “Feel the Bern.” 2. E. Brynjolfsson and A. McAfee, “The Business of Artificial Intelligence,” Harvard Business Review, July 2017. 3. O. Tanz, “Can Artificial Intelligence Identify Pictures Better than Humans?,” Entrepreneur, April 1, 2017, www.entrepreneur.com/article/283990. 4. D. Vena, “3
…
11, 2017, www.gsb.stanford.edu/insights/andrew-ng-why-ai-new-electricity. 3. J. Anderson, L. Rainie, and A. Luchsinger, “Artificial Intelligence and the Future of Humans,” Pew Research Center, Dec. 10, 2018, www.pewinternet.org/2018/12/10/artificial-intelligence-and-the-future-of-humans. 4. Two recent treatments of the ethical issues surrounding AI
…
to Neural Networking,” Economist, June 25, 2016. 3. M. G. Bellemare et al., “The Arcade Learning Environment: An Evaluation Platform for General Agents,” Journal of Artificial Intelligence Research 47 (2013): 253–79. 4. More technically, DeepMind’s program used what is called an epsilon-greedy method for choosing an action at each
…
. L. Samuel, “Some Studies in Machine Learning Using the Game of Checkers,” IBM Journal of Research and Development 3, no. 3 (1959): 210–29. 12. Ibid. 13. J. Schaeffer et al., “CHINOOK: The World Man-Machine Checkers Champion,” AI Magazine 17, no. 1 (1996): 21. 14. D. Hassabis, “Artificial Intelligence: Chess Match of the Century
…
York: Springer, 2003), 236. 17. Quoted in J. Goldsmith, “The Last Human Chess Master,” Wired, Feb. 1, 1995. 18. Quoted in M. Y. Vardi, “Artificial Intelligence: Past and Future,” Communications of the Association for Computing Machinery 55, no. 1 (2012): 5. 19. A. Levinovitz, “The Mystery of Go, the Ancient Game That Computers
…
Importance Weighted Actor-Learner Architectures,” in Proceedings of the International Conference on Machine Learning (2018), 1407–16. 4. D. Silver et al., “Mastering the Game of Go Without Human Knowledge,” Nature 550 (2017): 354–59. 5. G. Marcus, “Innateness, AlphaZero, and Artificial Intelligence,” arXiv:1801.05667 (2018). 6. F. P. Such et al., “Deep Neuroevolution
…
Press, 1996). 8. Marcus, “Innateness, AlphaZero, and Artificial Intelligence.” 9. G. Marcus, “Deep Learning: A Critical Appraisal,” arXiv:1801.00631 (2018). 10. K. Kansky et al., “Schema Networks: Zero-Shot Transfer with a Generative Causal Model of Intuitive Physics,” in Proceedings of the International Conference on Machine Learning (2017), 1809–18. 11. A. A. Rusu
…
.youtube.com/watch?v=n8m7lFQ3njk. 17. Hofstadter and Sander, Surfaces and Essences, 3. 18. M. Minsky, “Decentralized Minds,” Behavioral and Brain Sciences 3, no. 3 (1980): 439–40. 15: Knowledge, Abstraction, and Analogy in Artificial Intelligence 1. D. B. Lenat and J. S. Brown, “Why AM and EURISKO Appear to Work,” Artificial Intelligence 23, no. 3 (1984): 269–94.
…
from C. Metz, “One Genius’ Lonely Crusade to Teach a Computer Common Sense,” Wired, March 24, 2016, www.wired.com/2016/03/doug-lenat-artificial-intelligence-common-sense-engine, and D. Lenat, “Computers Versus Common Sense,” Google Talks Archive, accessed Dec. 18, 2018, www.youtube.com/watch?v=gAtn-4fhuWA. 3. Lenat notes
…
,” Bloomberg, Aug. 16, 2018, www.bloomberg.com/news/articles/2018-08-16/to-get-ready-for-robot-driving-some-want-to-reprogram-pedestrians. 5. “Artificial Intelligence, Automation, and the Economy,” Executive Office of the President, Dec. 2016, www.whitehouse.gov/sites/whitehouse.gov/files/images/EMBARGOED%20AI%20Economy%20Report.pdf. 6. This harks
…
. Campbell-Kelly et al., Computer: A History of the Information Machine, 3rd ed. (New York: Routledge, 2018), 80. 16. Quoted in K. Anderson, “Enthusiasts and Skeptics Debate Artificial Intelligence,” Vanity Fair, Nov. 26, 2014. 17. See O. Etzioni, “No, the Experts Don’t Think Superintelligent AI Is a Threat to Humanity,” Technology Review
…
, www.technologyreview.com/s/602410/no-the-experts-dont-think-superintelligent-ai-is-a-threat-to-humanity; and V. C. Müller and N. Bostrom, “Future Progress in Artificial Intelligence: A Survey of Expert Opinion,” in Fundamental Issues of Artificial Intelligence (Basel, Switzerland: Springer, 2016), 555–72. 18. N. Bostrom, “How Long Before Superintelligence?,” International Journal of
…
visual situations; see also Copycat artificial general intelligence, see general or human-level AI artificial intelligence: beneficial; bias in; creativity in; definition of; explainability; general or human-level; moral; origin of term; regulation of; relationship to deep learning and machine learning; “right to explanation”; spring; strong; subsymbolic; symbolic; unemployment due to; weak; winter Asimov, Isaac
…
, click here. Contents Title Page Copyright Notice Dedication Prologue: Terrified Part I. Background 1. The Roots of Artificial Intelligence 2. Neural Networks and the Ascent of Machine Learning 3. AI Spring Part II. Looking and Seeing 4. Who, What, When, Where, Why 5. ConvNets and ImageNet 6. A Closer Look at Machines That Learn 7. On Trustworthy
…
and Ethical AI Part III. Learning to Play 8. Rewards for Robots 9. Game On 10. Beyond Games Part IV. Artificial Intelligence Meets Natural
Rule of the Robots: How Artificial Intelligence Will Transform Everything
by Martin Ford · 13 Sep 2021 · 288pp · 86,995 words
the past decade, the field of artificial intelligence has taken a revolutionary leap forward and is beginning to deliver an ever-increasing number of practical applications that are already transforming the world around us. The primary accelerant of this progress has been “deep learning”—a machine learning technique based on the use of multilayered
…
team at the French bank Société Générale to create a proprietary stock market index that would offer investors a way to benefit directly from the artificial intelligence and robotics revolution. In my role as the consulting thematic expert, I helped formulate a strategy informed by the view that AI is becoming a
…
JEDI, an acronym for the Joint Enterprise Defense Infrastructure project, is a ten-year, $10 billion contract to host massive quantities of data and to provide software and artificial intelligence capability to the U.S. Department of Defense. The first kerfuffle occurred at Google, when its employees—who tend to have views positioned pretty
…
with Google’s DeepMind is a leader in pushing the frontiers of deep learning—offers a case study in the natural synergy between cloud computing and artificial intelligence. OpenAI will be able to leverage massive computational resources hosted by Microsoft’s Azure service—something that is essential given its focus on building ever
…
AI, ranging from university research labs to AI startups to practical machine learning applications being developed in large corporations, increasingly relies on this nearly universal resource. Cloud computing is arguably the single most important enabler of the evolution of artificial intelligence into a utility that is poised to someday become as ubiquitous as
…
entry barriers substantially, giving far more people the opportunity to utilize deep learning to solve practical problems. AutoML essentially amounts to deploying artificial intelligence to create more artificial intelligence and is part of a trend that Fei-Fei Li calls “the democratization of AI.” As always, competition between the cloud providers is a powerful
…
Play Store now offer a seemingly infinite number of apps to address nearly any conceivable need. The same sort of explosion is likely coming to artificial intelligence, and more specifically to deep learning. The emergence of AI as the new electricity will, for the foreseeable future, be driven by an ever-expanding
…
are likely to have a substantial first-mover advantage. That could well lead to winner-take-all scenarios as businesses with especially effective big data and artificial intelligence strategies gain a significant competitive advantage. Because data is so central to the effective application of AI, the first step toward an AI strategy is
…
while offering customers a better experience. Indeed, machine learning algorithms are being used to manage everything from inventory levels to product selection to placement of particular items within stores. All this allows physical retailers to begin taking advantage of the same kind of artificial intelligence that Amazon leverages so effectively in its online
…
the limitations of deep learning as one of the primary challenges holding back progress in the industry. “Supervised machine learning doesn’t live up to the hype,” he wrote, “it isn’t actual artificial intelligence” but rather “a sophisticated pattern-matching tool.”59 In other words, a system with the flexibility to offer
…
chemistry or materials science to deploy the power of AI without the need to first become machine learning experts. Artificial intelligence, in other words, is evolving into an accessible utility that can be wielded in ever more creative and targeted ways. An even more ambitious approach involves integrating AI-based software geared toward the discovery
…
over time. In other words, artificial intelligence has produced a solution based on the kind of “outside the box” exploration that is critical to meaningful innovation. Another important milestone, also announced in early 2020, came from the U.K.-based startup company Exscientia, which used machine learning to discover a new drug for
…
an especially vivid indication of just how rapidly specific applications of artificial intelligence are likely to continue advancing. Aside from using machine learning to discover new drugs and other chemical compounds, the most promising general application of artificial intelligence to scientific research may be in the assimilation and understanding of the continuously exploding volume of published research. In
…
contributions to the advancement of deep neural networks. This technology—also known as deep learning—has, over the past decade, revolutionized the field of artificial intelligence and produced advances that just a short time ago would have been considered science fiction. Tesla drivers routinely let their cars navigate highways autonomously. Google Translate
…
it together for a summer.”3 Attendees included Marvin Minsky, who along with McCarthy became one of the world’s most celebrated AI researchers and founded the Computer Science and Artificial Intelligence Lab at MIT, and Claude Shannon, a legendary electrical engineer who formulated the principles of information theory that underlie electronic communication
…
the forced introduction of estrogen. Depressed after selecting the second option, he took his own life in 1954. For the emergent fields of computer science and artificial intelligence, the loss would be incalculable. Turing was only forty-one when he died. In a more just world, he would have almost certainly lived to
…
played the game. Watson heralded a new age and portended machines that would finally begin to parse language and truly engage with humans, but 2011 would also mark the beginning of a dramatic shift in the underlying technology of artificial intelligence. Watson relied on machine learning algorithms that used statistical techniques to make sense of
…
information, but over the next few years, another kind of machine learning—based directly on
…
rise to dominate the field of artificial intelligence
…
time ago, have been an unimaginable scale, and it had become evident that the total volume of data generated globally would continue to grow at an exponential pace. This data gusher would soon intersect with the latest machine learning algorithms to enable a revolution in artificial intelligence. One of the most consequential new data
…
the decade, neural networks had so completely dominated the field that the media would often treat the terms “deep learning” and “artificial intelligence” as synonymous. CHAPTER 5 DEEP LEARNING AND THE FUTURE OF ARTIFICIAL INTELLIGENCE THE EMBRACE OF DEEP LEARNING BY THE WORLD’S LARGEST technology companies, together with the arrival of ever more compelling consumer
…
AI. It is, of course, not coincidental that most important AI research initiatives tend to be associated with large internet companies. The synergy between artificial intelligence and ownership of huge troves of data is often remarked on, but a critical factor underlying this symbiosis is the possession of a massive machine for
…
for the complex calculations required by deep neural networks. I think we can expect the same sort of idea explosion to happen in deep learning, and artificial intelligence more broadly, as the crutch of simply scaling to larger neural networks becomes a less viable path to progress. THE QUEST FOR MORE GENERAL MACHINE
…
—media fabrications that can be very difficult, or perhaps impossible, to distinguish from the real thing. Deepfakes are a critical risk factor associated with artificial intelligence, and we will discuss their implications in Chapter 8. GPT-2 was set up so that, given a text prompt of perhaps a sentence or two
…
common sense will emerge. That’s the main hurdle.”54 Understanding Causation Students studying statistics are often reminded that “correlation does not equal causation.” For artificial intelligence, and especially deep learning systems, understanding ends at correlation. Judea Pearl, a renowned computer scientist at UCLA, has over the past thirty years revolutionized the
…
to absorb all the workers displaced by automation in existing industries. Rather, future industries will be built on a foundation of digital technology, data science and artificial intelligence—and as a result, they will simply not generate large numbers of jobs. A second point involves the nature of the activities undertaken by workers. It
…
below four percent. Another important implication of tepid consumer demand is that it undermines productivity growth. Economists who are skeptical of the impact of artificial intelligence and robotics on the job market are quick to point out that if machines were indeed substituting for labor at a rapid clip, we should see
…
how software automation, often incorporating machine learning, is beginning to encroach on activities across a wide variety of white collar occupations. In the field of law, for example, smart algorithms now review documents to determine if they need to be included in the legal discovery process, and artificial intelligence systems are becoming increasingly adept at
…
over the United States. The country is investing massively and has made artificial intelligence a strategic national imperative. Its leaders appear to be both engaged and knowledgeable. In early 2018, Chinese president Xi Jinping gave a televised address from his office, and books on AI and machine learning were spotted in the background.2 The government is also
…
helping to fund hundreds of startup companies, many of which are valued at billions of dollars and are clear technology leaders. As China assumes its role as one of the world’s two primary centers of artificial intelligence research and development,
…
access to a deluge of data that can be used to train machine learning algorithms.17 The stakes of any perceived AI race between the United States and China are raised greatly by the obvious reality that the impact of artificial intelligence will by no means be limited to the commercial sector. AI will
…
to analyze and recognize human faces and other attributes, Chinese companies are at the absolute forefront of the field. As with other areas where artificial intelligence is being deployed in China, a critical driver of all this progress is access to a massive deluge of data that can be used to train machine learning algorithms.
…
adversarial example attack algorithms.”10 Adversarial attacks are specific to machine learning systems, but they will become one more important item on the list of computer vulnerabilities that can be exploited by cybercriminals, hackers or foreign intelligence agencies. As artificial intelligence is increasingly deployed, and as the Internet of Things results in ever more
…
or other non-state actors who might deploy them against civilians. BIAS, FAIRNESS AND TRANSPARENCY IN MACHINE LEARNING ALGORITHMS As artificial intelligence and machine learning are deployed more and more widely, it’s critical that the results and recommendations produced by these algorithms are perceived as fair and that the reasoning behind them can be adequately explained. If you’re using
…
suggests the possibility of an existential threat. This dark side of superintelligence is known in the AI community as the “control problem” or the “value alignment problem.” The control problem is not driven by fear of overtly malevolent machines of the kind portrayed in movies like The Terminator. Every AI system is
…
who studied innovation in the United States wrote, “Everywhere we look we find that ideas, and the exponential growth they imply, are getting harder to find.”6 This has to change, and artificial intelligence is the catalyst that can make it happen. In the face of these challenges, nothing could be more consequential
…
www.forbes.com/sites/jackkelly/2019/12/10/artificial-intelligence-is-superseding-well-paying-wall-street-jobs/. 28. “Top healthcare chatbots startups,” Tracxn, October 20, 2020, tracxn.com/d/trending-themes/Startups-in-Healthcare-Chatbots. 29. Celeste Barnaby, Satish Chandra and Frank Luan, “Aroma: Using machine learning for code recommendation,” Facebook AI Blog, April
…
’s recommendation rabbit hole,” Huffington Post, October 15, 2019, www.huffpost.com/entry/youtube-recommendation-rabbit-hole-mozilla_n_5da5c470e4b08f3654912991. 33. Stuart Russell, Human Compatible: Artificial Intelligence and the Problem of Control, Viking, 2019, pp. 173–177. 34. Stuart Russell, “How to stop superhuman A.I. before it stops us,” New York
The Means of Prediction: How AI Really Works (And Who Benefits)
by Maximilian Kasy · 15 Jan 2025 · 209pp · 63,332 words
benefits) / Maximilian Kasy. Description: Chicago : The University of Chicago Press, 2025. | Includes bibliographical references and index. Identifiers: LCCN 2025013812 | ISBN 9780226839530 (cloth) | ISBN 9780226839547 (ebook) Subjects: LCSH: Artificial intelligence—Political aspects. | Artificial intelligence—Social aspects. Classification: LCC Q335 .M369 2025 | DDC 303.48/34—dc23/eng/20250409 LC record available at https://lccn.loc
…
Humans Versus Machines 2. What the Old Story Misses 3. What This Book Does Part II. How AI Works 4. What Is Artificial Intelligence? 5. Supervised Learning 6. Overfitting and Underfitting 7. Deep Learning 8. The Exploration/Exploitation Trade-Off 9. Key Ideas to Remember Part III. Machine Power 10. Social Welfare
…
Toward Democratic Control of the Means of Prediction References Index Preface There are a lot of great books about artificial intelligence. Some of them explain in elaborate technical detail how the engineering, statistics, and computer science of AI work. Others focus on one of the many problems that AI might bring about,
…
I am currently a professor of economics at the University of Oxford, where I teach machine learning theory for graduate students and coordinate the machine learning and economics group. I come from a background in mathematics and statistics, as well as economics, and much of my research concerns questions of methodology. I draw on this background when reviewing
…
Walter Palmetshofer, Daniela Platsch, Carina Prunkl, Simon Quinn, Alvaro Ramos-Chaves, Anja Sautmann, Frederik Schwerter, Jann Spiess, Alexander Teytelboym, Martin Weidner, Ashia Wilson, Noam Yuchtman, and Chad Zimmerman. Part I Introduction Are you scared of artificial intelligence? You should be—if we are to believe some popular stories about the threat of AI
…
of AI for humanity, and the possibility of an intelligence explosion, where AI keeps improving itself once it has reached human level. The computer scientist Stuart Russell, together with his collaborators at the Center for Human-Compatible Artificial Intelligence at the University of California, Berkeley, has emphasized the so-called alignment problem—that is, the
…
AI-caused extinction of humanity. The academic version of the story, as told by computer scientists, also tends to feature a man and a machine, where there is a value-alignment problem of the machine (that is, a mis-specified objective) or a bias of the machine relative to its objective. 2 What
…
questions about how it is possible to learn from experience and how to act successfully in the world. To get started, we will need to agree on what we are talking about. What is artificial intelligence? “Intelligence” is a notoriously loaded term, and public perceptions of AI have oscillated from “an obscure academic
…
knowledge, were the dominant approach in AI. But most modern AI is based on machine learning. Machine learning uses data and statistical methods to build automated decision-making systems. To understand AI, we thus need to understand machine learning. One branch of machine learning is supervised learning, where the objective is to predict some outcome as accurately as
…
direction is desirable for society. This is a wider lens than most conversations about AI take, as it entails moving beyond the standard framework of machine learning and the optimization of a single given objective. In the narrow view of AI, based only on the logic of optimization, all problems can be
…
humans behind this platform have started to substitute themselves with machines pretending to be human. 4 What Is Artificial Intelligence? The goal of this chapter is to define artificial intelligence. The question of what constitutes artificial intelligence is closely tied to the equally loaded question of what makes humans intelligent. A key point to recognize
…
AI engineers. The rest of this book will avoid the notion of intelligence, but we cannot avoid having a definition of artificial intelligence. AI is first and foremost a field of research and of engineering. Public understanding of what AI is about has been subject to wild swings. Public understanding has oscillated from
…
for at least the last fifteen hundred years, from Merovingian France to Victorian England. Prediction is an important task in machine learning and AI—arguably, the most important task. Most of the time, however, machine learning does not rely on celestial constellations, hallucinogenic volcanic fumes, patterns of shells, or crystal balls. Instead, it is
…
carefully is one lever by which algorithms can navigate between these twin dangers. There are also alternative methods do so, and they go under many different names in machine learning: Regularization, penalization, shrinkage, and early stopping, among others. All these methods, despite their superficial differences, are used because they implement some version of
…
of the Jabirian corpus. There is something rather unsettling about this state of deep learning—at least for people with a taste for theory and an interest in machine learning, like me: Neural nets work better than they should, according to our theoretical understanding! As shown in figure 3, models that are
…
will tell, but one might conjecture that these architectures will be superseded in a few years, while most of the more general principles of AI and machine learning, which this book discusses, are likely to remain relevant in the long run. Today, virtually all large language models use the transformer architecture discussed
…
where AI might be used in matters related to fertility policy). In both examples, some combination of supervised learning and multi-armed bandits could be used to allocate services and treatments. And since machine learning algorithms are designed to maximize observed rewards, that means that in practice algorithms for labor-market policy might aim to
…
learning problem, are data on internet users. The rather mundane task of maximizing ad clicks based on user data is arguably the largest and most profitable application of machine learning to date. In 2023, Meta reportedly received over $131 billion in ad revenues, while Google’s ad revenues totaled $238 billion; see
…
a typical computer is its central processing unit, or CPU. The CPU is where all the actual calculations take place that let a computer run. Machine learning, and especially deep learning, requires a very large number of calculations during the training of new models. The calculations involved are surprisingly simple, however. For
…
-level control rights. There are strong reasons to think that privacy law cannot ultimately be effective, because of the centrality of data externalities for machine learning and AI. Privacy law is nonetheless an important domain where control rights over the means of prediction are being contested. Next, there is monopoly law
…
there were conflicts of interest between what was good for the public and what was good for Facebook. And Facebook, over and over again, chose to optimize for its own interests, like making more money.” Before discussing this alternative interpretation of value-alignment problems further, I will first review their standard interpretation. I will discuss
…
Mark Zuckerberg’s behavior to infer his preferences and by deriving from these inferred preferences the correct objective for social media feed selection, which is, again, what inverse reinforcement learning would suggest. There are fundamental limits on the extent to which the value-alignment problem can be solved by reward engineering or by
…
the same size, or a piece at all. Whose desire for more cake should AI be aligned with? The most salient value-alignment problems are thus not between a human and a machine. They are between different humans. Consider the algorithms that select what is shown in your Facebook feed. The problem with
…
are the tool available in the short term to align the objectives of AI with society, but they might not be a solution to the alignment problem in the long term. In the long term, we need to establish democratic control. The ideal of autonomy suggests that people should decide their
…
of uncertain harms) becomes redistribution (the transfer of resources toward those exposed to predictable harms) as predictions get more accurate. With data collection and (machine) learning, uncertainties are reduced, and harms become predictable, so that insurance against uncertain harms becomes redistribution to those in need. Critically, markets never deliver such redistribution; they can
…
Collective democratic decision-making is required to enable such redistribution. Insurance is not the only case where data collection and (machine) learning can have large distributional consequences. Another important example of the distributional impact of machine learning is individual pricing. Those who have a higher willingness to pay for some product, for instance because they
…
pay, because their prices go up. Those with lower willingness to pay are better off as a consequence of data collection and machine learning. Both the pervasive presence of data externalities and these distributional questions require collective decision-making to achieve socially beneficial levels of data collection. Individual property rights cannot ever solve
…
day, where AI, and tools based on machine learning more generally, seem destined to have big impacts in the workplace. We can only speculate on the impact that generative AI and related technologies will have. The recent debate has emphasized a few possible directions. For one, the ideal of artificial intelligence as an imitation of
…
the most. Becker’s book on discrimination was written long before the age of AI and machine learning, but its concepts of discrimination are reproduced by approaches that are in use today in machine learning. As we have discussed, AI and machine learning are concerned with optimization; failures of AI are understood as failures to optimize. Becker
…
, we can hope to make better decisions in the future. The leading approach to AI of the last two decades has been machine learning. Machine learning algorithms are trained on data and learn to make decisions based on statistical inference. Because the goal of AI is to solve decision problems, the notion of learning
…
Shades: Intersectional Accuracy Disparities in Commercial Gender Classification.” Proceedings of the 1st Conference on Fairness, Accountability and Transparency, PMLR 81 (2018): 77–91. Crawford, K. The Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press, 2021. Goldman Sachs. “Generative AI could raise global GDP by 7%.” April
…
Policy Choice.” Econometrica 89, no. 1 (2021): 113–32. Lattimore, T., and C. Szepesvári. Bandit Algorithms. Cambridge University Press, 2020. Manski, C. F. Partial Identification of Probability Distributions. Springer, 2003. Mitchell, M. Artificial Intelligence: A Guide for Thinking Humans. Farrar, Straus and Giroux, 2019. Robert, C. P. The Bayesian Choice: From Decision-Theoretic
…
Foundations to Computational Implementation. Springer, 2007. Russell, S. J., and P. Norvig. Artificial Intelligence: A Modern Approach. Pearson Education, 2016. Shalev-Shwartz, S., and S. Ben-David. Understanding Machine Learning: From Theory to Algorithms. Cambridge University Press, 2014. Stein, C. M. “Estimation of the Mean of a Multivariate
…
facebook-employee-frances-haugen-identifies-herself-as-whistleblower. Pessach, D., and E. Shmueli. “Algorithmic Fairness.” Preprint, arXiv, January 21, 2020. https://doi.org/10.48550/arXiv.2001.09784. Russell, S. Human Compatible: Artificial Intelligence and the Problem of Control. Penguin, 2019. Small, M. L., and D. Pager. “Sociological Perspectives on Racial Discrimination.” Journal of
…
–35; fairness in, 76, 111, 119, 164–69, 178, 195; limits on measurability of outcomes, 128, 130; and model complexity, 40–41; and social welfare, 169–74. See also automated decision-making Alibaba, 91 alignment problem, 4. See also value alignment AlphaGo, 61, 63, 64, 87 AlphaZero, 64 Altman, Sam, 3, 107 Amazon,
…
(MTurk), 20, 100–102. See also self-supervised learning American Dialect Society, 106 Anderson, Elizabeth, 200 antidiscrimination laws, 111 artificial brains, 43, 44, 50–52 artificial intelligence (AI): dangers of assuming the inevitability of, 1–2, 14–15, 115–17; defining, 9–10, 22–23; determining the intelligence of, 19, 21;
…
104 Buolamwini, Joy, 6 capabilities, 80 carbon taxation, 95 Carroll, Lewis, 199–200 causality, 178, 180–86 census data, 140–41 Center for Human-Compatible Artificial Intelligence, University of California, Berkeley, 4 central processing unit (CPU), 89 CEO compensation, 125–26 change agents, 14, 97–113; consumers as, 103–6; leverage
…
, 158–60 differential privacy, 15–16, 138–41, 144 diffusion models, 54–55 direct democracy, 197–99 discrimination. See antidiscrimination laws; bias and discrimination Disney, 121 distributional impacts, of machine learning, 146–47 Doctorow, Cory, 105 Dodgson, Charles (Lewis Carroll), 199–200 Dwork, Cynthia, 139 early stopping, 40, 42–43 Ecclesiastes, 124
The Coming Wave: Technology, Power, and the Twenty-First Century's Greatest Dilemma
by Mustafa Suleyman · 4 Sep 2023 · 444pp · 117,770 words
alive right now. The wave is about to hit and this is the forecast.” —Alain de Botton, philosopher and bestselling author “The Coming Wave offers a much-needed dose of specificity, realism, and clarity about the potential unanticipated and yet disastrous consequences of artificial intelligence, synthetic biology, and other advanced technologies. This important book is a vivid
…
human relations. Can we control these new technologies before they control us? A world leader in artificial intelligence and a longtime advocate for governments, big tech, and civil society to act for the common good, Mustafa Suleyman is the ideal guide to this crucial question.” —Jeffrey D. Sachs, University Professor at Columbia
…
computer science at the University of California, Berkeley “The Coming Wave is a realistic, deeply informed, and highly accessible map of the unprecedented governance and national security challenges posed by artificial intelligence and synthetic biology. Suleyman’s remarkable and in some senses frightening book shows what must be done to contain these seemingly uncontainable technologies.” —Jack
…
Be Possible Chapter 14: Ten Steps Toward Containment Life After the Anthropocene Acknowledgments Notes Index About the Authors _144835715_ GLOSSARY OF KEY TERMS AI, AGI, AND ACI: Artificial intelligence (AI) is the science of teaching machines to learn humanlike capabilities. Artificial general intelligence (AGI) is the point at which an AI can perform
…
two universal foundations: a wave of nothing less than intelligence and life. The coming wave is defined by two core technologies: artificial intelligence (AI) and synthetic biology. Together they will usher in a new dawn for humanity, creating wealth and surplus unlike anything ever seen. And yet their rapid proliferation also threatens to empower a diverse
…
gets into the details of the coming wave itself. At its heart lie two general-purpose technologies of immense promise, power, and peril: artificial intelligence and synthetic biology. Both have been long heralded, and yet, if anything, I believe the scope of their impact is still often understated. Around them grow a host of associated
…
technologies, the most profound in history. The coming wave of technology is built primarily on two general-purpose technologies capable of operating at the grandest and most granular levels alike: artificial intelligence and synthetic biology. For the first time core components of our technological ecosystem directly address two foundational properties of our world: intelligence
…
spent my career working on: AI. THE AI SPRING: DEEP LEARNING COMES OF AGE AI is at the center of this coming wave. And yet, since the term “artificial intelligence” first entered the lexicon in 1955, it has often felt like a distant promise. For years progress in computer vision, for example—the
…
-magic engineering one day is just another part of the furniture the next. It’s easy to become blasé and many have. In the words of John McCarthy, who coined the term “artificial intelligence”: “As soon as it works, no one calls it AI anymore.” AI is—as those of us building it
…
, of two waves crashing together, not a wave but a superwave. Indeed, from one vantage artificial intelligence and synthetic biology are almost interchangeable. All intelligence to date has come from life. Call them synthetic intelligence and artificial life and they still mean the same thing. Both fields are about re-creating, engineering these utterly foundational
…
the words of a New York Times investigation, this was a “debut test of a high-tech, computerized sharpshooter kitted out with artificial intelligence and multiple-camera eyes, operated via satellite and capable of firing 600 rounds a minute.” Mounted on a strategically parked but innocuous-looking pickup truck fitted with cameras, it was
…
achieve shared goals. Organizations too are a kind of intelligence. Companies, militaries, bureaucracies, even markets—these are artificial intelligences, aggregating and processing huge amounts of data, organizing themselves around specific goals, building mechanisms to get better and better at achieving those goals. Indeed, machine intelligence resembles a massive bureaucracy far more than it does
…
a mistake. Regardless of where we are with BSL-4 protocols or regulatory proposals or technical publications on the AI alignment problem, those incentives grind away, the technologies keep developing and diffusing. This is not the stuff of speculative novels and Netflix series. This is real, being worked on right this second in offices
…
coastline. It came instead from an unlikely source: the Commerce Department. The shots fired were export controls on advanced semiconductors, the chips that underwrite computing and so artificial intelligence. The new export controls have made it illegal for U.S. companies to sell high-performance computing chips to China
…
including IP, manufacturing equipment, parts, design, software, services—for use in areas like artificial intelligence and supercomputing are now subject to stringent licensing. Leading American chip companies like NVIDIA and AMD can no longer supply Chinese customers with the means and know-how to produce the world’s most advanced chips. U.S. citizens working
…
package. Legislatures are beginning to act. In 2015 there was virtually no legislation around AI. But no fewer than seventy-two bills with the phrase “artificial intelligence” have been passed worldwide since 2019. The OECD AI Policy Observatory counts no fewer than eight hundred AI policies from sixty countries in its database
…
of Rights with five core principles “to help guide the design, development, and deployment of artificial intelligence and other automated systems so that they protect the rights of the American public.” Citizens should, it says, be protected from unsafe and ineffective systems and algorithmic bias. No one should be forced to subject themselves to AI. Everyone
…
Information Processing Systems,” NeurIPS, papers.nips.cc. GO TO NOTE REFERENCE IN TEXT In the last six years “Research & Development,” in Artificial Intelligence Index Report 2021, Stanford University Human-Centered Artificial Intelligence, March 2021, aiindex.stanford.edu/wp-content/uploads/2021/03/2021-AI-Index-Report-_Chapter-1.pdf. GO TO NOTE REFERENCE IN
…
-bigotry. GO TO NOTE REFERENCE IN TEXT “As soon as it works” Quoted in Moshe Y. Vardi, “Artificial Intelligence: Past and Future,” Communications of the ACM, Jan. 2012, cacm.acm.org/magazines/2012/1/144824-artificial-intelligence-past-and-future/fulltext. GO TO NOTE REFERENCE IN TEXT They argue that AI may be slowing Joel Klinger
…
-learning-is-hitting-a-wall-14467. GO TO NOTE REFERENCE IN TEXT eminent professor of complexity Melanie Mitchell See Melanie Mitchell, Artificial Intelligence: A Guide for Thinking Humans (London: Pelican Books, 2020), and Steven Strogatz, “Melanie Mitchell Takes AI Research Back to Its Roots,” Quanta Magazine, April 19, 2021, www.quantamagazine.org/melanie
…
is a strong case Manuel Alfonseca et al., “Superintelligence Cannot Be Contained: Lessons from Computability Theory,” Journal of Artificial Intelligence Research, Jan. 5, 2021, jair.org/index.php/jair/article/view/12202; Jaime Sevilla and John Burden, “Response to Superintelligence Cannot Be Contained: Lessons from Computability Theory,” Centre for the Study of Existential
…
also LeCun Metz, Genius Makers, 58. GO TO NOTE REFERENCE IN TEXT NVIDIA wasn’t complaining Mitchell, Artificial Intelligence, 103. GO TO NOTE REFERENCE IN TEXT Two hundred and fifty passengers “First in the World: The Making of the Liverpool and Manchester Railway,” Science+Industry Museum, Dec. 20, 2018, www.scienceandindustrymuseum.org.uk/objects
…
. GO TO NOTE REFERENCE IN TEXT PwC forecasts AI will add “Sizing the Prize—PwC’s Global Artificial Intelligence Study: Exploiting the AI Revolution,” PwC, 2017, www.pwc.com/gx/en/issues/data-and-analytics/publications/artificial-intelligence-study.html. GO TO NOTE REFERENCE IN TEXT McKinsey forecasts a $4 trillion boost Jacques Bughin et
…
.ft.com/content/f2e88a9c-104a-4515-8de1-65d72a5903d0. GO TO NOTE REFERENCE IN TEXT Early analysis of ChatGPT Shakked Noy and Whitney Zhang, “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence,” MIT Economics, March 10, 2023, economics.mit.edu/sites/default/files/inline-files/Noy_Zhang_1_0.pdf. GO TO
…
-work-will-mean-for-jobs-skills-and-wages. Exact wording: “We estimate that about half of all the activities people are paid to do in the world’s workforce could potentially be automated by adapting currently demonstrated technologies.” Second statistic from Mark Muro et al., “Automation and Artificial Intelligence: How Machines Are Affecting People
…
of intelligence Richard Danzig first proposed this idea to me over dinner and then published an excellent paper: “Machines, Bureaucracies, and Markets as Artificial Intelligences,” Center for Security and Emerging Technology, Jan. 2022, cset.georgetown.edu/wp-content/uploads/Machines-Bureaucracies-and-Markets-as-Artificial-Intelligences.pdf. GO TO NOTE REFERENCE IN TEXT To get a sense of
…
the economy, locking large numbers of people in an equilibrium where they have no work, no wealth, and no meaningful power, the “Turing Trap.” Erik Brynjolfsson, “The Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence,” Stanford Digital Economy Lab, Jan. 11, 2022, arxiv.org/pdf/2201.04200.pdf. GO TO NOTE
…
of-public-space-ddc22d63e015. GO TO NOTE REFERENCE IN TEXT A team of leading researchers Shu-Ching Jean Chen, “SenseTime: The Faces Behind China’s Artificial Intelligence Unicorn,” Forbes, March 7, 2018, www.forbes.com/sites/shuchingjeanchen/2018/03/07/the-faces-behind-chinas-omniscient-video-surveillance-technology. GO TO NOTE REFERENCE
…
-work-plenty-of-warehouse-workers-are. GO TO NOTE REFERENCE IN TEXT Companies like Vigilant Solutions Kate Crawford, Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence (New Haven, Conn.: Yale University Press, 2021). GO TO NOTE REFERENCE IN TEXT Even your take-out pizza Joanna Fantozzi, “Domino’s Using
…
.org/wp-content/uploads/2021/08/FLI-Position-Paper-on-the-EU-AI-Act.pdf?x72900; and David Matthews, “EU Artificial Intelligence Act Not ‘Futureproof,’ Experts Warn MEPs,” Science Business, March 22, 2022, sciencebusiness.net/news/eu-artificial-intelligence-act-not-futureproof-experts-warn-meps. GO TO NOTE REFERENCE IN TEXT Some believe it lets
…
Hype,” Medium, Feb. 1, 2021, sts-news.medium.com/youre-doing-it-wrong-notes-on-criticism-and-technology-hype-18b08b4307e5. GO TO NOTE REFERENCE IN TEXT Promisingly, research on ethical AI Stanford University Human-Centered Artificial Intelligence, Artificial Intelligence Index Report 2021. GO TO NOTE REFERENCE IN TEXT Major shortfalls For example, Shannon Vallor, “Mobilising
…
.12427.pdf. GO TO NOTE REFERENCE IN TEXT In 2015 there was virtually “Legislation Related to Artificial Intelligence,” National Conference of State Legislatures, Aug. 26, 2022, www.ncsl.org/research/telecommunications-and-information-technology/2020-legislation-related-to-artificial-intelligence.aspx. GO TO NOTE REFERENCE IN TEXT The OECD AI Policy Observatory OECD, “National AI
…
, 164, 210 artificial general intelligence (AGI) catastrophe scenarios and, 209, 210 chatbots and, 114 DeepMind founding and, 8 defined, vii, 51 gorilla problem and, 115–16 gradual nature of, 75 superintelligence and, 75, 77, 78, 115 yet to come, 73–74 artificial intelligence (AI) aspirations for, 7–8 autonomy and, 114, 115 as basis of coming wave, 55
…
Inflection AI, 66, 68, 243, 244 information dematerialization and, 55–56 DNA as, 79, 87–88 Institute of Electrical and Electronics Engineers, 241 integrated circuit, 32 intelligence action and, 75–76 corporations and, 186–87 economic value of, 136 gorilla problem, 115–16 prediction and, 62 See also artificial intelligence interconnectedness, 28 Intergovernmental Panel on Climate Change
…
, 194 Meta, 69, 128, 167 Micius, 122 Microsoft, 69, 98, 128, 160–61 military applications AI and, 104, 165 asymmetry and, 106 machine learning and, 103–5 nation-state fragility amplifiers and, 167–69 omni-use technology and, 110–11 robotics and, 165–66 Minsky, Marvin, 58, 130 misinformation. See disinformation/misinformation Mitchell, Melanie, 73 Model T, 24
AI Superpowers: China, Silicon Valley, and the New World Order
by Kai-Fu Lee · 14 Sep 2018 · 307pp · 88,180 words
of the obligations that comes with my work as a venture-capital (VC) investor is that I often give speeches about artificial intelligence (AI) to members of the global business and political elite. One of the joys of my work is that I sometimes get to talk about that very same topic with
…
. First, it spoke to how AI has leapt to the forefront of our minds. Just a few years ago, artificial intelligence was a field that lived primarily in academic research labs and science-fiction films. The average person may have had some sense that AI was about building robots that could think
…
Today, Zhongguancun is the beating heart of China’s AI movement. To people here, AlphaGo’s victories were both a challenge and an inspiration. They turned into China’s “Sputnik Moment” for artificial intelligence. When the Soviet Union launched the first human-made satellite into orbit in October 1957, it had an instant
…
game to AlphaGo, the Chinese central government issued an ambitious plan to build artificial intelligence capabilities. It called for greater funding, policy support, and national coordination for AI development. It set clear benchmarks for progress by 2020 and 2025, and it projected that by 2030 China would become the center of global innovation in
…
artificial intelligence, leading in theory, technology, and application. By 2017, Chinese venture-capital investors had already responded to that call, pouring record sums into artificial intelligence startups and making up 48 percent of all AI venture funding globally, surpassing the United States
…
GAME CHANGER Underlying that surge in Chinese government support is a new paradigm in the relationship between artificial intelligence and the economy. While the science of artificial intelligence made slow but steady progress for decades, only recently did progress rapidly accelerate, allowing these academic achievements to be translated into real-world use-cases.
…
conquest of humankind, but other than boosting IBM’s stock price, the match had no meaningful impact on life in the real world. Artificial intelligence still had few practical applications, and researchers had gone decades without making a truly fundamental breakthrough. Deep Blue had essentially “brute forced” its way to victory—relying largely
…
the cognitive capabilities of machines. Deep-learning-based programs can now do a better job than humans at identifying faces, recognizing speech, and issuing loans. For decades, the artificial intelligence revolution always looked to be five years away. But with the development of deep learning over the past few years, that revolution has
…
finally arrived. It will usher in an era of massive productivity increases but also widespread disruptions in labor markets—and profound sociopsychological effects on people—as artificial intelligence takes over human jobs across all sorts of industries. During the Ke Jie match, it wasn’t the AI-driven killer robots
…
the technology and how it is set to transform our world. A BRIEF HISTORY OF DEEP LEARNING Machine learning—the umbrella term for the field that includes deep learning—is a history-altering technology but one that is lucky to have survived a tumultuous half-century of research. Ever since its inception, artificial intelligence has
…
power of AI to produce superhuman intelligence in narrow spheres. By the time I began my Ph.D., the field of artificial intelligence had forked into two camps: the “rule-based” approach and the “neural networks” approach. Researchers in the rule-based camp (also sometimes called “symbolic systems” or “expert systems”) attempted
…
centuries. But none of these changes ever arrived as quickly as AI. Based on the current trends in technology advancement and adoption, I predict that within fifteen years, artificial intelligence will technically be able to replace around 40 to 50 percent of jobs in the United States. Actual job losses may end
…
of us have been conditioned to derive our sense of self-worth from the act of daily work. The rise of artificial intelligence will challenge these values and threatens to undercut that sense of life-purpose in a vanishingly short window of time. These challenges are momentous but not insurmountable. In recent
…
But when it comes to research in the hard sciences, these issues are not nearly as limiting as many outsiders presume. Artificial intelligence doesn’t touch on sensitive political questions, and China’s AI scientists are essentially as free as their American counterparts to construct cutting-edge algorithms or build profitable AI applications
…
. But don’t take it from me. At a 2017 conference on artificial intelligence and global security, former Google CEO Eric Schmidt warned participants against complacency when it came to Chinese AI capabilities. Predicting that China would match American AI
…
would allow for far more ingenious ways to instantly accomplish its goals. Researchers refer to this as the “control problem” or “value alignment problem,” and it’s something that worries even AGI optimists. Although timelines for these capabilities vary widely, Bostrom’s book presents surveys of AI researchers, giving a
…
of business models, leaving no stone unturned in exploring everything that AI can do. Finally, the third catalyst is one that’s equally obvious and yet often overlooked: China. Artificial intelligence will be the first GPT of the modern era in which China stands shoulder to shoulder with the West in both advancing
…
capacity of nearly one-fifth of humanity to contribute to the task of distributing and utilizing artificial intelligence. Combine this with the country’s gladiatorial entrepreneurs, unique internet ecosystem, and proactive government push, and China’s entrance to the field of AI constitutes a major accelerant to AI that was absent for previous GPTs.
…
build AI algorithms than to build intelligent robots. Core to this logic is a tenet of artificial intelligence known as Moravec’s Paradox. Hans Moravec was a professor of mine at Carnegie Mellon University, and his work on artificial intelligence and robotics led him to a fundamental truth about combining the two: contrary to popular assumptions
…
The free market is supposed to be self-correcting, but these self-correcting mechanisms break down in an economy driven by artificial intelligence. Low-cost labor provides no edge over machines, and data-driven monopolies are forever self-reinforcing. These forces are combining to create a unique historical phenomenon, one that will shake
…
my deepest-held assumptions about what matters in life. But it was that process—and that pain—that opened my eyes to an alternate ending to the story of human beings and artificial intelligence. 7 ★ THE WISDOM OF CANCER The profound questions raised by our AI future—questions about the relationship among work,
…
with those around us. This near-death experience also gave me a new vision for how humans can coexist with artificial intelligence. Yes, this technology will both create enormous economic value and destroy an astounding number of jobs. If we remain trapped in a mindset that equates our economic value with our
…
Or maybe he would greenlight the project but then place someone else in charge of it. I imagined that the fate of artificial intelligence research hung in the balance, and maximizing the chances of success simply meant I had to be in that room for the presentation. I was in the midst of
…
meeting. It was a manifestation of the machine-like mentality that had dominated my life for decades. THE IRONMAN As a young man, computer science and artificial intelligence resonated with me because the crystal logic of the algorithms mirrored my own way of thinking. At the time, I processed everything in my life
…
was easier; it required typing out the romanized spelling of a Chinese word (for example, nihao) and then selecting the corresponding characters from a list. Artificial intelligence has further streamlined the process by predicting and automatically selecting the characters based on context. That technology has made typing Chinese almost as efficient as hammering
…
But as I recovered from my illness and awakened to the looming AI crises of jobs and meaning, I began to see things differently. In that touchscreen device and that unmet desire for human contact, I saw the first sketches of a blueprint for coexistence between people and artificial intelligence. Yes, intelligent machines will increasingly be
…
industries and displacing workers in the process. But there remains one thing that only human beings are able to create and share with one another: love. With all of the advances in machine learning, the truth
…
where I see hope. I firmly believe we must forge a new synergy between artificial intelligence and the human heart, and look for ways to use the forthcoming material abundance generated by artificial intelligence to foster love and compassion in our societies. If we can do these things, I believe there is a path toward a
…
connect the dots on four decades of work, growth, and evolution. That journey has spanned companies and cultures, from AI researcher and business executive to venture capitalist, author, and cancer survivor. It has touched on issues both global and deeply personal: the rise of artificial intelligence, the intertwined fates of the places that I’ve called
…
need for mutual understanding across national borders. AN AI FUTURE WITHOUT AN AI RACE In writing about global development of artificial intelligence, it’s easy to revert to military metaphors and a zero-sum mentality. Many compare the “AI race” of today to the space race of the 1960s or, even worse,
…
, https://www.theverge.com/2018/2/22/17039696/china-us-ai-funding-startup-comparison. first software program: Kai-Fu Lee and Sanjoy Mahajan, “The Development of a World Class Othello Program,” Artificial Intelligence 43, no. 1 (April 1990): 21–36. to create Sphinx: Kai-Fu Lee, “On Large-Vocabulary Speaker-Independent Continuous
…
Large Scale Visual Recognition Challenge 2012, Full Results, http://image-net.org/challenges/LSVRC/2012/results.html. for over $500 million: Catherine Shu, “Google Acquires Artificial Intelligence Startup for Over $500 Million,” TechCrunch, January 26, 2014, https://techcrunch.com/2014/01/26/google-deepmind/. harnessing of electricity: Shana Lynch, “Andrew Ng:
…
University, published March 7, 2018, https://www.youtube.com/watch?v=UF8uR6Z6KLc&t=785s. space race of the 1960s: John R. Allen and Amir Husain, “The Next Space Race Is Artificial Intelligence: And the United States Is Losing,” Foreign Policy, November 3, 2017, http://foreignpolicy.com/2017/11/03/the-next-space-race-is
…
-artificial-intelligence-and-america-is-losing-to-china/. Cold War arms race: Zachary Cohen, “US Risks Losing Artificial Intelligence Arms Race to China and Russia,” CNN, November 29, 2017, https://www.cnn.com/2017/11/29/politics/us-military
…
app-within-an-app model, 59 ARM (British firm), 96 Armstrong, Neil, 3 artificial general intelligence (AGI), 140–44 artificial intelligence (AI) introduction to, ix–xi See also China; deep learning; economy and AI; four waves of AI; global AI story; human coexistence with AI; new world order artificial superintelligence. See superintelligence
…
Association for the Advancement of Artificial Intelligence, 88–89 Austria, 159 automation in factories and farms, 20, 165–66, 167–68 Fink’s letter and, 215, 216 intelligent vs. physical, 166, 167 jobs at risk of displacement by, 157–60, 162,
…
traffic accidents in, 101 universal basic income and, 207 universal basic income (UBI), 201, 206–10, 218, 220, 222, 225 University of Modena, 191–92 University of Science and Technology of China, 81–82 “useless class,” 172, 230 utopians vs. dystopians, 140–44 V value alignment problem, 142 venture capital (VC) industry AI
…
42 Zhu Yuanzhang, 48 Zuckerberg, Mark, 22, 28, 33, 208 About the Author DR. KAI-FU LEE is the chairman and CEO of Sinovation Ventures and the president of Sinovation Ventures’ Artificial Intelligence Institute. Sinovation, which manages $1.7 billion in dual-currency investment funds, is a leading venture capital firm focused on developing
…
front page of the Wall Street Journal. He is the author of ten U.S. patents and more than one hundred journal and conference papers. Altogether, Lee has been in artificial intelligence research, development, and investment for more than thirty years. For more information on Kai-Fu Lee, visit www.aisuperpowers.com or follow
A Hacker's Mind: How the Powerful Bend Society's Rules, and How to Bend Them Back
by Bruce Schneier · 7 Feb 2023 · 306pp · 82,909 words
to Destruction PART 6: HACKING COGNITIVE SYSTEMS 43.Cognitive Hacks 44.Attention and Addiction 45.Persuasion 46.Trust and Authority 47.Fear and Risk 48.Defending against Cognitive Hacks 49.A Hierarchy of Hacking PART 7: HACKING AI SYSTEMS 50.Artificial Intelligence and Robotics 51.Hacking AI 52.The Explainability Problem 53.Humanizing AI 54.AI
…
far older than the invention of computers. It’s a story that involves money and power. Kids are natural hackers. They do it instinctively, because they don’t fully understand the rules and their intent. (So are artificial intelligence systems—we’ll get to that at the end of the book.) But so are
…
the wealthy. Unlike children or artificial intelligences, they understand the rules and their context. But, like children, many wealthy individuals don’t accept that the rules apply to them. Or, at least, they believe that their own
…
good. The trick is figuring out how to encourage the good hacks while stopping the bad ones, and knowing the difference between the two. Hacking will become even more disruptive as we increasingly implement artificial intelligence (AI) and autonomous systems. These are computer systems, which means they will inevitably be hacked in the same
…
ways that all computer systems are. They affect social systems—already AI systems make loan, hiring, and parole decisions—which means those hacks will consequently affect our economic and political systems. More significantly, machine-learning processes
…
, simply because there are more possibilities for unanticipated and unintended consequences. This is certainly true for computer systems—I’ve written in the past that complexity is the worst enemy of security—and it’s also true for systems like the tax code, financial regulations, and artificial intelligence. Human systems constrained by more flexible social
…
by digital systems that make predictions and decisions just like humans do, but they do it faster, more consistently, and less accountably than humans. Our machines increasingly make decisions for us, but they don’t think like we do, and the interaction of our minds with these artificial intelligences points the way to an exciting
…
and dangerous future for hacking: in the economy, the law, and beyond. PART 7 HACKING AI SYSTEMS 50 Artificial Intelligence and Robotics Artificial intelligence—AI—is an information technology. It consists of software, it
…
in operation today. This hacking will arise naturally, as AIs become more advanced at learning, understanding, and problem-solving. Def: AI / ā-ī/ (noun) - 1. (abbrev.) Artificial intelligence. 2. A computer that can (generally) sense, think, and act. 3. An umbrella term encompassing a broad array of decision-making technologies that simulate human thinking
…
all the same types of hacking to which other computer systems are vulnerable. But there are also hacks to which AI systems are uniquely vulnerable—machine learning (ML) systems in particular. ML is a subfield of AI, but has come to dominate practical AI systems. In ML systems, blank “models” are fed
…
model upon which the system is based. Others involve configuring the ML system to make bad—or wrong—decisions. This last category, known as “adversarial machine-learning,” is essentially a collection of hacks. Sometimes, they involve studying the ML system in detail in order to develop enough understanding about its functioning
…
open up the skulls of the police officers or ask them for explanations of their behavior. They look at the results and make a determination from that. 53 Humanizing AI Artificial intelligence systems will affect us at the personal level as well as the social level. Previously, I mentioned social engineering. The most
…
transform into an unparalleled architecture for social control. All of these systems are vulnerable to hacking; in fact, current research indicates that all machine-learning systems can be undetectably compromised. And those hacks will have increasingly large societal effects. 56 When AIs Become Hackers Hacker “Capture the Flag” is basically the outdoor game
…
a single wish, Midas asks that everything he touches turn to gold. Midas ends up starving and miserable when his food, his drink, and his daughter all turn to inedible, unpotable, unlovable gold. That’s a goal alignment problem: Midas programmed the wrong goal into his system of desires. Genies, too, are very precise
…
have been forwarded for new models of regulation that might effectively address the problems posed by the speed, scale, scope, and sophistication of artificial intelligences. AI technologists and industry leaders like Nick Grossman have proposed that the Internet and big data enterprises switch from a “Regulation 1.0” paradigm, where new ventures are deemed permissible
…
al. (28 Jul 2020), “ ‘Ghostwriter’ influence campaign: Unknown actors leverage website compromises and fabricated content to push narratives aligned with Russian security interests,” Mandiant, https://www.fireeye.com/blog/threat-research/2020/07/ghostwriter-influence-campaign.html. 50. ARTIFICIAL INTELLIGENCE AND ROBOTICS 206Marvin Minsky described AI: Marvin Minsky (1968), “Preface,” in Semantic Information Processing
…
, MIT Press. 206Patrick Winston, another AI pioneer: Patrick Winston (1984), Artificial Intelligence, Addison-Wesley. 206probably decades away: Futurist Martin Ford surveyed twenty-three prominent AI
…
, no. 2, http://psychclassics.yorku.ca/Miller. 214explainability is especially important: J. Fjeld et al. (15 Jan 2020), “Principled artificial intelligence: Mapping consensus in ethical and rights-based approaches to principled AI,” Berkman Klein Center for Internet and Society, https://cyber.harvard.edu/publication/2020/principled-ai. 214if an AI system: Select Committee on
…
Artificial Intelligence (16 Apr 2018), “AI in the UK: Ready, willing and able?” House of Lords, https://publications.parliament.uk/pa/ld201719/ldselect/ldai/100/100.pdf. 215Amazon executives lost enthusiasm: Jeffrey Dastin (10 Oct 2018), “Amazon
…
Tool, https://pair-code.github.io/what-if-tool/ai-fairness.html. David Weinberger (6 Nov 2019), “How machine learning pushes us to define fairness,” Harvard Business Review, https://hbr.org/2019/11/how-machine-learning-pushes-us-to-define-fairness. 53. HUMANIZING AI 217program called ELIZA: Joseph Weizenbaum (Jan 1966), “ELIZA: A computer
…
market: Laim Vaughan (2020), Flash Crash: A Trading Savant, a Global Manhunt, and the Most Mysterious Market Crash in History, Doubleday. 226these systems are vulnerable to hacking: Shafi Goldwasser et al. (14 Apr 2022), “Planting undetectable backdoors in machine learning models,” arXiv, https://arxiv.org/abs/2204.06974. 56. WHEN AIs BECOME HACKERS
…
automate the software vulnerability lifecycle,” Center for Security and Emerging Technology, https://cset.georgetown.edu/wp-content/uploads/CSET-Robot-Hacking-Games.pdf. 229research is continuing: Bruce Schneier (18 Dec 2018) “Machine learning will transform how we detect software vulnerabilities,” Security Intelligence, https://securityintelligence.com/machine-learning-will-transform-how-we-detect-software-vulnerabilities/. 230looking
…
Access 6, https://ieeexplore.ieee.org/document/8449268. 59. A FUTURE OF AI HACKERS 242novel and completely unexpected hacks: Hedge funds and investment firms are already using AI to inform investment decisions. Luke Halpin and Doug Dannemiller (2019), “Artificial intelligence: The next frontier for investment management firms,” Deloitte, https://www2.deloitte.com/content/dam/Deloitte
…
/global/Documents/Financial-Services/fsi-artificial-intelligence-investment-mgmt.pdf. Peter Salvage (March 2019), “Artificial intelligence sweeps hedge funds,” BNY Mellon, https://www.bnymellon.com/us
…
/en/insights/all-insights/artificial-intelligence-sweeps-hedge-funds.html. 244the precautionary principle: Maciej Kuziemski (1 May 2018), “A precautionary
…
approach to artificial intelligence,” Project Syndicate, https://www.project-syndicate.org/commentary/precautionary-principle-for-artificial-intelligence-by-maciej-kuziemski-2018-05. 60. GOVERNANCE SYSTEMS FOR HACKING 245AI technologists and industry leaders: Nick Grossman (8 Apr 2015), “Regulation, the internet way,” Data-Smart
…
, Sheldon, 169 administrative burdens, 132–34, 163, 164, 165 administrative state, 154 adversarial machine-learning, 209–10 advertising attention and, 183, 184–85 fear and, 197 persuasion and, 188–89 trust and, 194 AI hacking ability to find vulnerabilities and, 229–30 cognitive hacks and, 181–82, 201–2, 216, 218–19 competitions for, 228–29 computer acceleration
…
(AMT), 61 Amazon, 124–25 ambiguity, 240–41 American Jobs Creation Act (2004), 157 Anonymous, 103 ant farms, 1–2 antitrust laws, 185 architecture, 109 artificial intelligence. See AI hacking; AI systems ATM hacks, 31–34, 46, 47, 63 attention, 183–87 authoritarian governments, 174–75 AutoRun, 58, 68 Bank Holding Company
…
, 146–47, 158 lock-in, 94 loopholes deliberate, 146 legal hacks and, 112–14 systems and, 18 tax code and, 15, 16, 120 See also regulation avoidance loot boxes, 186 Luther, Martin, 72 luxury real estate hacks, 86–88 Lyft, 101, 123, 125 machine learning (ML) systems, 209 Malaysian sharecropping hacks, 116 Manafort, Paul, 26
…
MercExchange, 137 microtargeting, 184, 185, 216 Mihon, Jude (St. Jude), 255n mileage runs, 38–39 military enlistment, 188 Minsky, Hyman, 260n Minsky, Marvin, 206 ML (machine learning) systems, 209–10 money laundering, 86–87 money market funds, 75 monopolies, 93–94 Monopoly, 260n Moynihan, Donald, 132 multifactor authentication, 59–60 Musk, Elon
Co-Intelligence: Living and Working With AI
by Ethan Mollick · 2 Apr 2024 · 189pp · 58,076 words
Enlightenment Now: The Case for Reason, Science, Humanism, and Progress
by Steven Pinker · 13 Feb 2018 · 1,034pp · 241,773 words
Possible Minds: Twenty-Five Ways of Looking at AI
by John Brockman · 19 Feb 2019 · 339pp · 94,769 words
Nexus: A Brief History of Information Networks From the Stone Age to AI
by Yuval Noah Harari · 9 Sep 2024 · 566pp · 169,013 words
The Precipice: Existential Risk and the Future of Humanity
by Toby Ord · 24 Mar 2020 · 513pp · 152,381 words
Machine Translation
by Thierry Poibeau · 14 Sep 2017 · 174pp · 56,405 words
The Singularity Is Nearer: When We Merge with AI
by Ray Kurzweil · 25 Jun 2024
Architects of Intelligence
by Martin Ford · 16 Nov 2018 · 586pp · 186,548 words
Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity
by Daron Acemoglu and Simon Johnson · 15 May 2023 · 619pp · 177,548 words
More Everything Forever: AI Overlords, Space Empires, and Silicon Valley's Crusade to Control the Fate of Humanity
by Adam Becker · 14 Jun 2025 · 381pp · 119,533 words
The Ethical Algorithm: The Science of Socially Aware Algorithm Design
by Michael Kearns and Aaron Roth · 3 Oct 2019
12 Bytes: How We Got Here. Where We Might Go Next
by Jeanette Winterson · 15 Mar 2021 · 256pp · 73,068 words
This Is for Everyone: The Captivating Memoir From the Inventor of the World Wide Web
by Tim Berners-Lee · 8 Sep 2025 · 347pp · 100,038 words
The Thinking Machine: Jensen Huang, Nvidia, and the World's Most Coveted Microchip
by Stephen Witt · 8 Apr 2025 · 260pp · 82,629 words
What We Owe the Future: A Million-Year View
by William MacAskill · 31 Aug 2022 · 451pp · 125,201 words
If Anyone Builds It, Everyone Dies: Why Superhuman AI Would Kill Us All
by Eliezer Yudkowsky and Nate Soares · 15 Sep 2025 · 215pp · 64,699 words
Literary Theory for Robots: How Computers Learned to Write
by Dennis Yi Tenen · 6 Feb 2024 · 169pp · 41,887 words
Investing Amid Low Expected Returns: Making the Most When Markets Offer the Least
by Antti Ilmanen · 24 Feb 2022
Python for Data Analysis
by Wes McKinney · 30 Dec 2011 · 752pp · 131,533 words
Growth: A Reckoning
by Daniel Susskind · 16 Apr 2024 · 358pp · 109,930 words
HBase: The Definitive Guide
by Lars George · 29 Aug 2011
Inviting Disaster
by James R. Chiles · 7 Jul 2008 · 415pp · 123,373 words
Practical C Programming, 3rd Edition
by Steve Oualline · 15 Nov 2011 · 544pp · 96,029 words