<<

AI Magazine Volume 28 Number 4 (2007) (© AAAI) Articles In Honor of ’s Contributions on his 80th Birthday

Danny Hillis, John McCarthy, Tom M. Mitchell, Erik T. Mueller, Doug Riecken, , and Patrick Henry Winston

■ Marvin Lee Minsky, a founder of the field of arti- and the Institute of Electrical and Electronic ficial and professor at MIT, celebrat- Engineers and a member of the U.S. National ed his 80th birthday on August 9, 2007. This Academy of Sciences. He served as president of article seizes an opportune time to honor Mar- AAAI from 1981 to 1982. vin and his contributions and influence in arti- Our goal in this article is to provide readers ficial intelligence, science, and beyond. The arti- with personal insights of Marvin from mem- cle provides readers with some personal insights of Minsky from , John McCarthy, bers of our AI community along with some Tom Mitchell, Erik Mueller, Doug Riecken, brief discussion of his contributions. Aaron Sloman, and Patrick Henry Winston—all members of the AI community that Minsky Danny Hillis helped to found. The article continues with a When I first arrived as a freshman at MIT, I was brief resume of Minsky’s research, which spans determined to meet the legendary Marvin Min- an enormous range of fields. It concludes with a sky. I hung out around the AI lab hoping to run short biographical account of Minsky’s person- into him, but he never seemed to be around. al history. Finally, I heard a rumor that he was down in the basement, working every night on some new kind of computer. I went down one evening to take a look, and sure enough, there Personal Insights he was, surrounded by red-lined diagrams, arvin Minsky has been so influential wire-wrap guns, half-finished computer in so many fields that a brief summa- boards, and a few very busy assistants. I was too Mry cannot do him justice. He is one of shy to introduce myself, so I just stayed out of the founders of and the way and watched for a while, quietly exam- robotics, and he has also made significant con- ining the circuit drawings that were strewn tributions in psychology and the theory of around the room. computing. He has received many awards and When I noticed an error in one of the draw- honors, including the A. M. ings I bravely went up to Marvin and pointed (1970), the (1990), the IJCAI it out. “Well, fix it,” he said. So, I looked Research Excellence Award (1991), and the around and found the right tools and rewired Benjamin Franklin Medal (2001). He is a fellow the emerging computer to fix the problem. of the American Academy of Arts and Sciences Then Marvin asked me to fix something else,

Copyright © 2007, American Association for Artificial Intelligence. All rights reserved. ISSN 0738-4602 WINTER 2007 103 Articles

Figure 1. Marvin Minsky. Photo from S.P.A. Magazine, courtesy MIT Media Laboratory.

explaining to me how it was supposed to work. Minsky. Later, as I came to know Marvin as After a while, I guess he just assumed that I friend and mentor, I began to understand that worked for him. this was a pretty normal interview process for So, that’s how I started working for Marvin him. Lots of people would hang around and

104 AI MAGAZINE Articles say clever things, but he always paid attention I believe Marvin was right that day, and to the ones who actually did something useful. NASA was wrong. But the point of this episode is that it exemplifies what I admire in Marvin’s John McCarthy style. He avoided getting trapped in other peo- Marvin came to Princeton as a graduate stu- ple’s framings of the question, he thought the dent in in 1950, a year after me. problem through in his own way, and he came We quickly realized that we both were interest- to a different solution. And he had fun doing ed in artificial intelligence, as we later came to it. To me, AI has always been the most creative, call it. Unlike me, I think Marvin already had most ambitious, and most fun part of comput- some definite about how to achieve er science. Marvin helped set that tone for AI, (-level) AI. These then resulted in his and in this way he has touched all of us. design for the SNARC neural net learner and later led to his 1954 Ph.D. thesis. I had had Erik T. Mueller ideas different from his but didn’t consider The grand vision of human-level artificial intel- them good enough and didn’t come to a defi- ligence that Marvin presents has been enor- nite approach (through ) until 1957. Nei- mously important to the field and to my own ther approach has yet reached human level— research. At regular intervals I pick up and read nor have any of the others. a copy of The Emotion Machine to keep myself on track and prevent myself from losing sight Tom M. Mitchell of the goal— how the Marvin Minsky has had a significant impact on works. Marvin has taught us many things my own AI research over the years, despite the about artificial intelligence research. Four that fact that we have had relatively few opportuni- stand out are the following: ties for face-to-face discussions. How can a per- (1) We should think big and try to solve hard son with whom I’ve had little personal contact artificial intelligence problems like common- have such a strong influence? It’s easy—I have sense reasoning and story understanding. (2) been inspired by Marvin’s style of out-of-the- We shouldn’t attach too much importance to box thinking and by his vision and aspirations particular words, especially ones like conscious- for the field of AI. His work with his students ness and emotion that refer to complex sets of on machine learning, integrated robot/lan- phenomena. We shouldn’t rush to define guage/perception/planning systems, and frame them. Minsky says “words are designed to keep representations has helped shape the field and you from thinking.” (3) If we can’t find an ele- no doubt my own thinking about it. But for me gant solution to an artificial intelligence prob- personally, Marvin’s strongest influence has lem, we shouldn’t give up. An elegant solution been serving as an existence proof that we mor- is unlikely given what we know about the brain tals are capable of great ideas and great aspira- and the fact that it is a product of many years tions. He has strengthened my own courage to of evolution. (4) We should build large-scale tackle problems that I might otherwise not and systems consisting of many different methods to avoid spending too much time on incre- that work together. mental extensions of well-worn ideas. Marvin I deeply appreciate Marvin’s support over is a creative genius, full of ideas and willing to the years for research on the unsolved prob- pursue them to lengths that others might not. lems of the field. Many members of our com- I recall one episode, when I served with Mar- munity working on building computer pro- vin in the 1980s on a technical advisory panel grams with common sense, emotions, and to NASA regarding a potential space station imagination have received the benefit of his project. Marvin suggested the space station constant advocacy for research on these aspects should be self-assembling so that it wouldn’t of human intelligence. His work is an inspira- require costly and dangerous involvement of tion to us all. astronauts. This was heresy to the NASA estab- lishment, which was then justifying the space Doug Riecken shuttle in part as the key to assembling a space “What is your theory?” This was the first ques- station. But Marvin stuck to his guns and pro- tion from Marvin following our being intro- ceeded to lay out how he would design gener- duced in 1986. I replied that I was working on al fittings that snapped themselves together a machine-learning music composing system automatically, self-propelling parts, and point- called Wolfgang. Marvin asked if Wolfgang ing out that teleoperation of such smart composed interesting music, to which I replied devices would work fine despite multisecond that it appears to, but there is an issue. You see, communication delays with earthbound I programmed how Wolfgang was going to human controllers. learn to learn (based on current machine-learn-

WINTER 2007 105 Articles

ing techniques at that time). I then explained which our interactions became more frequent, that Wolfgang should be put aside for a newer including visits he paid to Birmingham and version system based on a different approach meetings at conferences and workshops. to bias learning and reasoning in a composing Although I keep learning from Marvin, we system; the newer Wolfgang system should do not agree about everything. His emphasis develop biases for things it wants to learn based on “common sense” focuses on the uniquely on system characteristics representing “its human aspects of what do, which I emotions and instincts.” That was the flash think underplays some of our most important point for a friendship and scientific journey evolved abilities shared with some other ani- with Marvin that continues to be the opportu- mals, especially perception and manipulation nity of a lifetime. of complex three-dimensional structures with My graduate studies with Marvin, and the complex causal interactions. Explaining com- years since, have enlightened me to Marvin’s mon sense strikes me as probably the hardest insight, creativity, and perpetual curiosity; his unsolved problem in AI, with implications for unique thinking and comments are always many other aspects of intelligence. We have many steps ahead, and as many colleagues both worked on requirements for “layered” have indicated, it is typical to obtain a eureka architectures for humanlike systems. Marvin’s effect from a Minsky suggestion after several work showed that I was not making enough days of thought. While we recognize Marvin distinctions between levels of processing, part- for his numerous contributions, it is most com- ly because I was more concerned with how pelling to consider how he continues to point fit into a space of possible designs, us as a community towards many of the most which includes knowing how we overlap with demanding questions. He has always com- other species, and also how the architecture mented “that you can not learn something evolved and how it develops from infancy until you learn it many different ways.” Thus it onwards. This difference leads to different ways is essential we develop significant theories of of structuring discussion of architectures, but architecture that will demonstrate this multi- without substantial disagreement. His latest learning-reasoning-representation ability. Mar- book is a major achievement, despite stretch- vin’s thinking and publications (such as his lat- ing the word emotion to refer to too many dif- est book, The Emotion Machine: Commonsense ferent control functions! But the important Thinking, Artificial Intelligence, and the Future of thing is not the label he uses but his myriad the Human Mind [Minsky 2006]) provide impor- insights into what we are and, above all, how tant insights by which to advance our field of we work. We have reached some similar con- research. clusions by different routes that bring out Perhaps one of Marvin’s most compelling insights that I had missed, for instance in his contributions is his legacy as a teacher. Please paper “Interior Grounding, Reflection, and visit Marvin’s website1 and locate his list of stu- Self-” (Minsky 2005) arguing dents (located from the Minsky homepage that apparently simple sensations must be through a link labeled “people”). It is an “complex reflective activities.” Finally, his web- impressive, large list of many of the top AI site is full of treasures, for example the extraor- researchers in our community. I am most grate- dinary BBC Three Music Interview in 2004. ful for Marvin’s friendship and his unique thinking and framing of the hard research Patrick Henry Winston problems that matter. In my first term or two of graduate school, all I really knew was that I wanted to understand Aaron Sloman thinking, so I stumbled around studying all Over 35 years ago, Marvin’s writings, such as sorts of subjects, including neuroanatomy, psy- the amazing “Steps” paper (Minsky 1961b), chology, control, , and which contained many important ideas that operations research. Then, one day I wandered illuminated a major breadth of later work in into one of Marvin’s lectures, and I felt, for the the field of AI, convinced me that the “design first time, that I was listening to a genius think- stance” should be used to address philosophi- ing out loud. By the time the lecture was over, cal problems, that is, trying to understand how the die was cast, and my career objectives were various mental processes could actually work: settled, once and for all. He said things then the basis of my 1978 book. Later, I learned from that I’m still thinking about now. his frequent contributions to comp.ai.philoso- A few months later, I scribbled out a thesis phy (which should be collected somewhere). In proposal vaguely aimed at dealing with analyz- 1995 he agreed to join the Philosophical ing scene descriptions and building on Adolfo Encounter at IJCAI (with McCarthy), after Guzman’s pioneering work on understanding

106 AI MAGAZINE Articles

Photo courtesy Jim Hendler. Minsky During a Break at AAAI-05 in Pittsburgh, Pennsylvania. line drawings. In a footnote, I mumbled some- hardy to disagree with him, as I do sometimes, thing about learning from description differ- but then I remind myself that if we switched ences. When Marvin thumbed through the sides in any argument I would still get crushed proposal, the footnote caught his eye, and with some refulgent insight, expressed con- turning to , he said, “Well, this cisely and with astonishing clarity. Anyway, learning hack looks interesting.” In that few when a computational account of intelligence seconds, my thesis proposal was reduced to the is finally fully worked out, my bet is that Mar- 20 words in the footnote. vin’s ideas and those he champions will be at Many years later, Danny Hillis came into my center stage and everything else will seem like office, and as we talked, we noted that we both epicycles. had many experiences in which Marvin fash- ioned good ideas from our not-so-promising Minsky’s Research and confused thinking. The most extreme, but common scenario goes like this: You think you and Contributions have a great , so you try to tell Marvin Minsky’s research spans an enormous range of about it. He has a very short attention span, so fields, including mathematics, computer sci- he leaps ahead and guesses what the idea is. ence, the theory of computation, neural net- Invariably, his guess is much better than the works, artificial intelligence, robotics, com- idea you were trying to explain. Danny point- monsense reasoning, natural language ed out that something similar might go on processing, and psychology. when we talk to ourselves. The words and For his bachelor’s thesis in mathematics at phrases provide access to ideas that, when , Marvin used the topology expressed in words and phrases, provide access of knots to prove a generalization of Kakutani’s to still more ideas, sort of like Marvin’s K-line fixed-point theorem (Minsky 1950). At Prince- activation cycle. ton University, he focused on neural networks. Eventually, I have come to know a lot of In 1951, he and fellow graduate student Dean geniuses, but Marvin is the only genius I know Edmonds built the stochastic neural analog who is so smart it’s scary. I worry that it is fool- reinforcement computer (SNARC), the first

WINTER 2007 107 Articles

hardware implementation of an artificial neu- ligence and psychology. It also discussed frame ral network (Bernstein 1981). His Ph.D. thesis systems and default assignments and how all in mathematics presented theories and proved these could be applied in such areas theorems about learning in neural networks as control of thought, imagery, language and (Minsky 1954). story understanding, learning and memory, In the following years, he published a num- and vision. The memo is also well known for its ber of results in the theory of computation. He appendix criticizing the logical approach to proved theorems about small universal sets of artificial intelligence and its computational digital logic elements (Minsky 1956), proved inefficiency, insistence on consistency, and that tag systems can be universal (Minsky monotonicity. His criticism of logic’s monoto- 1961a, 1967), and discovered a four-symbol nicity triggered work on nonmonotonic seven-state universal (Minsky and led to the field of formal nonmonotonic 1962). He later published a textbook on the reasoning (Brewka, Dix, and Konolige 1997; theory of computation (Minsky 1967). Ginsberg 1987). With John McCarthy, Nathaniel Rochester, Starting in the early 1970s with Seymour and , Marvin organized the Papert, Marvin began developing his most 1956 Dartmouth Summer Research Project on famous theory of how the mind works, the Artificial Intelligence that is considered to be society of mind (Minsky and Papert 1972, pp. the birthplace of the field of artificial intelli- 92–99). Minsky continued to evolve the theo- gence. Over the next few years, he wrote a ry in several papers (Minsky 1974, 1977, 1980) number of papers on artificial intelligence. and published his acclaimed book The Society They included the seminal “Steps Toward Arti- of Mind in 1986 (Minsky 1986). Singh (2004) ficial Intelligence” (Minsky 1961b), which pre- summarizes the society of mind theory and sented the major problems for future research work inspired by it, including combined sym- in symbolic artificial intelligence—search, pat- bolic-connectionist methods, the method of tern recognition, learning, planning, and derivational analogy, and society-of-mind-like induction—and “Matter, Mind, and Models” programming languages. In 2006, Minsky pub- (Minsky 1965), which investigated introspec- lished a sequel to The Society of Mind titled The tion and free will. Emotion Machine (Minsky 2006). In 1968, Minsky published the influential Both The Society of Mind and The Emotion collection Semantic Information Processing, Machine are giant catalogs of what Minsky calls which presented the state of the art in symbol- “resources” and “ways to think” that allow the ic artificial intelligence at the time (Minsky human mind to be resourceful. Here is a sam- 1968). M. Ross Quillian’s chapter on semantic pling: networks spawned much further research on A censor is a resource that prevents an idea the topic in psychology and artificial intelli- from coming to mind. gence. The book included a chapter by Thomas G. Evans on a geometric analogy program, a A k-line is a resource that records what chapter by Daniel G. Bobrow on a program resources were used to solve a problem so that that solves algebra problems, and chapters on they can be used to solve future similar prob- question-answering systems by Bertram lems. Raphael and . The collection also A paranome is a resource that connects several contained reprints of several papers by Minsky representations so that the mind can easily as well as John McCarthy. switch between those representations. In 1969 with Seymour Papert, Minsky pub- A selector is a resource that engages a set of lished the controversial book (Min- resources likely to help with a problem recog- sky and Papert 1969), which is summarized by nized by a critic. Rumelhart and Zipser (1986) as follows: A suppressor is a resource that prevents a dan- The central theme of this work is that parallel gerous action from being performed. recognizing elements, such as perceptrons, are beset by the same problems of scale as serial A trouble detector is a resource that engages pattern recognizers. Combinatorial explosion higher-level resources when the usual ones for catches you sooner or later, although some- achieving a goal do not succeed. times in different ways in parallel than in seri- al. (p. 158) A value is a resource that is used by resources that deal with self-conscious reflection. In 1974, Minsky published the famous AI Memo No. 306 titled “A Framework for Repre- The emotion of fear is a way of thinking whose senting Knowledge” (Minsky 1974). This semi- evolutionary purpose is to allow us to protect nal memo introduced the notion of frames, ourselves. which was highly influential in artificial intel- Aspects of the society of mind have been

108 AI MAGAZINE Articles implemented by Minsky’s students (Riecken of educational for children) and 1994, Cassimatis 2002, Singh 2005). Minsky’s Thinking Machines Corporation (maker of the books will continue to serve as a treasure trove supercomputer). of ideas to be mined for years to come. He has written about possibilities for a digi- tal physics with particles and fields existing within cellular automata (Minsky 1982). His Minsky’s Personal History interest in science also extends to science fic- Minsky’s wide-ranging scientific and mathe- tion. He served as a technical consultant for the matical curiosity started in his childhood in movie 2001: A Space Odyssey. With Harry Har- City where he amused himself by rison, he wrote a science fiction novel, The Tur- taking apart his father’s ophthalmological ing Option. instruments. He thoroughly enjoyed the Marvin and his wife, Gloria Rudisch, a pedi- opportunity to focus on his interests at the atrician, have three children, Julie Minsky, Bronx High School of Science in the challeng- Henry Minsky, and Margaret Minsky, and four ing company of several classmates who went grandchildren. His living room, immortalized on to become Nobel Prize–winning physicists. in the Society of Mind CD, is packed with inter- After spending a year at , esting things, including two pianos, a harp, Andover, he served in the sculptures, paintings, old Macs, a SNARC neu- in the final year of World War II. His under- ron, a small rocket, mementos from many dis- graduate experience at Harvard University was tinguished personalities/friends (such as Bono academically dazzling; he studied classical from U2, Larry Bird of the Celtics, Gene mechanics with Herbert Goldstein, mathemat- Roddenberry and the cast from Star Trek, and ics with Andrew Gleason, neuroanatomy with so on), and plastic storage bins full of fun com- ponents, gadgets, and toys. Marcus Singer, neurophysiology with John Welsh, and psychology with George Miller. At Notes Harvard, he first envisioned a machine that 1. www.media.mit.edu/~minsky. could learn, an idea he has pursued for his entire career. He went on to graduate work in References mathematics at , where Bernstein, J. 1981. Profiles (Marvin Minsky). The New the department chair Solomon Lefschetz gave Yorker 50–126, December 14. him great freedom to develop his interdiscipli- Brewka, G.; Dix, J.; and Konolige, K. 1997. Nonmo- nary ideas on learning and intelligence. After notonic Reasoning: An Overview. Stanford, CA: Center receiving his Ph.D. in 1954, he continued this for the Study of Language and Information (CSLI). work as a junior fellow at Harvard and at the Cassimatis, N. L. 2002. Polyscheme: A Cognitive MIT Lincoln Laboratory where he wrote one of Architecture for Integrating Multiple Representation the first papers on artificial intelligence, titled and Inference Schemes. Doctoral dissertation, Pro- “Heuristic Aspects of the Artificial Intelligence gram in Media Arts and Sciences, School of Architec- Problem.” He joined the MIT faculty in 1958, ture and Planning, Institute of Tech- where along with John McCarthy he founded nology, Cambridge, MA. the preeminent world laboratory for artificial Ginsberg, M. L., ed. 1987. Readings in Nonmonotonic intelligence research. He is currently a profes- Reasoning. Los Altos, CA: Morgan Kaufmann. sor of and computer sci- Minsky, M. 1950. A Generalization of Kakutani’s ence and the Toshiba Professor of Media Arts Fixed-Point Theorem. Bachelor’s Thesis, Department and Sciences. of Mathematics, Harvard University, Cambridge, MA. Marvin has boundless energy and creativity. Minsky, M. 1954. Neural Nets and the Brain Model He studied musical composition with the com- Problem. Doctoral dissertation, Department of Math- poser Irving Fine and has theorized about the ematics, Princeton University, Princeton, NJ. appeal of music (Minsky 1981). In his spare Minsky, M. 1956. Some Universal Elements for Finite time, he often improvises Bachlike fugues on Automata. In Automata Studies, Volume 34, ed. C. E. Shannon and J. McCarthy, 117–128. Princeton, NJ: the piano. He invented a machine that com- Princeton University Press. poses and plays its own music, the Triadex Minsky, M. 1961a. Recursive Unsolvability of Post’s Muse, with Edward Fredkin (U.S. Patent Problem of “Tag” and Other Topics in the Theory of 3,610,801). His other inventions include the Turing Machines. Annals of Mathematics 74(3): 437– confocal scanning microscope (1955, U.S. 455. Patent 3,013,467), the first head-mounted Minsky, M. 1961b. Steps Toward Artificial Intelli- graphical display (1963), the of a bina- gence. Proceedings of the IRE 49(1): 8–30. ry-tree robotic manipulator (1963), and the ser- Minsky, M. 1962. Size and Structure of Universal Tur- pentine hydraulic robot arm (1967). He was a ing Machines Using Tag Systems. In Recursive Func- founder of Logo Computer Systems (publisher tion Theory, Proceedings of Symposia in Pure Mathemat-

WINTER 2007 109 Articles

Agents. Communications of the ACM 37(7): 107–116, 146. Updated Calls for Many AAAI-08 Rumelhart, D. E., and Zipser, D. 1986. Feature Dis- covery by Competitive Learning. In Parallel Distrib- Programs Are Now Available! uted Processing: Explorations in the Microstructure of , Volume 1: Foundations, ed. D. E. Rumelhart, Please Visit the Conference Website for Details: J. L. McClelland, and PDP Research Group, 151–193. Cambridge, MA: The MIT Press. www.aaai.org/Conferences/AAAI/aaai08.php Singh, P. 2004. Examining the Society of Mind. Com- puting and Informatics 22(6): 521–543. July 13–17 2008 Singh, P. 2005. EM-ONE: An Architecture for Reflec- Chicago, Illinois, USA tive Commonsense Thinking. Doctoral dissertation, Department of Electrical Engineering and , Massachusetts Institute of Technology, Cam- bridge, MA.

Danny Hillis, Ph.D., is a scientist and currently cochairman and chief technical officer of Applied Minds, Inc. Hillis also developed the Connection Illustration © Julie Ridge. All rights reserved. ics, Volume 5, ed. J. C. E. Dekker, 229–238. Provi- Machine, along with many other significant inven- dence, RI: American Mathematical Society. tions, and is a cofounder of Thinking Machines Corp. Minsky, M. 1965. Matter, Mind, and Models. In Pro- ceedings of the International Federation for Information John McCarthy, Ph.D., is a scientist, a Turing Award Processing (IFIP) Congress, 45–49. Washington, DC: recipient, a former president of AAAI, and a Spartan Books. cofounder of the field of artificial intelligence. Minsky, M. 1967. Computation: Finite and Infinite McCarthy is a professor emeritus of computer science Machines. Englewood Cliffs, NJ: Prentice-Hall. at Stanford University. Long ago he originated the Minsky, M., ed. 1968. Semantic Information Processing. Lisp programming language and the initial research Cambridge, MA: The MIT Press. on general-purpose timesharing computer systems. Minsky, M. 1974. A Framework for Representing Tom M. Mitchell, Ph.D., is a scientist and a preemi- Knowledge. A. I. Memo 306, Cambridge, MA: Artifi- nent member of the machine learning community. A cial Intelligence Laboratory, Massachusetts Institute former president of AAAI, Mitchell is the Fredkin Pro- of Technology. fessor of AI and Machine Learning, and is chair of the Minsky, M. 1977. Plain Talk about Neurodevelop- Machine Learning Department at Carnegie Mellon mental Epistemology. In Proceedings of the Fifth Inter- University. national Joint Conference on Artificial Intelligence, ed. R. Erik T. Mueller is a scientist at the IBM Thomas J. Reddy, 1083–1092). Los Altos, CA: William Kauf- Watson Research Center. He is the author of two mann. books and creator of the ThoughtTreasure common- Minsky, M. 1980. K-lines: A Theory of Memory. Cog- sense knowledge base. His research focuses on com- nitive Science 4: 117–133. mon sense reasoning and its application to story Minsky, M. 1981. Music, Mind, and Meaning. Com- understanding and intelligent user interfaces. puter Music Journal 5(3): 8–44. Doug Riecken, Ph.D., is a scientist and department Minsky, M. 1982. Cellular Vacuum. International Jour- head of the Commonsense Computing Research nal of Theoretical Physics 21(6/7): 537–551. Department at IBM Research. Riecken developed the Minsky, M. 1986. The Society of Mind. New York: Wolfgang and M systems. His research is in common Simon and Schuster. sense reasoning, agent-based systems, and intelligent Minsky, M. 2005. Interior Grounding, Reflection, user interfaces. and Self-Consciousness, Proceedings of an Interna- Aaron Sloman, Ph.D., is a scientist and honorary tional Conference on Brain, Mind, and Society, professor of artificial intelligence and cognitive sci- Tohoku University, Graduate School of Information ence at the University of Birmingham. Sloman devel- Sciences, Sendai, Japan. oped Poplog. His research focus is in the study of Minsky, M. 2006. The Emotion Machine: Commonsense mind, along with agent architectures, motivation, Thinking, Artificial Intelligence, and the Future of the emotion, vision, causation, and consciousness. Human Mind. New York: Simon and Schuster. Patrick H. Winston, Ph.D., is a scientist and Ford Minsky, M., and Papert, S. 1969. Perceptrons: An Intro- Professor of Artificial Intelligence and Computer Sci- duction to Computational Geometry. Cambridge, MA: ence at MIT. A former president of AAAI, Winston is The MIT Press. also chairman and cofounder of Ascent Technology, Minsky, M., and Papert, S. 1972. Artificial Intelli- Inc. Winston’s research is focused on developing an gence Progress Report. A.I. Memo 252, Cambridge, account of human intelligence, with particular MA: Artificial Intelligence Laboratory, Massachusetts emphasis on the role of language, vision, and tightly Institute of Technology. coupled loops connecting the two. Riecken, D. 1994. M: An Architecture of Integrated

110 AI MAGAZINE