The HLF Portraits: Edward A. Feigenbaum

Video thumbnail (Frame 0) Video thumbnail (Frame 14554) Video thumbnail (Frame 29108) Video thumbnail (Frame 43662) Video thumbnail (Frame 58216) Video thumbnail (Frame 72769) Video thumbnail (Frame 87304) Video thumbnail (Frame 93667)
Video in TIB AV-Portal: The HLF Portraits: Edward A. Feigenbaum

Formal Metadata

The HLF Portraits: Edward A. Feigenbaum
Title of Series
No Open Access License:
German copyright law applies. This film may be used for your own use but it may not be distributed via the internet or passed on to external parties.
Release Date

Content Metadata

Subject Area
The Heidelberg Laureate Forum Foundation presents the HLF Portraits: Edward A. Feigenbaum, ACM A.M. Turing Award, 1994 Recipients of the ACM A.M. Turing Award and the Abel Prize in discussion with Marc Pachter, Director Emeritus National Portrait Gallery, Smithsonian Institute, about their lives, their research, their careers and the circumstances that led to the awards. Video interviews produced for the Heidelberg Laureate Forum Foundation by the Berlin photographer Peter Badge. The opinions expressed in this video do not necessarily reflect the views of the Heidelberg Laureate Forum Foundation or any other person or associated institution involved in the making and distribution of the video. Background: The Heidelberg Laureate Forum Foundation (HLFF) annually organizes the Heidelberg Laureate Forum (HLF), which is a networking event for mathematicians and computer scientists from all over the world. The HLFF was established and is funded by the German foundation the Klaus Tschira Stiftung (KTS), which promotes natural sciences, mathematics and computer science. The HLF is strongly supported by the award-granting institutions, the Association for Computing Machinery (ACM: ACM A.M. Turing Award, ACM Prize in Computing), the International Mathematical Union (IMU: Fields Medal, Nevanlinna Prize), and the Norwegian Academy of Science and Letters (DNVA: Abel Prize). The Scientific Partners of the HLFF are the Heidelberg Institute for Theoretical Studies (HITS) and Heidelberg University.

Related Material

Axiom of choice Gateway (telecommunications) Complex (psychology) Group action Multiplication sign Decision theory View (database) Coroutine Mathematical model Computer programming Medical imaging Mathematics Semiconductor memory Core dump Endliche Modelltheorie Physical system Electric generator Rational number Software developer Gradient Physicalism Bit Lattice (order) Process (computing) Right angle Reading (process) Geometry Row (database) Point (geometry) Server (computing) Virtual machine Mass Theory Goodness of fit Term (mathematics) Data structure Metropolitan area network Artificial neural network Library catalog System call Euklidische Geometrie Word Software Integrated development environment Personal digital assistant Physicist Video game Game theory Family
Discrete group Group action Building National Physical Laboratory Multiplication sign System administrator Set (mathematics) Water vapor Shape (magazine) Disk read-and-write head Mereology Formal language Neuroinformatik Web 2.0 Optical disc drive Mathematics Spherical cap Semiconductor memory Bus (computing) Mathematician Endliche Modelltheorie Office suite Hill differential equation Position operator Eccentricity (mathematics) Physical system Decision tree learning Predictability Programming language Collaborationism Software developer Computer simulation Computer scientist Flow separation Demoscene Cognition Data mining Arithmetic mean Process (computing) Telecommunication Order (biology) Computer science Website Self-organization Right angle Ranking Freeware Resultant Reading (process) Spacetime Turing test Point (geometry) Existence Service (economics) Vapor barrier Virtual machine Student's t-test Graph coloring Field (computer science) Theory Number Frequency McCarthy, John Term (mathematics) Gastropod shell Computer-assisted translation Metropolitan area network Focus (optics) Information Validity (statistics) Artificial neural network Projective plane Plastikkarte Volume (thermodynamics) Basis <Mathematik> System call Word Elementary arithmetic Software Personal digital assistant Universe (mathematics) Library (computing)
Computer chess Coccinellidae Group action Thread (computing) Range (statistics) Disk read-and-write head Computer programming Neuroinformatik Expected value Inference Different (Kate Ryan album) Hill differential equation Propositional calculus Identity management Physical system Collaborationism Programming paradigm Software developer Inference engine Electronic mailing list Sampling (statistics) Bit Sequence Cognition Electronic signature Message passing Arithmetic mean Process (computing) Telecommunication Heuristic Pattern language Text editor Quicksort Spacetime Point (geometry) Web page Wage labour Computer file Mass Student's t-test Rule of inference Mathematical induction Number Frequency Goodness of fit Boundary value problem Representation (politics) Data structure Computing platform Stapeldatei Information Artificial neural network Expert system Plastikkarte Knowledge engineering Basis <Mathematik> Multilateration Line (geometry) Limit (category theory) System call Word Integrated development environment Video game Principal ideal Multiplication sign Process modeling Mereology Endliche Modelltheorie Area Algorithm Electric generator Knowledge base File format Moment (mathematics) Computer science Self-organization Right angle Procedural programming Resultant Turing test Functional (mathematics) Observational study Virtual machine Field (computer science) Hypothesis Power (physics) Centralizer and normalizer McCarthy, John Task (computing) Shift operator Forcing (mathematics) Mathematical analysis Planning Database normalization Logic Formal grammar Object (grammar) Communications protocol Spectrum (functional analysis) Inductive reasoning
Computer chess Axiom of choice Multiplication sign View (database) Execution unit Set (mathematics) Water vapor Mereology Food energy Computer programming Inference Bit rate Different (Kate Ryan album) Series (mathematics) Stability theory Algorithm Knowledge base Physicalism Sound effect Flow separation Process (computing) Numeral (linguistics) Exterior algebra Right angle Spacetime Point (geometry) Statistics Functional (mathematics) Civil engineering Virtual machine Branch (computer science) Student's t-test Rule of inference Field (computer science) Power (physics) Number Natural number Robotics Operator (mathematics) Energy level Form (programming) Domain name Validity (statistics) Information Artificial neural network Projective plane Expert system Interactive television Maize Film editing Personal digital assistant Game theory Table (information)
Statistics Group action Virtual machine Combinational logic Computer programming Neuroinformatik Number Mathematics Machine learning Natural number Boundary value problem Representation (politics) Software testing Data structure Area Shift operator Programming paradigm Surface of revolution Symbol table Category of being Process (computing) Computer science Right angle Quicksort Object (grammar) Resultant
Internet forum Bit Quicksort
the a little bit of a all right so that you can do you have a because if I begin with finding the man and the boy in the child what kind of a child were you and what kind of a child a conscientious spurious Jewish for who have was not very good at sports right and couldn't care less and when his friends were playing stickball on the urban street I was at home a proving Euclidian geometry syrups because I thought that was wonderful and I would stay up all leaving doing that so that was really is so essential no no I mean I was was perfectly normal have a kid except my friends would do a lot more very things that I would but I always generally it's clearly marked by curiosity that seems to be almost the core of your whole life this curiosity I think I you encouraged and that as a child of the the earliest memories I have of getting into science was the elegance and the beauty and the satisfaction of Euclidian geometry proving those servers why how you could do that and they were really true literature you at this while or maybe 8 9 and whatever whatever geometry comes in the world but maybe maybe 1st year in high school so don't know I I went to college at 16 so I was in high school from 12 to 16 I like to know whether your family is interested in the same kinds of things are you know they not at all not at all and it was not a single scientist in the family they were all commercial businesspeople of small businessmen and typical Jewish small business people lowered gotten transported up by 1 generation from the Lower East Side of New York into New Jersey some nevertheless they also went to Brooklyn that someone who uses it right and there they were New Jersey and they you know to them and the right career to follow it's easier to be a doctor or a lawyer a rabbi and they and never ever no 1 had heard the word physicists so I didn't become a physicist they had heard the word engineer and they might engineers to get jobs yeah so that was OK right so you have a mental or at any point I the metaphor later and will will get it that party but as a child who or in high school not nearly itself that the competent high school teachers but the nobody stands out as a as a mentor or they were they were careful a serious people but down and 1 teacher of biology was pretty poor and that destined me not to go into violence I know you a scholarship to Carnegie so you must have demonstrated a pretty good abilities in high school and high school 3 days straight is of you got a scholarship of war you thinking about your future at age 18 17 18 and my future was going to be an engineering job OK so I mean that they were my family background had no other paintings in it that would or images that would give me any idea that the when it wasn't like I sometimes think if I were living in manhattan I could have gone to the Bronx High School of Science and that would have been a completely different atmosphere but this was just a routine college-prep high school and then things worked out pretty well we have so you got the comedy and scholarship and I'm beginning to sense that this is when your future begins to be shaped as as as we come to know at what happens in Carnegie well you're intellectual development so and currently that's an interesting question I went to the 1st year of engineering school in electrical engineering incidently because that was the closest engineering discipline to physics as far as I could tell us of and what a mass in it so I sail through the year the freshman year and with almost always and up and then I get to the end of my freshman year and I ask myself I remember asking myself is that all there is this is it so I went to see my advisor Edward Shultz who later became provost the work currently tech wonderful wonderful person and he looked at my grade record in set with your grades you can do anything you want so I pulled out the catalog of concurrently catalog and I looked all the way through it and I all I found an interesting course called ideas and social change and you know what it was in a thing called the Graduate School of Industrial Administration and I'm a freshman right I just going into myself for years so but I took the court since said I could do anything I went over there and I sat in on the course and the the person teaching the course now 1 of the most famous social scientists of the 20th century James March I was a brand-new assistant professor it just shown up and the 1st thing here this is a graduate course and I'm only a software or any hand in hand so that that was the 1st reading of the course was a von Norman and Morgenstern's Theory of games and it just blew my mind at night never occurred to me that such a thing could exist and then we went through a whole lot of essentially mathematical models or other social sciences complex social science models with Jim March so March was and what you might call the pied piper I the wonderful early meant to our in a not shaping he was like gateway into the behavioral sciences and at some point and and then March gave me a job in the summer and so an assistant Miss for experimental assistant he was doing some experiment on next small group decision making and I was his research assistant I was but it must spend of software here and in March and Herbert Simon were in the next room of working out the details of the famous book that they work together called organizations and at some point March decided I really needed to meet son and Simon you deserve that yes so he introduced me to Herb Simon and then in my senior year I actually took heard Simon's course call mathematical models in the social sciences it was a graduate seminar at about 6 people in it the Simon had published a book called models of men which was a a compendium of the papacy agreed not his mathematical models of various psychological and social phenomena including 2 of which were the seminal in my view the seminal papers of artificial intelligence will but they were recognized as such at that time 1 was called a behavioral theory of rational choice and the other was called rational choice and the structure of the environment and kind of summarizes AI and and you as a young person you're burning with this this new possibilities and well well so Simon what we we were in a semester system so we met after Christmas for the finish of for a 1st term and Simon comes in after christmas OSes January 1956 and he says over Christmas Alan Newell and I invented a thinking machine Allen Newell is a fairly advanced the young scientist although he was still in the Ph.D. program actually Allen Newell are both in the Ph.D. program in we both studied together for the qualifying exams but the Newell West a pretty savvy scientists paratype said that Allen Newell and I invented a thinking machine and of course that and boggle the mind what could he possibly mean by that so he he explained what he meant by a machine which was a
computer Carnegie had no computers at the time it hadn't it is not for so they were scheduled to get 1 in the summer of 56 that but he Simon gave us all copies of they IBM Model 7 0 1 manual that was IBM's 1st introduction that was IBM's entered to the UNIVAC by IBM's 1st big-scale computer and I remember taking that home and reading it straight through the night and bygone at dawn's early light I was a born again I knew what I was going to do I just and I was gonna stay at Caribbean was gonna work with Simon and I was going to do this thing with this competitor but that that really it that wasn't really the 1st introduction to computing it wasn't brand new in that day my father was an accountant a hard-working accountant who would carry his calculating machine home with him from his job site to our home these were the heavy more and freedom yes counter there is a large was non-mechanized you had to turn the crank and but the Freedman have an electric motor and that these were pretty heavy and I became aware this calculating machines and whereas those friends of mine I was telling you about that that played stickball industry and they were in on it they were good at that when I was good at doing was this calculating machines so every once in a while I would carry this thing on the school bus going down to high school and the 30 pound thing and I would spend the day showing off my skill through it all comes together you're not PhD student under this brilliant man biting the become a PhD student to leave until the September 56 56 Simon Azure gets on supervisor how do you decide on the dissertation color but it's very that's totally simple but Simon didn't believe in courses you just learn what you need to learn when you need to learn and I walked into Simon's office and said here I am he got me a fellowship know there I am OK what now any starts explaining a problem to me which I had to do with what we would now call a computer simulation model but in fact the whole sub-discipline at the time was called computer simulation of cognitive processes that was a sub-field of AI that was focused on models in psychology as opposed to artificial intelligence which was focused on engineering models whether it was like people are not the right and focus beyond the on the question of models of human memory how did people memorize items that word but semantically disconnected from the rest of the world these were called nonsense syllables rightist three-letter cap trigrams it didn't make any sense at all but how do but they there was a 50 years of experimentation in psychology and very stable results that you could get in the laboratory with any people and the idea was construct a model which would then predict all of those were post predict all those results of 50 years in psychology that model was very many to that that is the field that you think you're involved in formerly psychology at this point at this point yes in fact I would I became a card-carrying member of the American Psychological Association of him and Simon himself and Newell went on to win what is essentially the Turing Award of psychology called him distinguished Scientific Contribution Award psychology they also were both Turing Award winners so they straddled both fields and I think in fact this is a good point to save it down not only was Simon a genius but but he may have been the single greatest intradisciplinary genius of the 20th century he was as honored in political science and as he was in psychology as he was in computer science and he won the nobel prize in economics for the invention of what is now called behavioral economics in Siemens terminology behavioral did not behave as the economists viewed mathematical right to say that assume they can write is so your dissertation in the end is in advance and this deal in terms of what you are now saying as a young of doctoral student about human thinking more human memory and an era well that particular model the it's called the pan am elementary ProSieben memorizes was the best model in existence of that set of phenomena of that it turns out that information processing language which we would now call computer language it was the best language in which to express that model mathematics was not the best language mathematics was a straitjacket you could you couldn't discuss complex and discrete processes in mathematical language but you could easily and simulated model of those in in computer language with and this is a language you had come up with the world that not just me but it's that the mean the originated were really Simon and Newell and a set of colleagues at the time of Marvin Minsky at MIT being 1 of them of which you do your heart is a young but I was just some some people call me a 1st registered and and will I'm just guessing here to your dissertation was a success so with this success and I and Simon I continued working on it but the 1st we did to model I ID 2 models in my faces EPM 1 in the pen to Simon and I developed an a better model that fixed up well typical sciences refine your models all the time that you can 3 and currently there are 2 psychologists in Britain who were doing an anniversary of what volume to celebrate well something like the 50th or 60th anniversary of service of the development of the pact and its experimental of validations over the years and it's an MIT press both of they OK so you're him ending so you need a job all you need to know the position now in academia have is that it happened well there are 2 places I could go 1 it is into psychology and of that's kind of unlikely since this this kind of thinking is very new to psychology and it psychology departments were would have to be very innovative if they were a higher universe like this or you could just follow you union card my union card said Graduate School of Industrial Administration and there was a group at Berkeley in the business school at Berkeley I call the Management Science Center that just love to what we were doing you pointed out over your shoulder literally that's where it is that at the end of and they hired me and they hired another of her Simon students during Feldman and we both went to Berkeley I had decided that was on the basis of travel around the country that I was a San Francisco Bay area I want to live in in any in in any case in Berkeley offered me this job so right now I have great and I have a house in the Berkeley hills and this gorgeous view of and and then Stamford data set of metallurgists have I will get there that suffer for 5 years well I got there and so I have full right in no 59 60 OK as a kind of a separation year between at the end of working with heard Simon on a PhD at Carnegie and the beginning of a job at berkeley but so I went to the National Physical Laboratory in England which is where Allen Turing had ended up after his Bletchley park really work he had come there to design a real machine and that was called the ACE machine and in fact when I got there in in 59 and the there was still a machine they're called the Pilot ACE although English Electric had already built a follow machine called the dues the following the yes this during had already moved from National Physical up to Manchester I have an aunt had died of course frantically it was during already a sacred name of these no no he wasn't at all he was just a a a warm memories warm up because all those people at the National
Physical lab were his friends they work with them and they knew him so that will in over teacher and they talk about Turing and his eccentricities think that as the founder of and now I know that now the preceding now the idea of Turing being the founder of Computer science came later it was basically an ACM invention that all we have to have founded there needs to be a founding story in every in every year when the needs to be founding story so we shall we who shall we made allows all let's make Turing the founder of and you know he did this wonderful abstract things so satisfy the mathematicians and you also had this wonderful paper on on thinking about by machine know and so that will make the AI people happy and so he was annoyed to the founder and a Turing Award was set up and 1 of the cart people won the 1st Turing Award Allen Perlis I was head of the Computing Science of that hidden Berkeley will get you started with the real news begins but in Berkeley are you there 5 years for 5 years you're there as a business psychologist I mean was what I was I talked the 1st artificial intelligence courses at Berkeley did you and that of course I had to earn my bread and butter when that was member that book I mentioned organizations by Martin Simon yes this I talk organization theory out of that book I and that book was also being used in the School of Public Administration and also Simon's earlier book Administrative Behavior was being used at Berkeley but I taught artificial intelligence and Ayatollah organization theory it's it's hard to retrospectively we describe a period of excitement but did you have a sense at this early stage of growing interest and Sudan's drawn to these questions or were you feeling pretty much alone in your interest the there was a lot of excitement out among the students and that the the cost that I offered and sometimes jointly with Julian Feldman was a very busy courts the literature at that time was scattered but there were a few papers here in a few papers there and Xerox xerography had not been invented yet believe it or not but here we are talking about it would get old machines in it was difficult for the library to reproduce so but Julian Feldman II put together a an anthology of readings book for the students called Computers and Thought published in 1963 sometimes called a 1st book yes of and it is still impressed it's still being sold by MIT press and it it's insides are still fresh insights still fresh people remark that well I we put it on the Web as well so it's available right for free for downloading people downloaded say this is remarkable the so why are you reading this beautiful college-town to travel all the way across the water to Stamford to I want to marry you that contain prediction 2 reasons the 1st reason was that the well in then in the year 19 sixties 3 64 so I was being considered for 10 years at berkeley accordion us over there was an acceleration and due to some peculiar circumstances having to do with salary in what rank and salary were tied in the Berkeley system so in order to get me an appropriate salary to bring me in a slightly higher rank that made me eligible for tenure as a year earlier to our 2 years earlier and so I was being considered for tenure I would I was a computer what we would now call a computer scientist who was building models in psychology who was associated with center from for human learning at berkeley so my ties were more psychology than they were to the business school and yet I was being cats considered for tenure in the business school and it was kind of like a fish out of water what was I doing a business school anyway right the end if I looked across campus what I found was a chaotic scene in which the electrical engineering department in the math department were furiously arguing about who would get this new discipline that was emerging which had didn't yet have a name but finally came to have the name computer sites these were a vigorous wars between mathematics and electrical engineering it ended up the the the 2 departments and then finally of a Chancellor came actually Al-Baraka from Stanford came in his chair and said this is ridiculous we're gonna make 1 Department of EECS and at that point Berkeley settled out but it took them years to settle down so in that area and while they were chaotic and I was searching around Stanford was starting a new department John McCarthy another Turing Award winner famous the co-inventor of I knew me from across the barrier of course and recommended to George Forsyth the founder of the department that there this guy Berkeley was looking for a job when a higher integer over with here yes of course so they offered you 10 year and said the wealthy where the US odds in such a professor like that Stanford no I think my timing may be wrong and it's that you begin also the time and create a laboratory it as you as you begin your work at Stamford is that is that you will well I what happened at Stanford was John McCarthy and I were were both funded by this new age in new of funding agency and the Defense Department which was then called later called Dr. I was being funded for my AI modeling McCarthy was being funded for time-sharing development we both knew that we need a lot more money and we needed a big computer that would that would energize shouldn't be the computer for artificial intelligence work so we got together and put in a joint proposal because you put in a joint proposal for a a computer and the computer became the center of the Stanford Artificial Intelligence Laboratory at all but there was no space on campus we we were already overflowing space are little space that they gave us to begin a department and so the set for Artificial Intelligence arbitrary got moved off miles away into the foothills into a remote building that was gifted to Sanford by the company called G T any which doesn't exist anymore General telephone Electronics that they had decided to put a research lab out there and then abandon it so the the shell of the building to Sanford instead for finish it Officer the Stanford Artificial Intelligence arbitrary and it was therefore for many years but until we moved back onto campus now my problem with that was that down I was already deeply engaged in an interdisciplinary collaboration with Joshua Lederberg and it was a geneticist and and with a number of chemists in the Stanford mass spectrometry that headed by Codd Djerassi the we needed interactive work with the computer and it was a the deal for our chemists to get in their cars drive all the way up there and spend time there getting in there can't find a parking place back at it was just a big deal so we just brought brought of our project back down to campus use the Stanford campus machines for a while but that was not satisfactory was we we needed a more cutting edge kind of machine so were they to Bergen I put in a proposal to the National Institutes of Health since he was a famous biologist in the medical school and we started of a computational laboratory called some makes stanford university medical experimental right computer the i in medicine at what's in a little more of an aim was a national community we convinced the National Institute of Health that with the advent of modern networks ARPANET did not exist at the time but there were other networks like Telenet and time meant that we could bring in other people from universities all over the country utilizing common software and shape and that was I am part of some mixing keratin
I'll just backtrack 1 thing to the Call it inside call it a natural development where you can see that the collaboration with a biologist only that itself is already I suppose following the interdisciplinary model but what led you to think that as you proceed with the think about computers that getting together with the biologist was the right way to go point you asked that question I would never about it here if you I would say to the as a message to young people always think in an interdisciplinary way this disciplinary boundaries are artificial and I was able to learn that Edmund at an early stage from hurts I thought the most interdisciplinary scientist that existed at the time if you look at the introduction to the book Computers and Thought it I happened right at lower cost of the book was due segment I wrote the introduction the last paragraph or 2 talks about inductive processes that we need to work on inductive hypothesis formation most of the other work and I had been deductive and and mostly dealt with puzzle-solving or mathematical analogs so that like proving trigonometric identities but no 1 had worked on the broader issue of how do we asked people or machines as a eyes ingested a lot of data from the world and form a hypothesis a continuing hypothesis of what's going on out there what is what is all that data me so I although I was working on papers with her Simon on the PAM during that period in the back of my mind was constantly how online investigate that problem and in particular had a white create what officially would be called an experimental environment for experimental computer science although I called it a sandbox Hadaway create the right kind of sandbox in which to explore the questions I had learned from Newell and Simon you don't explore these questions in the abstract you explore them and what were called task environments and they had originally chosen white in Russell's print keepyour Mathematica propositional logic as the 1st task environment they later moved on to I mention trigonometric identity as an chest as they were that they had the 1st major chess program I call the NSS chess program so I learned about task in environments and sandboxes I had to have an experimental environment in which to study this question so in 1964 a psychologist the name Karp Pribram who had gotten infected with this germ of information Processing models of cognition and had published a book with George Miller and going to call plans in the structure of behavior coral but in this psychology department ran a Saturday seminar in the Center for the Advanced Study in the Behavioral Sciences which is up on the hill behind Stepford near the golf course so every Saturday morning people with interests like that would gather once a month not every Saturday morning once a month among those people brought there by John McCarthy was this amazing biologist Joshua Lederberg and when I explained cost nobel prize winners an absolute genius like and and when I explain to labor what I was looking for what what could I use as a sandbox to study I already come to the conclusion that if you want of to conclusions for you wanna find out how people think Best that and it's part of the Newell Simon work program with would which had to do with what are call thinking aloud protocols you can tease out of thinking people's minds how they think even though they probably have thought into U.S. that's part correct and you could work out the details probably the people I want to work with most initially at least were scientists because they were professional paid hypothesis force that's what they are they are they it's the experimental scientists look at a lot of data and they're constantly forming hypotheses about that but in particular areas about that data so why not talk to labor about that so we did on 1 Sunday morning and he said I have a great idea for you it's going to be in the area of mass spectrometry and there was a reason for that labor was working on the use of a mass spectrometer as a probe for life on mars on the 1st robotic platform that would land on mars Viking platform that was being designed and labor was the principal of 1 of the principal investigators of the part of that study that was going to design instruments to find life what were they looking for then we're gonna look for amino acids and the instrumentation Imagen doing wet chemistry in a lab on Morris crazy right so they were going to do it using dry chemistry electronic chemistry which was mass spectrometry fragmenting molecules collecting the fragments by their of their fragmental molecular weight and then inferring what the structure of those fragments implied did they imply a particular kind of amino acid like at inferring it's of the operative words there yes inductive inference so labor had become a minor expert on the mass spectral fragmentation of amino acids because that's what he was the our and when that he said look when you get to Stanford on in early January he knew I was by the time he knew I was coming set and did but let's get together so as soon as I got Sanford I called up later birth and he said OK let's start no I have an algorithm which is capable of generating from the overall molecular formula which is easy to get but that's not the best more of a computational problem right and a out I can I have an algorithm which can generate just like a legal move generators chess can generate all the possible answers to this problem in a in a complete and a redundant fashion he happened to do that because he was deciding to learn modern computation he was a geneticist who had been involved in a little bit of complication when it was all cards back in the thirties but he wanted to get into modern computation and he picked this problem that being a genius that picked this problem Our Acadia which is also pretty smart some side maybe and from that we can layer on to his algorithm just like in it just like in playing chess we can layer on the heuristics the the rules of good judgment the rules of good guess signature that would enable the pruning of that space to allow the only thing that would come through all that generation was a short list of the most likely candidates structural candidates 4 that particular fragmentation patterns that was seen by the mass the common that we didn't we didn't give the the user of the chemist a single answer we gave a chemist a few answers believing that there may be some basis for the chemist can
know some things that he hadn't yet told us from which he or she could pick the right answer out of the few whatsoever typical number would be 3 5 and and Codd Geraci who later became mass spectrometry it would look at it and say yeah it's that 1 because I know we've got a sample from René this is a marine sterol and we got from resample and and this was a really this is a marine object and he would not no why did we involve Jurassic because almost immediately we ran out of ladybirds knowledge he had a very limited body of knowledge about mass-spectrometry claim that amino acids but when we gave the program the simplest possible alcohols or ethers or file users easily you know chemistry 101 l organic chemistry 101 simple things the program was terrible it knew nothing it was like 1 of these really dumb programs you talk to on the phone and it knows nothing it so the reason was because it knew nothing so we had to find someone who really knew mass spectrometry well the single most knowledgeable per person perhaps in the world have to was a Codd Djerassi before that formerly the inventor of the birth control pill and later head of the central nested country lab and he had a bunch of people PAGE and post postdocs who knew a great deal about mass spectrometry and systematically we began to mine the knowledge out of their heads and in fact the the then in the act of doing it became known as knowledge engineering that's what we were doing we were my and lighting these gold nuggets out of the heads of practicing practitioners what if that formerly considered the 1st demonstration of that experts say yes so then draw all this program DENDRAL named after later birds the original algorithm which was dendritic algorithm long and and when we turned it into an AI program was called Heuristic DENDRAL but since that was clumsy we just kept calling it then wrong Gendreau was the 1st expert-system in 1968 we were asked to give a paper at a conference on an annual workshop in Edinburgh that was run by a British AI scientists who had worked as a young person had worked with Allen Turing and Bletchley Donald making Donald Asyst he knew about his work and he asked us to give a paper on it and what I did there was to assemble the full range of empirical results remember I was behaving as an empirical scientist and I labor an hour and bruised and our team we were empirical scientists studying results of computational experiments that and we were trying to infer from those experiments what was general about that I knew all and 1 of his students had just published a book on the so it was called on generality and problem solving and we were looking at the Kwun that was the question of how much of a I represented the power of reasoning general reasoning that was a big idea general reasoning GPS was the name of Newell and Simon's program but McCarthy had dedicated his whole career to the formal methods of reasoning is this a such lot of logic and Newell and Simons was not strict logic Newell and Siemens with a heuristic program called means and analysis and that shifts from expectations where so when we looked at our results our results said it was in the power of solving more and more complex problems in organic chemistry was not a function of how powerful the reason there was it was a function of how much the program knew about versity of discourse small knowledge knowledge is power and I sort of stolen have of all related well I stole it from Bacon Bacon didn't use it in connection with with the the used in a political sense but I I turned it into in the knowledge lies the power it's knowledge that counts not the power of the inference engine that kept only having 1 batch of experimental evidence leading to this I called it the knowledge is power hypothesis but later on as we build many many many more expert systems and as thousands were built in the world and they all reflected exactly the same thing I change the name of it to the knowledge principle it's really a foundational principle of AI that In the knowledge lies a power not in the reasoning that I was to interject a human moment because here you come to this well very important of inside are you met with conferences subsequently met with solving our hallelujahs hallelujahs absolute this was a whole year of hallelujahs and in 1984 this member word I'm talking 1968 right then we developed all these systems base sequence of experiments at stanford that involved at least a dozen systems but many other places around the country particularly in this AI I am community in medicine right and commercial Our expert systems the I lost the thread will anyway quasi 84 the editors of the Journal of Artificial Intelligence asked by a number of luminaries in the field to write up what they considered to be the the main results of the last decade of artificial intelligence research so that was basically they were writing this in 83 so 73 through a 82 and Allen Newell who is 1 of the founders of the field and this is 1 of the early geniuses in a Turing Award winner said it was clear to him that expert which was the number 1 all of the advances in the field during that period here's here's what I wonder whether user universal this is so eventually of course 1 of the reasons we're so that your work but this was really radical shift to go from thinking AI in from logic to knowledge and I'm wondering whether or people are resisting this whether they are it instantly getting the importance of the ship eventually of course the the answer is that the shift was gradual OK and it was announced to the world as of course all great things are by MIT people this Seymour Papert in 1 of his colleagues wrote a paper on it in which they announced quote the shift to the knowledge-based paradigm without was due to ups but but it took a period of they publish their paper in 1977 OK so it took about almost 10 years for the transition to be complete the resistance came from the people who simply didn't believe the idea of which John McCarthy was 1 of the where those people he was a believer in formal methods were the power of AI was in the formal methods not in the knowledge so when McCarthy was talking about common sense reasoning the obvious question that we would ask him in and indeed 1 of our students is now quite famous in the field that Lennart I was working on common sense reasoning obvious question John McCarthy is so why are you worried about the logic of common sense reasoning why aren't you worried about the representation of common sense knowledge and Lennart went off to the actually I worked for research lab for a while and then started his own company called psych Corporation is assumed it was the end of the line and psych Corp. built a huge knowledge base of millions of pieces of common sense knowledge I think that if you've all actually refer to it as the initial of of potentially as the continues this of this is because where he's going or maybe he's already here I don't know that the status of the research of but is actually thinking machines in the broadest sense rather than just expert the procedure what is the difference between common sense thinking as applied machines that extra however
the the difference is the narrowness of the domain but when we 1st started working on medical diagnosis we did not take on all of medicine the 1st EC exploration with by short a graduate student and the PhD student on the diagnosis of blood infections spend their treatment by and about that turned out to be 400 rules of expertise typical numbers would be a few thousand rules of expertise in a narrow field I I I did 1 here in San Francisco on pulmonary function diagnosis and that turned out to be a few hundred rules and turned out to diagnose these cases perfectly well the doctor who we were working with I asked to see what what the knowledge was and I showed it to him and he said is that all there is and we said No that's all there is and out he will often work in another field for 5 years and he was so shaken by that simplicity that that's all they know that that that's the so it's the narrowness of the domain and common sense reasoning is not narrow I if I have the if I'm drinking a glass of water and I over and the water runs over the table and runs on the rug and this nothing about that that you're going to infer from physics right it's just that you know that you've seen it before the machines have historically know that and if you're if you're out if you're self-driving car and you're driving by a in Mountain View California where they drive most of the time driving by a big part that has a soccer field you'd better worry about a kid running out in the street chasing a soccer ball at the latter infer that from psychology of physics or anything like that that's what common sense knowledge isn't or another basic piece of common sense knowledge is a vital chapter stable and have I have 2 tables if I chopped the half I have 1 air right or if I we used in fact this rule by in 1 1 medical diagnosis programs if the patient Our is presenting certain symptoms and of the patient is Mikhail then rule out all kinds of things that would be female related for example related to pregnancy right of course so it how else would machines know that so it is common sense the next the achievement of our that absolutely absolutely no when I when Serie was Ooh Serie company was formed out of the DARPA project at SRI and then it was bought by steve jobs for Apple and what 1 of the people who had been working in my lab Thom Gruber became chief technical officer of Syrian when they moved over to Apple and Apple was going to put it on the phone I said the time remember the lesson of knowledge is power you're only going to succeed if you focus on certain things it's Serie will do my travel or hotels or some specific things that you can focus on in fact I even suggested that 1 thing with series should know more than anything else in the world is the iPhone itself instead of me having to do a lot of things pushing buttons it should not I I should be able to ask you to do things on the iPhone energy do a form in which it does not mean of series set the alarm for 8 o'clock in the morning work but when Tschira tried to do generalize questions it did know anything it didn't know it knew it could look up the answers in in civil evil knowledge base but otherwise it didn't have any common sense knowledge at all and so it became the butt of a lot of jokes in the early series gets now Google took the whole issue a lot more seriously can they did build in enormous knowledge base that underlies of Google and when you do a Google search and on the right-hand side you get a lot of information there that's because Google really does know what you're talking about it's a natural point ask you and we began to talk about really the future of AI and the the validity of the machine that is smarter than we are all these questions that the lady population at any rate is interested in where where you see the placement of the machines that can think in a common sense way in the future use of the broader broad nature of our culture the the dystopian view of that is circulating in the world due to outrageous comments by Elon Musk and Stephen Hawking fueled by an AI colleague at Berkeley of Stewart Russell and several others of the these dystopian view this ignore the scope and power of what is to be human in the in the in a about this I mean that in a technical sense not in a humanistic sets this the technical sense is that we people know any enormous amount of things when we become by the time we become an adult we have seen everything we have we have 10 to the 15th Noren's appear and their operating real slow and we just store a tremendous amount of experience and we bring it out of using methods may be like the statistical methods of neural networks I'm not sure about that but we bring out that experience and we have an enormous amount of experience especially about the interactions between as people so and machines are very stilted that of course comics and comedies and even dystopian tragedies get a lot of effect out of the stupidity of the of the use of AI robots doing more idiotic things because they just don't know care they don't get in fact to the EC the with this movie on the EC slash enough of actually was quite a good movie because the person who did that movie did get I really did understand what it meant to have that knowledge that's what so spooky about that woman artificial woman woman that will be if she had this so far that's number 1 thing is that it is going to be a very long time before programs have that amount of knowledge of the world the players if you give it if you give me any specific problem you want if it's a legal problem a medical problem of business problem you name it we can nail it with an expert system we know that technology inside and out but if it's everything in the world who know I mean it's just so many things that they will be room for humans and their machines for a long time to come now there are things for which machines are definitely better than people 1 of them is systematic numeration people are very bad at being systematic but so 1 of the reasons then role was so good is it never miss anything it had that underlying general algorithm in which guaranteed that we had all the legal moves in that space of molecules right but more 1 of the 2 so every once in a while a company works of of a of a problem in the branch unit so the problem gets to be beyond what people can cope with so when cost far off lost that game to people that was because the blue went down a level or 2 deeper than cut costs far-off looked at a few hundred excellent choices in the maze people didn't have as much knowledge of chess but it looked at 200 nm I think it was something like 280 million alternatives it found 1 that had never before been seen by mankind which was elegant and beautiful and when it when the move was exhibited to the chess masters in the audience who were watching this it was just a murmur of all are through the audience that strikes me as the degree of computational all of the ability but not I mean in in the
sense of which numerical testing I get as something that
beyond what human beings can do and can aid us but it's not thinking it's not this world if you've introduce us to where speculation is conceivable for machines speculation as opposed to I that we have yes I wasn't I wasn't careful enough in my in where worded it the boundary is symbolic representation and symbolic manipulation of if its computational that's great that belongs to other people in computer science at do it better than we do right and the the the boundary is very at the boundary called machine learning which I is a marriage between statistics and AI and amazingly the statistics as work really well and so remarkable things are going on at the boundary largely computational was rather than inferential right but there will be but the pendulum swings and whereas in the sixties and seventies it was mostly inferential now it's largely computational they will merge together into a combination of artifacts that know how to categorize and identify objects in the World based on large numbers of features of those objects which is what we call Machine learning there they can go up from thousands to millions of right features because we have what amounts to boundless computation of people don't realize that but we have incredible amounts of computation living out there in in were called warehouse schools warehouse computations and millions of computers and Microsoft and so on we have enormous amounts of computation and Google employs them for their what's called people talk about Google having a great day I grew but they also have great computers supporting the AI group so where there will be a marriage and it will not only be between inferential processes balanced and machine learning statistical processes but between people and those processes that is the end of it all where humans are assisted by these things that programs do well and programs are assisted by the things that people do well but is knowing everything do you have advice less young people contemplating a career in computer science my advice would be 3 things number 1 is if you're orienting yourself toward a career in computer science then take very seriously working in artificial intelligence the world has given you 1 of the greatest problems ever know which is the problem of the nature of intelligence I always call artificial intelligence to manifest destiny of computer science number 2 if you find a problem area in which many people are working for example in in AI these days it's machine learning then go somewhere else work on something else don't publish incremental papers incremental papers are soul-deadening you work all year long to get an experimental result you write 2 or 3 papers on it move on to the next incremental result soul-deadening and the 3rd is you gotta take Thomas seriously in his book The Structure of Scientific Revolutions be a paradigm shift look at the big problems don't do something small be elated work be assignment be Jurassic do something small do something really big the changes sort of a a
little bit of a
sort a a a then you know you have to
really know the