The HLF Portraits: Judea Pearl

Video thumbnail (Frame 0) Video thumbnail (Frame 10585) Video thumbnail (Frame 21170) Video thumbnail (Frame 31755) Video thumbnail (Frame 36919) Video thumbnail (Frame 42760) Video thumbnail (Frame 44680) Video thumbnail (Frame 52744) Video thumbnail (Frame 60808) Video thumbnail (Frame 64714) Video thumbnail (Frame 78734) Video thumbnail (Frame 88733) Video thumbnail (Frame 95463)
Video in TIB AV-Portal: The HLF Portraits: Judea Pearl

Formal Metadata

The HLF Portraits: Judea Pearl
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: Judea Pearl; ACM A.M. Turing Award, 2011 Recipients of the ACM A.M. Turing Award. the ACM Prize in Computing, the Fields Medal 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

Point (geometry) State of matter Multiplication sign Forcing (mathematics) Lattice (order) Student's t-test Mereology Revision control Estimator Process (computing) Integrated development environment Computer cluster Right angle Theorem Musical ensemble Freeware Family Metropolitan area network Plug-in (computing) Social class
Context awareness Euler angles Orientation (vector space) Multiplication sign Direction (geometry) Decision theory Source code Shape (magazine) Mereology 6 (number) Mathematics Semiconductor memory Bus (computing) Office suite Exception handling Social class God Electric generator Block (periodic table) Software developer Moment (mathematics) Gradient Bit Proof theory Process (computing) Right angle Quicksort Service (economics) Observational study Divisor Student's t-test Graph coloring Field (computer science) Power (physics) 2 (number) Hypothesis Goodness of fit Causality Term (mathematics) Natural number Authorization Nichtlineares Gleichungssystem Booting Inheritance (object-oriented programming) Planning Word Personal digital assistant Physicist Video game Musical ensemble Family
Ocean current Point (geometry) Statistics Group action Multiplication sign Decision theory Direction (geometry) Workstation <Musikinstrument> Bell and Howell Computer programming Number Hypothesis Neuroinformatik Power (physics) Vortex Semiconductor memory Term (mathematics) Different (Kate Ryan album) Computer hardware Energy level Office suite Computability theory Mathematical optimization Physical system Theory of relativity Information Physicalism Bit Maxima and minima Probability theory Dean number Degree (graph theory) Type theory Process (computing) Voting Software Logic Physicist Computer science Video game Right angle Quicksort
Point (geometry) Computer chess Functional (mathematics) Decision theory Multiplication sign Virtual machine Mereology Rule of inference Computer programming Attribute grammar Mathematics Strategy game Term (mathematics) Mathematical optimization Thumbnail Search theory Information Artificial neural network Linear regression Mathematical analysis Computer scientist Dirac delta function Performance appraisal Category of being Process (computing) Integrated development environment Universe (mathematics) Computer science Heuristic Right angle Game theory Whiteboard Coefficient Resultant
Computer chess Point (geometry) Divisor Multiplication sign Virtual machine Electronic program guide Hidden Markov model Replication (computing) Rule of inference Computer programming Field (computer science) Neuroinformatik Programmer (hardware) Semiconductor memory Energy level Data conversion Physical system Shift operator Moment (mathematics) Expert system Price index Limit (category theory) Process (computing) Video game Right angle Game theory Whiteboard Identical particles
State observer Pixel Context awareness Group action Multiplication sign Decision theory Boom (sailing) Set (mathematics) Numbering scheme Propositional formula Dimensional analysis Computer programming Subset Semiconductor memory Bayesian network Different (Kate Ryan album) Physical system Potenz <Mathematik> Mapping Inference engine Moment (mathematics) Sound effect Bit Probability theory Category of being Message passing Order (biology) Right angle Resultant Spacetime Point (geometry) Three-dimensional space Statistics Connectivity (graph theory) Similarity (geometry) Student's t-test Number Power (physics) Element (mathematics) Goodness of fit Programmschleife Propagator Causality Natural number Authorization Energy level Computer-assisted translation Metropolitan area network Computer architecture Module (mathematics) Matching (graph theory) Artificial neural network Neighbourhood (graph theory) Expert system Approximation Word Software Network topology Point cloud Game theory Table (information)
State observer Decision theory Direction (geometry) Execution unit Workstation <Musikinstrument> Range (statistics) Replication (computing) Food energy Perspective (visual) Dimensional analysis Subset Formal language Neuroinformatik Estimator Machine learning Bayesian network Different (Kate Ryan album) Oval Arrow of time Diagram Endliche Modelltheorie Data conversion Hill differential equation Social class Programming language Curve Gradient Fitness function Sound effect Special unitary group Flow separation Degree (graph theory) Message passing Angle Order (biology) Right angle Video game console Reading (process) Asynchronous Transfer Mode Point (geometry) Statistics Functional (mathematics) Connectivity (graph theory) Virtual machine Event horizon Hypothesis Power (physics) Goodness of fit Cross-correlation Causality Root Term (mathematics) Energy level Associative property Curve fitting Domain name Complex analysis Information Graph (mathematics) System call Subject indexing
Internet forum Bit rate
[Music] professor pearl take us back to your childhood where are you as a child what is your family like my family came from Poland my father came in 1924 with a few families of Pioneer that we call them today and established a new town the town name is brave rock which is a biblical town which my grandfather thought it would be nice to reinvigorate it used to be a place of learning according to the Mishnah and and they did it this time a British band British Mandate they had no problem coming in no because at that time the British were kind and the I it was twenty nineteen twenty four as before there was a conflict with the Palestinian at that time it started around a time in 1929 but I think it was almost free immigration from Poland the family was committed to agriculture well my grandfather was a merchant textile merchant in Poland but when they came over to Iraq they had to work the land they grew radishes and when the radishes became too cheap they decided to do a dairy and they said my grandfather to Damascus to buy the cows and we had to learn to grow so your father was my father no my father came with 14 years old right he was 14 14 by the time you were born just because we don't have so much time I need to get you born so you know born it I was wrong and I can tell to 6-way and at this time your father and mother are professionals they are because the environment how father was working in him the citrus all chilled then he shifted to became a city secretary yeah he changed profession and so I was born two families that did not finish college even my mother was lot of he came from a cultural house but not a academic so I'm gonna make you ten years old what do you like a ten years old are you are you interested in school teachers are you surrounded by books I think I was a good student but not brilliant I was not among the first or second in class I was third in 404 yeah I did answer the questions that they asked sometimes I asked them in innovative way remember that but nothing to be proud of as a matter of fact I remember the clear day when teacher wanted to meet my mother and I thought oh here he goes again and complain about me that's even know me I was surprised when she came back and said yes he knows when you think you're doing fine me fine yeah somebody knows notice me all right so you're early you're not yet used to being a distinguished not at all are you getting by the time of high school a good education in your estimation that is the point of which I never understood I got the best education one can think Oh tell us about it that is came to my mind when I after getting my tool in the world I happened to see it by Bill Gates it's some meeting that he had with leading of science educators and yes we anything even your childhood that would you would like to share with us and maybe wish spacious or something I said no nothing how did I caught myself by George I had I was lucky to be part of the greatest educational experiment in Maine claimed in history I know it well just think I said if instead of California forces every professor to spend five years teaching high school yeah imagine the kind of education that high school kids will get now imagine further and they're doing it not because they're forced to but because this is a seat is their mission yes that would happen to me my teacher came from Heidelberg really tidal plug in Berlin and you know great scientists at the time that were displaced by Hitler and had to leave Germany and came to Israel and they couldn't find the academic job so they told high school and we will the victims actually not victory although of course they looked at us as replacements for their scientific dreams Wow because they knew you know that we are going to establish a new state a new society where science education is part of being excellent a part of excellence so you respond well to the teaching yeah I responded well yeah yeah I did well I wasn't shiny no but you were receiving I was receiving constantly quite well and also nice thing about them they taught us in historical man chronologically not logically so every theorem and every invention was connected with a scientist face so for
us science is not a collection of facts it is this constant human struggle with a mysteries of nature beautifully said so you are deciding to become a scientist sources no no no I thought I'm gonna go to humanities in the 6th grade of high school and then I switched back to science luckily I don't think I would do very very new merit would I my career was determined on the I went after I went to the army I went to the army name it's called the knuckle copse which was partly agriculture top part is serving in icky boots and partly serving in the army and I thought I'd become a member of a kyboot all my life and I need to study agriculture agar-agar onami and that was my thinking do my army service but then I guess I wised up how I ask people a little bit I'm a little good in math which profession should I go to utilize it let me go start the engineering so I did I went to engineer it was hard for me because it was considered a betrayal the kibbutz idea yeah I said it was I remember a day when I told the the Secretary of the kyboot that I decided to go and study instead I said I guess you're gonna start the music he said no I'm gonna study engineering so engineer may gonna leave the kibbutz it was unheard of how are your parents encouraging this decision or discourage it good enough to go along with anything that I decided to do so now you're an engineering school which is Technion Technion which is a very good engineer a good engineering school I didn't know it at the time but looking back well I think I had the same experience my teachers they're just superb well giants in science giant education and then dedicated I'm waiting for the next moment you take fire or maybe the first moment when you you see maybe it's an engineering college maybe after a direction for your mind for your where will you invest yourself right now you were studying to be an engineer any particular kind of engineer little engineering Electrical Engineering okay I decided to go to electronic we call it low current at the time an exploring engineer because they told me it has more math and I wanted to utilize what I could well I got it so I went to electronic engineering I didn't have any orientation the stuff at I wanted to do was say at the time already but it interests me anyhow and again I wasn't brilliant I was always the fourth in class I couldn't compete with those students who knew already what kind of questions will be asking the exam I didn't know and and eventually I graduated and what I should say is I liked what I studied like because I I participated in the discovery of the things that I study and it because I had a teacher that was or have been several teachers that we're really part of the discovery like Professor Warren Dorf who was one of the great electron microscope developer and so I felt that I am part of that and they gave us the illusion that each one of us could make a discovery even the high school even your high school I get this illusion with the keep on dreaming in to us each one of you can discover another proof of Pythagoras see yes what year is this now that you are the the 1966 I am finished the power me yeah then I went to college okay yeah now what is your next step going to be like step isn't for me it was clear I'm going to go to graduate school okay for that so you're now committed to developing this this knowledge in yourself because it's a way of all flesh because I really want it to you start enjoying it anyone will develop it so you go to where other people go they went to graduate school some people decide it's enough education I'm now gonna make some money yeah I was close to that exception jobs at Israel at that time bright in Japanese so instead of being a clerk in the of the postal office I decided to go to get my masters in Brooklyn Poli okay so now you're gonna leave Israel I'm gonna leave Israel I get married first I met my wife also in college I was in the same class and at the time did this seem like a radical decision to go on the cross to leave the country okay you get good grade you go and get your master and PhD maybe but that plan was to spend one year maybe a few years so get a PD and either used to become a millionaire and then you come back okay that was the plan it was a plan the plan went awry I accomplished this a school I mean because I'm still waiting for the second what happened in Brooklyn in terms of your intellectual career Paulie I was my first encounter to American
education and then I discovered that I am damn good I'm gonna compare myself to others Thank You me might be making some thank you for something and also I will very pool here when we came we had meals in the automat for 29 cents a day and so I was thinking about invention I'd invent something a patent it and I'll become accomplished the second goal invention invention so I was thinking about invention and my wife tells me that I was waking up at night and what kind of crazy idea one of the really appreciate it I came up with a awareness that Maxwell was wrong in his equation yeah the fourth equation is wrong and I couldn't sleep all night until I woke up in the morning and he was right so that was my mood at the time yes what else can I tell you well I need to make you a world-class scientist soon so I need to get you err so how do I get you from the the Brooklands student whose finds he's pretty good who starts being interested in the idea of innovation of some sort yes to somebody who pretty soon is gonna do some groundbreaking oh yeah see what when I look back and people ask me for this kind of question yeah I wrote back and I said there were those sparks of audacity as I can see sparks here it's what they're what I got a job in I'll see a laboratory and I remember that my work was reinvented something and I got an invitation to speak in the American physics society in Cleveland for my first conference so I speak to them and all the senior guys in our salable so you ask me aren't you excited so how do you have a stage fright you know you know what it means it's to be invited to give a talk you know I said come on guys I'm gonna tell those physicists something they don't know and I should have a stage fright what did I get it but it's very factor in your life this attitude yes I think so yes and you know somebody who automatically salutes orthodoxy no no the other way around the other way around so where did I get this audacity to speak in front of leaders of the field or other fields and tell them what they should do how they should do things better right I think I got it from the fact that I grew up in my block no I I drove the bus the first bus at connected between miles down to Tel Aviv and I my uncle was the first one who brought a water pump to the town and I went to high school who was establish it was established yet why are you before I register right everything was new yes and my family and my neighbors and all the people of a town knew whatever they don't do for themselves no one and he didn't have expertise there was no authority there no expertise to guide you yes innovate yourself and they knew that but the shape they think the way they to shape their life in their towns is gonna be the shape of the next generation and the country so we know now that you are in your own mind capable of thinking in new ways not over listening to people who tell you yes or no but you still have to develop a career do you stay in the labs do you then go up to more graduate work how do you isolated in the I had a good arrangement well that I could work on my thesis in the LCA laboratory really and not here are trying a case of audacity I don't want to brag but it's unheard of I went to Brooklyn Poli just for one month and then somebody found the cheaper College when we know College of Engineering so I was there and I visit my uncle in the hospital and it's a Rutgers University there not too good in general but it's why don't you switch the paulie they're good okay I'll try so I went to Brooklyn pouring I said what does it take to Cameron to graduate from your school we say where to take classes in all they said no way I'm gonna give you the privilege of having me as your student you I should use this world in words why because I can offer you something you don't have I worked and I'll set up a tools we have equipment you don't don't think of it so would you like me to be your student okay hey for you I would like to start doing my thesis without taking any classes you gotta take some time not percent said no one pass the time without classes so that means right it turns out I'm a past examiner yes then how did they do is it flying color I came home I said what is flying and you fell that was good I found out it was good and they gave me PhD without taking any classes my god doing my doing my thesis in house a laboratory and descending it and I did differently so now defend it for me what is the subject of your thesis oh it was great that was super conductivity okay super conductivity yeah we worked on super conductive memories right cause magnetic cores were getting too slow and too bulky and even to string them if you make it on superconducting sheets and
you put permanent currents one way or the other you have a bit of 1 or 0 you got yourself a good memory and fast so everybody at that time worked on superconductive memory is the IBM Bell lab and I'll see a level toys and that was my thesis about so for the time it was it was answering a need of the time it was written responsive to the science of the day correct right yeah it was a technological need to find a new phenomena in physics that could serve as a memory as a basis for memory well complete a memo it was before the semiconductors right which wipes away all your all right expertise correctly when do the semiconductor I should tell all so I tell you just a 15 something which you do you know that there is a vortex named after me it's true I'm immortal yeah I thought of a thesis yeah it turns out that my thesis I investigated a new type of voltage in spin film that only twenty or thirty years later physicists started looking at it and they looked they found my thesis they call it pearl vote okay so now I have a star on my name their galaxy I am immortal but your expertise is still gonna be wiped out by semi pierce the technology also to conduct my own it was like them completely I went to that was at a three years after all later of course after I worked for three years on the grated wire my Marines magnetic the semiconductor just bide everything happened yeah so within the world it's a new world I was working here in California in a company named electronic memories and they came where they told me no jobs luckily I had a friend in UCLA or relative that kept the phone number so I called him in jail estrin was his name and I said they I was really audacious not yet UCLA do you see this now when I talk to you look if I can see already they get this was fun I wanted to go to computer software it's tricky no I don't things and they don't think yes he wanted so I went to UC then I try to interest the Dean in hi anyway so he said what did you do in salsa I said nothing but I know I can will said hey crazy he said he asked me I have the people who spent 30 years of their career doing software and you want me to hire you and you've done even a single program I said yes I can prove myself I proved myself whatever I go I don't know where I got it today I wouldn't say that I
just say that it worked it was not no he didn't hire me at you he didn't know you are the others he kicked me out his office yeah so do you say like you see ladies I said let's let him grab a slot I didn't know what the but he said they come over and the next day I was working here at UCLA in the lab and then eventually they want me to teach and I started teaching and how did you what was the direction well in computer science I started teaching the hardware ok computer memories programming think which I didn't do before but so at the time now what year is this it was exactly in I am beginning of 1970 ok so look around with me at the time and computers in 1970 and tell me what the situation is now well I will still at the dream of getting into soft work ok and we didn't materialize because I moved to the different Department the apartment of engineering systems were sort of interdisciplinary and everybody working on a different aspect of life and science and they asked me to teach decision theory and that was a good experience I quit computers and I started teaching statistics decision theory what else probability theory yeah it was a good education for me is this and remember I'm a layman so I may not ask this the right way you know is this statistical logic that you are entering into as a as a discipline at this point you can call statistical logic because I don't know of any other statistical logic you know I had to make decision optimal decisions right what how to deal with uncertainty right satty geez that SATA stations developed in terms of optimizing actions under uncertainty from books like savage from foundation of statistics from books like Rafer in Schleifer ferguson was influential so i've had all this statistical decision theories books and I taught it and I liked it because you know it had a philosophical power to it yeah it has to deal with scientific methods you act in the world the world gives you some information you have to utilize the information is to its maximum degree and you have to decide what is relevant and what is not relevant and make a right decision so centimeter presented the philosophy of science cast in statistical terms now you're going to fold this back into computer theory quite soon or not how long not yet it took me a few years before I was rehired by computer science into into computer science maybe it was a bit meat seventeen okay I started teaching I continued teaching decision
the--with to computer science but I dressed it up in artificial intelligence hi yes because artificial turfs at that time began to worry about uncertainty and I think this this statistical decision theory was only need formal way of addressing it and so I taught decision theory to computer scientists under the cloak of artificial intelligence of course I also taught elementary artificial intelligence which was search theory game playing heuristics and I married it to my first book was about heuristics rule of thumb because many people watching this will be at an early stage in their career I'm just wondering you you're still a young scientist and this you are looking around in the world are you beginning to construct questions that you want to
answer yeah this was a silly question that bothered me today to do so well in a foreign environment human being yes the world is full of uncertainty the amount of information that you hit a batter world is miniscule and despite all this missing information we do so well with rules of thumb heuristics and the embodiment of heuristics was in game play chess playing though all those games require that you look ahead you assign a dirty evaluation function at the home you roll it back and they decided this move which is a metaphor for thinking until today it's a very powerful metaphor to the way we think so in the process of thinking about machines at this point and artificial intelligence being very generally spoken of as the attempt of machines to make machines think the way we do as human beings people are expecting a great deal of machines at this point what what is the hope for machine is the term machine learning used yet just beginning just beginning there was only I think one paper by Samuel about machine learning checker his ideal world you playing the machine with this function it evaluates the board strength and it's a delta function does not activate but you learn it so you look at the several moves I think he use linear regression and he adjusted the coefficients there much like they do today with deep learning he adjusted the coefficients of the attributes which make up that function and optimize it so you can play better mm-hmm so I still have his original paper job so what role are you gonna play in this quest analyzing another analysis what makes one heuristic better than another okay what what how sensitive is the effectiveness of again the probability you make the right move to the sloppiness of the heuristic and these were all mathematical results about about a game playing endeavor that most AI people at the time were investigating but most people dealt with it not mathematically and I try to introduce mathematics to that dirty game right so you are very much of a movement that wants to mathematize the scruffy and you achieve this I mean yes certainly I made some results I'm proud of well please be proud in front of me tell me what the breakthrough in machine learning all right in game playing I could point out one or two breakthroughs one of them is the automatic learning of heuristics and the idea were the heuristic is just simplify you make simplifying assumption about the world when you find the optimal strategy under those simplified assumptions the strategy that you find under these simplification becomes in the strategy it becomes your heuristics and this happened to be very lend itself very well to auto automation so people in planning out doing it now still using this idea that university is nothing else but optimization under simplifying assumptions now is the machine now of the future that people are developing a servant of here the human will what is what is the notion of where machines will fit into the process of human thinking I think if this time people thought that if they get a good machine playing program then they solve a huge part of human intelligence really yes because they saw human intelligence in
what you would might now call a very limited way yes yes and that time it was big wish machine playing it was a huge chunk of AI at the time and this look at the textbook a text bubble totally game playing that you to believe this I mean you are participating in this intellectual moment correct yes so maybe it's crazy to jump so fast but it seems to me as I read your work and so forth that your leap is from thinking about intelligence in this way to intuition which is not intelligent I did not only intelligence what intuition plays a role in machine in the game playing so you are already dealing with the computers anyway intuition okay the intuition that a chess master will have looking at the board and say how strong it is Oh what makes one move stronger than another yes you have intuition just look at the board and see it is stupid move hmm that intuition we wanted to capture in a machine yeah and are you successful doing this this is a stage I don't think there was then a breakthrough in machine playing like we have today with alphago technology was not there and the maybe the memory I mean it was also yeah it was a big limitation resource limitation yeah of course but then I moved out - totally - because expert system came on board yes and that was a shift in the emphasis in machine intelligence people were just automating the doctors and mineral explorers and not doing anymore and we had customers machine playing didn't have a many customer that expertise in it which is the replication of how in a profession and any field you go about making developing insights is that essentially what expert systems are a system is a system that emulates emulates a professional right a chemist how a chemist might chemists or lead or whatever the physician is the best example our physician distinguish between symptoms that indicate malaria from those indicators Lou that's it's a process that the idea was we want to emulate the doctor or a sister doctor in advice but we are gonna save a tremendous amount of time of money because those professionals are being paid so the game was to replace the highly paid professionals and you're you're in that game no no I heard about this game the level was how do you represent uncertainty because in all these professions you have uncertainty and they the technology was illicit from the doctor to the professional his method of reasoning rules that guides him in his professional in his professional life yes like you see a symptoms symptom a and then you feel that there is a hunch that you have malaria with a certain certainty factor it's saying that you know highly likely to have malaria right but then you have another symptoms which makes you highly likely to have flu and you combine them together and you come up with another certainly factor that you may have just ordinary headache right so the physician had used to its be interviewed like I'm being interviewed now and revealed the way he thinks about all these rules that drive him that guide him in his professional life and a programmer would sit next to the professional write down all the rules in program them and emulate it on the computer and then there will be some conversation between the professional no I wouldn't do that at this point I will change my mind but you told us that you go to you did using this rule and Ethel it's not that Bank coincide with what you're saying now when I change the rule and that would be back and forth eventually you'll have a expose system which emulates a professional so certainty because of course I've read something about your work as it has proceeded to
this day so I'm gonna echo what I think
your sets of stages are and say that this is the stage in the causal ladder of observation this is noticing and seeing but no yes that was at that time they wanted to predict in order to diagnose diagnosis was the main thing okay you are facing with a bunch of observations right and you got to decide if what the chances are that you find oil in this thing right so you're translating the collection of observation to belief in a certain proposition right that was the name of the game at that point I noticed that people are not doing it formerly the and I said we should do it correctly what is correctly well what whenever you hear scientists say let's do it correctly it means that it matches his vectra's own prejudice right yes my prejudice was statistics and decision theory and the probability theory because this is what I told him College 1970s so let's do it correctly and I notice that we cannot do it correctly because the memory space requirement under exponentially exploding and same thing with the time that you take me to eat traditionally traditionally means to build a table of probabilities and then manipulate it so I decided decided I simplified things and I thought what would be the kind of architecture that will enable us to overcome those memory it's only exponential memory exponential time requirement and it's time that we should learn learning from the way human doing it and of course we don't have exponential time exponentially memory here and I read a paper by a rumelhart Roman hot the 1976 a paper paper about how children read text and he postulated
that they are doing and checking that is ahead checking their architecture there between the semantic levels of the pixel pixel level the letter level the word level and the sentence level and they all collaborate in past messages to each other and eventually the system after passing matches relaxes to the decision I'm reading the cat or the word is car not cat okay depending on the context Jennifer and that was very appealing to me and if we if each five-year-old child can do it and obviously the pillow can do it right so let's think about an architect that will give you the let's results restricted by this message passing architecture I found one the three if you have a three and you pass messages then eventually things relax to the correct probability I was very happy with it I was a very happy and I didn't even publish the paper and saying this is the origin of the prop probably yeah that paper was named Reverend base on inference engine okay in front of an engine was a name he used for the excellent system but everybody discounted probability they had we came to ill repute because of the exponential demands and so I published this paper how to do what with probabilities of what people says you can't so you're challenging again in your audacious way the general assumptions which are assumptions I simply felt that we should do it number one correct and number two efficiently and mainly to marry the two if possible and it came up with a belief propagation system and 1985 I even called the name Bayesian network and then and it's received well I mean people do listen to this not yet not yet you know it's it's well among physicians but not in the AI community not the AI you know when statisticians it was a group from Denmark Louis named his company and they built Exorcist a very good medical diagnosis system based on his ideas and but then I made or they should step another one the book I gave a homework well we couldn't deal with loops trees and Polly trees were okay but with loops we had a problem because we have a belief propagation messages propagating constantly repeating itself not cut sometimes converging to the right answer so I gave it our work to students find out the effectiveness of this approximation scheme with message are passing from one node to another as if they were situated in a poetry forgetting the fact that they already residing in a loopy Network see what's happening it was a slip of the tongue that male Bayesian network work of course no students could solve it right but people who notice it and it's right it it's worth beautiful so repeat because I think you've said a bit of this how why the artificial intelligence community is resisting this and what at what point do they begin to embrace it as an idea they resist it is because first of all because of the bad reputation this probability and yes and secondly because there was suppose in the middle of the tension between the Scruffy's and in Metis then it is wanted to understand what they are doing also having at least some formal element to it so they can govern have a guarantee of performance and the scruffy said let's build a system that works and as long as it works it's good enough was we have a similar situation today but so they need not in command Scruffy's well that was you recently think that it first Bayesian network but not received well but then with the success of the belief propagation explain explain that success people could was easy to problem all you had to program is what message one note sent to the other it was a synchronous you didn't have to tell when the message are being sent yes wake up in the middle of the night look at your neighborhood send a message whenever they wake up they will do the same right it's very easy to problem its if it converges you're lucky most of the time it is convert to the right answer it was boom everybody could program it and it was transparent because the way that the network was arranged matched your cousin's understanding of the world so and that made a big difference as opposed to the expert systems was rule-based he could people could program diagnosis of the car car engine the way they understand that the car works using a mapping of the component the visual component of the cloud which component effect another and so it was a threat terrific advantage to be able to redo things when you know when a new trial engine comes in you didn't have to change the entire system you only had to have to change the module that was normal that reconfiguration ability made this has made people want to use Bayesian network and for expert system so at this point now intellectually you're now at peace with or they are at peace with you the artificial intelligence community I mean they have accepted these principles essentially yeah it's essentially a Bayesian network made me famous in the at least known in the official intelligence community right so clearly a man of your nature is not going to be happy being accepted by the orthodox community so no the moment they became commercial I lift them because I realized something yes I realized the reason for this reconfigurability feature and this explain ability feature it was that all the Bayesian network even though it wasn't causal yes because it's probabilistic it was matching our causal intuition about the world and that we didn't realize that that was a power thing so the causal element was really responsible for lots of the acceptability of Bayes net so when I realized that causality is not relieved it's a the third dimension to understand causality you have to go to dimension above probability that properties could not capture causality I shifted to study Authority so is this
now the next stage of the exploration of causation and machines that and that's what I published a book ok about Bayesian network it's a very successful it was very successful book it was called probabilistic reasoning and it was published in 1988 and at that point I felt like an apostate or did it what I say in my book yes yeah from crop into a bastard yes because I realize already that my love was with another girl we'll probably the causality and that probability he knows caption causality as we have to go to another dimension also altogether what is that dimension of thinking explain it to me now we've we've left probability at the dance you're now embracing causation at another level what is this new level level of cause an assumption I call it okay console listen which means the correlation is not causation that we all know but it's much more profound and it's it's means that you need to have causal assumptions before you get causal conclusions you cannot get causal conclusions from correlation alone from observation alone no matter how you manipulate the observation no matter how smart you are in look at them from one angle to another manipulate them if you don't have intervention or assumptions about hypothetical intervention or assumption about how the world works from the causes perspective you cannot get because in conclusion it was a clash of language a class of computer languages the computer levels language he chlorophyll statistical analysis requires another component and this is causing assumption so you say I'll accept causal assumption good you know I'll do causal reasoning with close it assumptions you can't we can because you need to express that assumption formally in what language we don't have a computer language for it well I discovered that we don't have this language for this essential component of human intelligence you look at science it is no language to say the rooster crow doesn't call the Sun to rise right it just correlates with that you can say correlates but that's not the same yeah you and I know it does not cause the sun rooster crow does the toll of the Sun even proceed and correlates but it doesn't cause it right so just to say I'm not trying to prove it yes I'm just to say that sentence doesn't require the formal language and that language it doesn't exist in the statistical marina it must be something extra statistical yes once you realize that you say what is it well is this where we are now we are now yes in 1988 the unit I left probabilistic reasoning and I went to causal reasoning and theis by that challenge indeed linguistics challenge yes and I found that there is a language like it was done by civil right geneticists in 1920 it's called causal diagram we just wanted to say that the rooster crow doesn't cause the Sun to rise don't put an arrow between putting a growl and don't put an arrow between the roots a crow in the center as though it was leading to that yes hello means that it has a cause of power yes not having arrow means it cause a leave me relevant hey using these arrows you can augment the data and event it with is caused an assumption that are some badly needed so now we are in the business of marrying qualitative language of diagrams with sophisticated data and statistical analysis on that marrying the two because you don't want to leave that the qualitative the grammatical some Shinzon the on you you want to combine the read data and get quantitative estimates to what degree does a cause be what degree one disease cause symptoms one treatment kills cancer these are complicated domain we cannot but turns out that you need only qualitative model conceptual correlative who calls us home and that is enough for you to combine the data and get the answer you don't you phones we come to the end of this you come out with a a resounding call for understanding this in the book of why yes just came out i think very very recently it just came out this month and is this basically a call to the AI community to redirect their their in their energy their philosophy their hopes for the machine i
mean what what are you asking it's a call for several communities okay the machine learning community is one of them because the machining techniques without being too so today we're all statistical mode in statistical modes which I call not derogatory but I want to use the term care fitting not in derogatory things but in order to transfer the idea in the same way that you pass a curve to a bunch of points this is all what is done in machine learning in the sense that you don't supplement it with any information which is not in the data yes yes restrict yourself the information the data can supply you and you're trying to manipulate the data by fitting very good function complex function with hills and valleys right crazy function so the machine learning community needs to hear your message yes who's what other community the island of resistance there are islands of resistance in economics in statistics but do not believe in graphs and they believe in the difference first of all they don't even know about causality most of the stations do not know how to handle cause effective range because most other stations are hooked enchanted by the same siren songs as the machine learning people let's get it all to the data okay they trust the data alone to provide the answers yep for two reasons machine learning people because of that it's beautiful you don't have to think it's a decision that do you think for another reason they don't have any other language all the world is to summarize that I know that when such stations will listen to me today I'm worried about school yes I'll worry about it informally it's true who else will employ it's a station who cares about data they care about the effect of treatment they see so the highest attestation to decide cause effect relationship if a decision says I'm sorry all I can give you associations okay somehow they manage the salary even though there is a cheating here you know and because such decisions is bound by the language he or she learned in grad school and the language of Statistics is void of causal relationship you think any statistic textbook and a search for cause or causality in the index and you will find none and so that has led the artificial inquiry in the wrong direction in terms of getting replication of human thinking officials are not worried about human thinking they are worried about getting the drugs correctly ok proving the right drugs ok but those who are who want the machine to approximate human thinking are still not thinking in causal terms correct mmm thinking in terms of my fitting fitting data fitting data and extracting as much information that you came from the data thank you it's the beginning of the conversation but we ended now thank you