From Codebreaking to Computing: Remembrances of Bletchley Park 50 Years Later

good day my name is Jan Lee it is my pleasure on behalf of my colleague professor Golda Holtzman and the department's of computer science and statistics of Virginia Tech to welcome you to view a series of interviews with Jack good and Donald Michie conducted by David Kahn and Karen Frankel professor good was born in London England in 1916 and was educated at Jesus College Cambridge under the supervision of G H Hardy during World War two he was seconded to the government code and cypher school at Bletchley Park where he worked as a statistician with Alan Turing Hugh Alexander a max Newman in the highly successful efforts to break the German high command codes known as AG nygma and fish Donald Michie born in 1923 in Rangoon Burma was also located at Bletchley Park working on the naval codes educated at Rugby School and post-war at Balliol College Oxford dr. McKie specialized in linguistic techniques to break codes with Max Neumann together doctors good and Mickey enhance the mechanical methods of code breaking initially using a series of machines with the unfortunate name of bomb and later with the sequence of machines named the Robinsons with Newman and touring they conceived of the construction and usage of high-speed electronic devices which ideas were implemented as the Colossus machines these machines were operational by d-day 1944 and significantly contributed to the finalization of the war in Europe since World War two doctors good and Mickey have developed their talents in slightly different tracks doctor good after working at the University of Manchester during the development of the first door program machine became a preeminent scholar in Bayesian statistics and is one of only eight on remembers of the International Statistics Institute he has served as a university distinguished professor of statistics at Virginia Tech since 1960 and holds adjunct positions in philosophy and the Center for the Study of science in society Dr Goode recently appeared on the BBC television programme the strange life and death of dr. Tory Donald Michie completed his studies in human anatomy and physiology and earned a doctorate in mammalian genetics in 1953 later he returned to the study of the interrelationships between human intelligence and computers found in the centre for machine intelligence at the University of Edinburgh and later founded the touring Institute in Glasgow Scotland dr. Micky is perhaps best known for his farsighted work in artificial intelligence and for his being on the losing side of a bet with David Levy the chess master on the date by which a computer program would beat a world master at the game dr. Mickey's work with Alan Turing and his early efforts in artificial intelligence were recently highlighted in the joint BBC PBS television series entitled the machine that changed the world David Kahn is an editor for Newsday specializing in the op-ed page which faces the editorial page he's a world renowned expert on code breaking and has published several key books on the topic his latest book seizing the Enigma appeared in the top 10 lists of the New York Times ratings for nonfiction for several weeks in 1991 dr. Cohn is also an editor of the specialised journal Cryptologic Karen Frankel is a senior editor with Association for Computing Machinery and has conducted a number of interviews with recipients of the Turing award the highest award for technical merit in computing generally regarded as a Nobel Prize of the computing field Karen is one of a very small number of writers who has co-authored a book with a late Isaac Asimov their joint effort was entitled robots machines in man's image funding for this program was provided by the departments of Appeal science and statistics at Virginia Tech the College of Arts and Sciences and the division of research and graduate studies additional funding for travel was provided by the National Science Foundation and the virginia academy of sciences dr. McKee's visit to the United States was funded by the Garvin visiting professorship of the College of Arts and Sciences at Virginia Tech the event views took place in the TV studios of Virginia Tech on a 30th of April 1992 dr. good I thought that we might start with a little background as to your early years perhaps you could tell us how you became interested in what would become computer science probably mainly during World War two I should think when I got used to certain kinds of computers I'm not sure whether I read it much science fiction before the war to get me interested in such matters how did you land at Bletchley Park well I was reserved they say that is to say they didn't seem to want to call me up into the army I wasn't the right type or some some office from the gentleman or something and a few of us doing research were put on the so called reserve list until jobs were found for us it took more than a year before they did interview me for a job in the Civil Service India Bletchley and I remember the interview for example what sorts of things did they ask you in the interview I don't think they needed to ask anything in particular the postulate already done a security vetting so they knew rather a lot about me in advance Alexander was one of the interviewers he was the British chess champion three-term is a very intelligent guy and I had known him in the chest world so we would to some extent I suppose friends we didn't know each other very well yet at that time but at least he knew that I was a fairly strong just they're not in his class although in five minutes yes as a matter of fact I could often beat him at that time he wasn't probably able to make very clear what the assignment would be at Bletchley as men although my friend Bernhard Scott had somehow guessed what what was being done in though I don't know where he got his information and we were interviewed at the same time well it was rather kind to me because he said wear your scarf inside your coat not outside just so you don't look like an undergraduate even though we were both possibly being interviewed for the same job he still gave me adverse so then you must have been just out of Oxford and Balliol well I wasn't in bay Lille I was at Jesus College as a graduate student I don't think I had yet got my PhD I had a prize in mathematics but I haven't taken my officially taking my doctorate at that time I've been a student at Jesus College for about five years doctor good tell me what did you first hear about code-breaking and when did you first learn what you're going to do when you arrived at Bletchley Park when I first learned about code-breaking in general just as part of our general knowledge when I was I suppose about 12 or 13 years old the notion of a simple substitution appealed to me that's a very trivial code of course or cipher whichever you like to call it although it's a lot more sophisticated than the Caesar alphabet at any rate right and I used to enjoy myself just breaking easy simple substitutions when there was a child but at Bletchley I first heard what was going on officially from Hugh Alexander and make me at Bletchley Station and as we crossed the field he told me what was going on there of course he shouldn't have been so open when we weren't inside the office did that did that talk make any kind of an impression on you oh yes an enormous impression I it made it I was extremely excited we were just beginning to get into the naval enigma at that time mm-hmm what was the time of that when was that that was the same day the Bismarck was sunk I think the 27th of May 1941 that's right I remember someone coming in when I was in the office with his thumb down during the day we knew that the Bismarck was being chased that was just public knowledge so all he had to do was to make that sign and we knew what had happened let's see now the Enigma was a German cipher machine is that right used by all the German armed forces and the army the Air Force and the Navy and the army and or particularly the airforce enigma had already been solved by that time but the naval enigma was a more difficult task because of its keying method and you were assigned to work on that is that correct yes there were also eight wheels in the library so to speak three in the machine but eight available and I think they I'm not sure but I think at that time there were fewer on the earth force that's correct the Air Force and the army had only five we had eight yes well that increased the number of possible real orders to I think three 36 8 times 7 times 6 instead of 60 which is 5 times 4 times 3 that's correct yes a ratio of 5 was a extra just from that alone and what did they have you doing I mean how are you going about cracking this machine what were you doing sitting at a desk I suppose you were seated at a desk what was your actual work what did it consist of well the week I was in a kind of a mathematical set subsection of hot 8 which is where the naval enigma was attacked there was also a linguistic section we were connected by a little hatch and most of my work in that section was on a process called Banbury smus Banbury smus yes it was named for a named after the town Banbury where these sheets for the process were printed as you say a lot of the work was sitting at a desk but also one had to get up frequently to have a look at the punch sheets that went with the process which were in the middle of the room and comparing we would compare them to make sure that no mistakes have been made and what were these sheets used for were they were those the ones that were placed one on top of another or something like that if I remember correctly yes they had punched holes so that when you compared them well on top of the other you could see where the same let the same cipher letter appeared on both sheets if you had a black background for example so it was very convenient for finding so-called repeats and you use this then to determine the settings of the code wheels that form thee as a where the heart of the Enigma as you mentioned the eight gold wheels three at a time yes the the number of holes that you could see and whether they were close together and so on would give you a probabilistic information of whether in fact the comparison was just chance or whether it was it were they're really represented that the two messages were insightful in the same position right now who are the some of the people you whom you were working with at this time in her eight which was the section solving the naval Enigma Turing was the head of the six shall ensuring yes there was who Alexander guy named Kendrick Peter twin John Clark who nearly married during the one time is now a married her name is Joan Mary now she was a featured if that's the right word in the in the play breaking the code which were course was about cheering right did Alan I'm sorry Karen what was your first impression when you met Turing do you recall the first conversation say I was somewhat awed by him I don't really feel that I could do a thumbnail sketch on him it except mentioning his idiosyncrasies that I discovered later on you see he seemed a little reserved a little shy perhaps so he already had a reputation you had already heard of his his genius and sir Leon I'm not sure about that I may be misremembering and had he made had he made any substantial contributions to code-breaking to the solution of the Enigma as compared to his on computable numbers and other contributions to mathematics oh yes well he had a very neat idea concerning the so called bomb which was a machine cryptanalytic machine used for the breaking of of the Enigma and he did have a rather fundamental idea that I think could have missed being missed for quite some time you know during the war it wasn't like peacetime research when you had several months to think of our ideas and work them out in detail it was a question in many cases it would be a question of who's happened to think of something first but I think we could have missed I think we could have missed his idea for quite about several months what was this idea can you explain it to us well basically it was based on the mathematical fact that from contradiction a logical contradiction you can deduce everything and curiously enough although that sounds rather like logic chopping it turned out to be rather important in this application in its lacked it's like glass of course I mean that alone doesn't explain that degree but that's really the basis of the idea and of course he was familiar with that idea and had actually argued with the philosopher Vidkun Stein that that idea was important and that constrain would not accept that it was of any importance so it turned out to be of considerable importance and how did it manifest itself in these bombs which were the code-breaking machines for the Enigma well it meant the from a given assumption if you were making it if if you're making an incorrect assumption about the settings of the machine from a good a given assumption provided that your crib if I may use the expression was of reasonable length you would as you would assume something about the plugboard of the machine and then the bomb would make all possible deductions for the same platter the same plugging of the letter that you were looking at and on the other hand if you were happen to be looking at the correct position of the settings then it would only make may one assumption it couldn't contradict itself because it's it started with a correct assumption so all you had to do I would say all I mean this is an electronic problem how you do it would Orin let electromagnetic perhaps would be to look at the row if I may use the expression corresponding to the different possible assumptions and see whether they were nearly or linked or nearly awful so there was a very quick test of whether you made the right assumption and that saved a factor of twenty six in the running time I see so if I understand this correctly what was going on was that the British guessed the plaintext or the original German of a message matched it against the German the intercepted German code message and then the bombe ran through all these permutations of these wheels and using this theorem or concept of Turing's if this Row was filled up the machine would continue on because it meant that the assumption was wrong as you said from a single logical contradiction everything would flow and if it were correct then the machine would stop and he would be able to detect which were which was at least one possible setting of these wheels which might be correct and then you were tested on naturally yes I missed out one possibility and that is that the settings of the wheels are correct setting the fee is correct but to make an incorrect assumption that was a case that I hadn't covered and then you would get all deductions except well except the correct one right and that would also create a stop in the machine yes yes and using this technique then and with the help I believe of some captured documents you are eventually able to go on and pretty much read the Germany naval enigma for a good part of the war there was a blackout period and then you continued once again but then you moved on to the other area of this new German machine the Colossus well Tunney was the German machine Tony was the German machine I'm sorry correct and now why were you chosen for that amongst all the people who were working in hut 8 on the german naval enigma or perhaps in the other huts working on the army or Air Force in it well you picked up but it was largely because of the crib situation had become so good that member isthmus became rather unnecessary I think that was the main reason so the people who are doing Bamber is Messala needed in that yes that's right and they tended to be dispersed right and so you moved over then to there's no machine how did that happen and who was in charge and how did it go from there when I arrived in what was called the Newman area because max Newman was in charge in Newman read yeah yeah we'd like to use fancy words like that there was a fairly small section I think there were 16 wrens there was one machine called a Heath Robinson which had just started being used and of course there was max Newman there was some engineers there and there was Donald Michie who was the only other Crypt analyst there at that time and in fact he'd been a classic scholar at Baylor College so doing this kind of work was somewhat new to him but he took to it very well and um and was very enthusiastic now how far away first of all who was max Newman he was a mathematician interested in logic and in topology he'd written a book he was a fellow of the Royal Society and I suppose at that time he was probably in his early forties and he was from Cambridge yes right and why was he chosen for this job do you have any idea well partly he came voluntarily to Bletchley because he wanted to help with the war and he was in a section at first called or perhaps earlier called the test story he may have been in the research section this is major testers tester II that's right yes but he didn't take to the work there he didn't like it he felt what were they doing is what were they doing there they were working on Tommy of course but they were using hand methods right and some of those were very looking very laborious and he felt that they were some of them the ones that didn't involve any language at all ought to be mechanized and he felt very strongly about this he also felt that he wasn't really competent at doing the work himself because I think he compared himself with some of the other people there who were very very good and thinking especially of Peter Hilton right for example so he became discouraged so he had an extra reason for wanting to mechanize the process and get get out of having to do it by hand so he went to Travis or Abril Travis and said that we ought to try to do this work by electronic machinery he probably knew of the existence of the bombe which was of course being used against the against the Enigma and Sir Edwards of course encouraged him and that's how it got started and Edie excuse me I just want to clarify once again with Travis was the head of Bletchley Park the head of the entire will actually services code-breaking he was not in charge of the diplomatic that was in charge of Alastair Denniston who had been his predecessors head of the whole oh yes that's right well Denniston may have been may have been in charge I may be wrong when I said that he went to Sir Edward Travis it mapping denistoun what what time was this when was this that would probably be in the mid forties I should think well by then under the Denniston who was then commander Denniston I mean Amanda Travis 1940 I meant not married Fortas in the middle of 1940 the 1940 Denniston was still in charge I see yes but anyway that's what we went to yes Karen I'm sorry how closely did you work with Donald Michie were you deciphering or working on the same messages together how close was the teamwork at first we realized that the machine called the Heath Robinson was not working too well sometimes it caught fire but there were also problems of not really knowing what runs should be done we did some research on the statistical properties of language and of the characteristics of the patterns of the machine and we worked very well together we were friends from the start what would happen when a message came over I guess the wires and how did you know which ones - how did you prioritize the work the prioritization if I may use a horrible expression as was done to some extent by the intelligence people I mean for example much later we were told that the most important link was usually link called jelly fish between us I think Stuttgart or Strasbourg and Paris Strasbourg is not the city in France Strasbourg was a commune extent - communication center just outside Berlin oh yes it was a different Strasbourg yet similar names I'm sorry nicely okay I didn't know that in fact some of that traffic was a very great importance later on that particular link it happened to be easier to solve than some of the others less of a challenge it was a bit disappointing when we were told that we should concentrate on jellyfish so when a message came can you give it an example of what it was like when you were working on a message together and working closely it's hard to say I mean we didn't exactly work on a message by ourselves we were helped by other machinery what we did was deciding what should be tried next if something went wrong and also find out what why it had gone wrong sometimes it's simply because certain tapes have been incorrectly produced by the rings I mean they're they look fallible it was difficult to produce the tapes in a way because you had to count the number of characters there was a little hand counter for counting the length of a tape and it had the exactly right how long what a message beyond average in tally their very often thousands of characters long we called those messages they were really under underlying it they were really civil mists is placed in to end in plain language they liked sending out a lot of stuff all than one go on the long take the longest tape I read recently was 60,000 characters long but we quite often had them would later say 15,000 what were some of the greatest challenges in the longest messages that you worked on well the longest messages were the easiest out because they gave more information so that was fine I don't know what the shorter specialties were we'd probably never even tried to break in so to speak on the short message I don't know I don't remember did you did you use the same technique with Tony as you did with the enigma in other words guessing at cribs and then trying to match them up and then going from that to a reconstruction of the original keying of the German machine cribs would have been looked at only in the test tray in the machine section we never used or hardly ever tried to use any kind of linguistic information so what did you use how did you go about breaking it it was statistical you analyzed the statistics of the intercepted message and from those attempt to deduce the settings of the key wheels and the positions of the pins and lugs and all of that on them yes yes that's right can I go back a little bit earlier on where was the neumann ray and what did it look like originally it was a hot you know one of those those Quonset huts over there they weren't actually there were long narrow buildings right yes I don't know what they were constructed out of but they looked as if they were constructed out of cardboard or less but then when we began using large machines like the Colossus we moved into a brick building we still were called hot F by the way and then later on we had so many colossi that we needed yes another brick building Lord out age the word hat was still used right and how many people worked in at the start at least in the Neumann reef well when I arrived as I say there were about 16 rings two or three if you include human three Crypt analysts maybe half a dozen maintenance engineers or a little more and of course there were also engineers in the background who weren't working at bletchley much of the time and later on when the colossi were being built half the Dallas Hill Research Station which was the telephone research station was working on producing colossi right but they they didn't even know what they were working most of them they were not they were not at Bletchley Park they were up at dollars he'll miss that's right s so if you walked into the Newman REE what would you see would you see desks would you see people working with screwdrivers to put these machines together and what would you hear and what would you smell what would it be like in there well it was rather hot because of the heat produced by the Heath Robinson in the early days of course I'm referring to now you did ask me you know the size of the department in the early days although I slipped into saying that half the Dallas hill Research Station was involved that was much later of course there was an air of informality but devotion to work would you expected during World War two there was hardly anybody I think in the country who didn't have that feeling that this guy this mega murderer was someone that needs to be disposed of and wait were you seated at desks and were there people as I say putting machines together in the same room no only maintenance people came into that particular room there was another room I think when they worked on other types of equipment and of course they also had to produce a simulated Tunney machine because when you did break into a message you wanted to decipher it and you didn't want to do that by hand there's thousands of letters long so this simulated tanning machine was simply called Tanny sometimes now the the bombs which were used to decipher this entirely different machine the Enigma were huge things they were about five or six feet high if I remember maybe a little more and about seven or eight feet wide and a foot or two thick if that's correct yes right on except I think I was only once in the place perhaps where they had these machines and I think it was taller than a man you may be right I think it may have been about seven feet high mm-hmm now how big was the heath robinson and how big then was Colossus in comparison with these the Heath Robinson this big or oh no it wasn't the larger than that it was something like 4 feet wide I should say about 5 feet high and about 18 inches thick and did it have flashing lights and things like that on it oh yes it's not impressive oh yes at that time in history and I think it probably had about 30 vacu well they weren't the vacuum tubes that perhaps not recall and choose they were thyratrons which were guests filled tubes of the order of 30 but that's just a guess on that part that was the throb the Sun mm-hmm and then you moved from that into the Colossus is that right yes and that was how big the Colossus I should say was about mmm about 11 feet wide about 6 feet high and about 4 feet thick something like that the first Colossus had fifteen hundred thousand men which made it very probably the largest electronic machine that had ever existed at the time and did it did it have flashing lights and did did it have sprockets and things running around such as tapes of intercepted messages anything like that oh yes well only Heath Robinson there were two tapes for input and this meant that you had to drive them by their sprocket holes in order to keep those engage and that tended to stretch the tape somewhat that was one of the difficulties with Heath Robinson now Tom flouse who designed the Colossus realized that they that this problem of having two tapes could be overcome by putting vastly more electronics inside the machine to represent one of the tapes that's why it had well 1,500 valves the first the second one had 2,400 I think and since the tape was now being driven by pulleys and not very sprockets there wasn't a serious problem of the tape breaking they did occasionally but the rarely break and was it where presumably it was worth it to replace one of these tapes with the vacuum tubes is that right I mean you it seems like there's a much greater expense and danger of these tubes burning out and all of that as compared to an occasional tape breaking the expense didn't matter much the speed of producing it was the important thing but the number of Valve's did discourage some people they didn't believe that the machine could possibly be reliable with that number of valves or tubes but tom flowers knew that if you did not turn the machine on or off that it would become reliable after the initial teething troubles I see and is this move from tape to to the vacuum tubes with material I suppose stored in them is that any kind of a step towards the development of a computer was that a major step i think the engineers who worked on Colossus probably learnt quite a lot of electronics and that probably pain that some of them will in later life if they got involved in computing I think some of the information must have got through and especially through Tom flowers and the people who worked with him that's what I was going to ask how did your experience with Colossus condition your work later on with computers well in my case I think my experience with Colossus led me to suggest an idea just to suggest to not work it out in detail which would later be called that macro programming because the idea there was to construct basic instructions by means of very elementary boolean operations which is in effect what more or less what the Colossus had been doing so it was very natural or some suggestion to make Wilkes didn't know about that I don't think when he suggested Macra programming and our aims were somewhat different he thought of this process is something that would be of use to the hardware engineers constructing a machine whereas I thought of it as something that the user of a computer might want to do if he felt the need for a new type of elementary instruction instead of writing a subroutine which is what would be done today in nearly every case he would construct basic instructions as the idea and different users might want to construct different basic instructions for example counting the number of ones in a word or something which hadn't had not been thought worthwhile having that be useful for SATA for one of the users so if you just set this up and if he later got on to different work he could change the basic instructions that was the idea and I think almost certainly that that I had this idea because of my experience with Colossus it wasn't influential I managed it was ignored completely except that Torah that Newman did mention it but I think he had rethought it for himself about six months later and discussed it with the people at the National Physical Laboratory so it might have got through you know through the grapevine and after all by the time I told three people and Newman are told a few more that was already about seventy percent of the people working on electronic computers in England what what I was getting at when asking about the move from tape to electronic tubes was whether this was a move towards stored stored programs or stored programming is that can it be looked at that way well the trouble is are not quite sure how to define stored programming really me either so it rather depends obviously there was a program there that was plugged up it wasn't something that would change in the course of around accept that we could as a matter of fact by hand change a program but because just because it was plugged and then we had toggles rashes and so on whereas installed programming you you don't do that you ran a program with the stork program or in the machine so I'm not quite sure really quite how to answer it but I'm sure Donald Michie could answer that better than I can tell us while you were working on this I know you weren't able to think 30 years ahead and imagine computers as they are today but how were you working or concentrating on the tasks assigned you namely solving German messages apparently you were not the ones who were actually translating these messages from German into English or running them through the machine so that they would be in German but what were you doing I mean the messages I know would come in by tape either radioed in or sometimes brought in by courier what would you do with these tapes once you got them we would choose a tape very often being influenced as I said just now by the intelligence sections as to what was of high priority and we would then stick it into a loop so it could run round and round and then the reins would put this on the machine on the pulleys and we would start off with various runs now at first we had to experiment a bit with what were the best rounds to do so we would sit at the Machine and be in synergy with the machine and with the ring so you had man woman and machine in synergy shouting out instructions for what changes should be made in the program such as what would you shout out you'd say oh flag 3 to level 3 to level 2 thing like that and meanwhile the machine was printing out stuff on an electric typewriter I'm talking about the Colossus and it was giving you the results of runs if if you've got a very high score and around you'd believe that there was a high probability that you'd had success on that run and then you'd have to decide what should be the next run later on we made decision trees so that the rains could take over and simply do according to certain fixed rules a sequence of a sequence of runs without the help of a Crypt analyst and if then it failed in the printout and the alphabet in a certain sense would be tacked out and this would be given to a duty officer to consider what should be done next if we had not succeeded at that stage god Karen I'm sorry I was going to jump ahead perhaps and ask what was the most important strategic message that you cracked well I can't be sure about that about to in tally it was probably some jellyfish messages as I said between Paris and what was it Strasbourg yes yeah that had enormous impact on the war in Europe apparently about movements of infantry or battleships what kind of will not battleships in the case of Tammy this would be the German army I see it it was highly influential according to Sir Harry hingeless history of British intelligence and World War two that was a very important period the exact message of course I don't know we we are bathed a principle of the need to know for the most part as in any intelligence organization you weren't told things that your so-called superiors or actual superiors thought you didn't need to know when you did know how did it feel to have made that kind of contribution well it made you feel gave you a Napoleonic complex you said at one point I think I heard you say earlier in one time earlier that you and Donald Mikey had been working at Colossus one in front one in back and you were attacking some particular problem and achieved some kind of a breakthrough I don't know what this was I don't know what the problem was or what the breakthrough is no wonder if you could tell us well the whole purpose of the Colossus was to carry out our machine processes that had been done by hand but of course doing the match faster which of course is the purpose of an electronic computer right now amongst these hand processes there was one called setting then we call it when you know what the patterns of the wheels are you set messages well mister time and there's another process which it which requires the finding the solving of the patterns which would be fixed maybe Friday at first I think for a month and then later for a day or so and that was the so-to-speak theum the break into the day now as I say both of these were being done by hand but we thought of the purpose of Colossus as the setting problem when we got into a rut and we it didn't occur to any of us until Donald Michie thought that perhaps the finding of the patterns also might be solvable on the same machine so he came in one morning with this bright idea and as you said he worked at the back of the machine and I worked at the front helping him and really the effort was certainly mainly his he made a mistake which I noticed so I made some contribution apartment just working at the front of machine and that vastly extended the value of the Colossus that was a very important breakthrough and in fact of the the method led to additional equipment that was installed in later kalasa Ivor Tom flowers John Klaus was an extremely good you can't imagine anyone who could be more helpful and he was regarded as an absolute wizard at electronics as soon as he heard of a problem he would draw out roughly what the circuits ought to be and there to tear up the sheet this is all in the literature and give it to his his band of brothers as it was called sometimes in an article by Coombs and they would get to work and they would produce what was required and he nearly always I think perhaps even always got things right first time I wonder you know the Germans had the decoding machines as well a code-breaking machines they were always mechanical I wonder wore electromechanical I wonder why the British moved from electromechanical machines into purely electronic do you know who took that giant step it's difficult for me to tell but I cannot guess that it was something to do with the telecommunications Research Establishment who may have been using electronics in connection with radar and the Heath Robinson whether the first design there was someone the head of of the engineering was a man named Wynn Williams who works at the at TR early so that could have been it that Turing happened to know about flowers touring was the man who put Newman onto flowers as being someone who might be able to produce a better machine he knew a Turing had got to know of flowers reputation I think it's because he considered the possibility of of making an electronic out of the bomb which was electromagnetic and Keurig authorities yes I think that's why he got into contact with Adonis Hill and then with flowers but that was decided against it was decided that wasn't a good investment but at least that's how he had the personal contact with Laos it's interesting that that was considered not a good investment do you have any idea why and the other one was well the other was a kind of analog machine that was doing a lot of work simultaneously it was a it was really a very parallel machine right and it so it got its speed by by being a very parallel machine and apparently to make at that stage in electronics to have produced an electronic machine doing the same thing would have been unnecessary or no not quite convincingly worth the effort perhaps and what was the actual when we get back to the classes what was it actually doing I mean what kind of tests was it running why did it need to input tapes this isn't clear to me the Heath Robinson did need to detect it yes one of them was the original cipher tape and one was something to do with the settings of the machine you know some function of the of the wheel patterns if there if the wheel patterns were known so there were these two tapes that in some sense rather had to be matched against one another and that meant running them round and round it also meant that the lengths of the two tapes had to be mutually Prime in order to make sure that all relative positions were were rather icy right and so at all possible positions of these two tapes then I ran them through two tests to see whether certain statistical patterns would emerge is that what it was in a nutshell yes or whether certain tests statistical tests were satisfied significance to and if they were satisfied what would that show you it will show you that you Pro up probably correct you could even work out perhaps the probability by using so-called Bayesian methods of statistics you could say the probability of this is correct is such-and-such if you wanted to but in so called non days in statistics you simply say that a certain p-value as its called tail area probability had been reached and it was too surprising to not to believe in effect but of course there were marginal cases right and sometimes you would do the next round in the hope that it would show that things were better than had been clearly shown in the first round and then once you've got these statistical tests came out say very strong in particular cases how would this lead you back to the settings the starting setting of settings of the wheel and the gaim schreiber and the positions of the pins and all of that so long as you knew the patterns then you were setting then when you got a highly significant result it meant that you had found the correct initial settings positions angular positions right of the of the wheels that were used in that particular round that's in effect what you were doing finding the settings of the wheels later on this other process in a nutshell for a modern statistician we were putting it little misleading there but more or less we were finding the singular vectors of a matrix that makes it very clear no I understand yes well that as I say is a nutshell right I see and did you get any feedback from higher-ups about successful messages which is what Karen was asking about earlier well by the time you'd finished all all your work you just knew it was right and the decipherment on our simulated tally would actually show you some German mm-hmm so you knew then that you had things substantially right at least of course you might have a 1 instead of a 0 or something like that on the patent in which case this would come it out as it as a garble as it was called in the German but since there were garbles anyway through transmission right not to some extent I don't know whether that would have made much difference to the to the pitter to the edge of the intelligence before however I imagine that if and this and just reconstructing the rhombus I don't remember this I imagine if they found too many garbles like that they would say oh you must have something a bit wrong here we would recheck I see now dozens of messages no doubt were coming in each day and you couldn't solve every day's messages I don't believe there were days when the statistical tests wouldn't work or something didn't produce anything how did you know when to stop working on a particular day's traffic and to shift over to something else and begin working on a new day's traffic well for the brain the breaking of the patents problem of the so called wheel braking problem there was a significance test which I worked out a little bit sophisticated and that's how we could tell whether the beginning run was successful later on it would make itself clear whether it was successful or not if the significance test failed to work we would often try anyway and nearly always when the significance test failed to work we didn't succeed in in in later runs it were apparently the significance test proved itself in the eating as it were I just have one final question which is this was this work satisfying to you and if so why it was entertaining as a game to begin with and of course it was satisfying because we knew we were helping to win the war we were doing something substantial and something much more important than we had ever done before I think that's very true and if I may just conclude my own researchers have shown that certainly in the Battle of the Atlantic months and thousands of lives were saved and I think it certainly must have been true for the gaim Schreiber and Tunney where you must have shortened the war considerably by the intelligence you were gaining which could be applied effectively by the Allied command and so saved thousands of lives and people now alive today might not have been but for your work that's true I think we've enjoyed now by Donald Michie who is director for Advanced Study at the turing institute in glasgow and who during the war was a codebreaker working among other people with Alan Turing and I'd like to ask Donald who was a classical scholar at Balliol and not a mathematician Bailey ol Oxford how he got into the code-breaking work particularly on the mathematical side I volunteered to an establishment in Bedford to be taught a Japanese course in the early summer of 1942 having left my high school early and but I got the date wrong time of year wrong and the interviewing officer had pity on me and said that if I packed my bags and got there within the next 48 hours I could sign on meanwhile for our codes and cyphers course which might interest me it interested me very much I didn't have anything else to do in the evening but to go back and push on with the course and after quite a short time a few weeks the colonel Pritchard came over from Bletchley because he was recruiting in connection with the setting up of the test3 and they push from Newman which followed shortly after that why do you think they can told you about codes and ciphers instead of had a shooter cannon or something like that well this government training school I went to that happened to be the next course that was starting on a Monday and I was being interviewed on a Thursday so they said if you if you're quick about it we'll let you into this course to fill in the time but it turned out to be more than time feeling because Bletchley started to relay the recruiting and they know I got pulled into the net and then you went on over to Bletchley immediately yes right and you went where you're over there immediately and no I was parked for a - couple of weeks or so two or three weeks in the main building learning some basics and before I left I I remember one of my chores was to teach I'll try to teach Roy Jenkins who you may have heard of to read paper tape code teleprinter and paper tape and I went directly over to the tester II where he followed her almost immediately mmm and what did you do in the tester II what was your job there our main work was doing exploiting by hand methods certain kinds of mistakes that the German operators made when they didn't obey their own instructions and produced what's referred to in hims Lee's book in which we also called deaths namely one in the same message in the easiest case gets transmitted twice on the same machine the same cipher the same patterns the same setting except the second time the tape is put in at a slide of so many characters relative to their first transmission they think something's gone wrong with the first transmission so just for luck they send it again and then what you get is if you add the two together by a certain kind of logical addition the key is common to both so that disappears and you're left with a message in German military German which has been added to itself at a certain stagger so that was the easy case more difficult stuff is where you've actually got two completely different messages that they've been transmitted on the same machine same patterns same setting and then what you have is the boolean sum of two quite different messages one added to the other and by certain amount of low cunning it's sometimes possible to unscramble the - mm-hm in either case either the easy case or the difficult case then you have an opportunity to produce pure key so I suppose that was the commonest thing that we were doing but there was also an operation called Turing isthmus which was derived initially from Alan Turing and it was in connection with that that I consulted him and then you went on over to the new memory yes I was humans first at point EU and he was setting up that must have been in May 1943 when the first Heath Robinson machine was delivered and Jack showed up about two or three weeks later and what was your job in the Newman read what were you doing there well our official job from Max Newman who is a very driving personality was to use the Heath Robinson in the style that it's been conceptualized envisaged to get results because he was somewhat beleaguered having got funds and he wanted results to justify it and Jack and I in the day time obedient they went through the motions but in our hearts we knew that so many things were prototype and provisional and as yet unresearched that unless the equipment could be used to do fundamental research on the statistical structure of the problem the chances were not too good so we got into the habit unknown to max Newman as far as I know with one or two helpers in staying on in evening shifts using the Heath Robinson to do this type of research and collect data of a kind which was then absolutely invaluable on which systematic methods could be built and which did in fact justify human how would you describe the way that you work together it was it a collaboration or just healthy teamwork or how even healthy competition between you what was it like between Jack and me is to individuals yes Oh collaboration that's typical scientific collaboration how did you divide up what you were going to the way you would tackle something to people working close together get a sixth sense about each other's profiles strong and weak points fill in for each other's gaps and stuff and problems subproblems get broken up past backwards and forwards it's hard to be very specific I think that's fair enough except for one little job that I did on the side and that Bri message that I mentioned but I don't think we needs to go into details on that particularly haha would you like to talk about some of your strengths and how you balanced the qualities that the other lacked well this is difficult but Jacky's was it is one of the outstanding applied mathematicians of the generation next newman said of him to me and i believe these are the right words actually the metaphor of the gun that jack wood is like as a mathematician like a loaded gun just pointed in the right direction and he could demolish anything and so that anything which required other than surface mathematics obviously that was for jack and however large the section became and however mighty many mighty mathematicians came into it including Henry Whitehead Jack mentioned if some real hard not had to be cracked in the mathematical sense it tended to land on Jack's plate I think it's much easier to talk about one's weaknesses and then strengths which I'm not sure I know too much about but certainly we we fitted and made a good team after you work that to the end of the war is that right hmm and then what happened did you continue your team work elsewhere did you go your own separate ways what happened at the end of the war we went our separate ways professionally we just stayed in touch as friends on and off ever since we made one or two rather uncompleted efforts a collaboration on one or two problems and we even published a joint paper I remember on Becky's paradox yes that was about eight years ago yeah but in general we just been so busy with our own stuff mm-hmm that the opportunity has arose well what did you do and then I'd like to ask Jack what he did what did you do after the war where did you go and what interest drove you there and did this in particular did this war I'm experienced prepare you in that direction oh the wartime experience completely disabled me for the career that I had plan which was to follow up my classical scholar sure which I had to Oxford and arrived in the fall of 1945 right at Oxford yes paleo and dutifully began to apply my nose to the grindstone again in Latin and Greek language history philosophy and so forth after the bletchley experience I just found that bearable unbearably dull let down and went to my advisor I think in the college called him a moral tutor but every student had one and said it's no good I really can't face any anymore I'm trying my best so he said well what would you like to do sorry I said I don't really know but maybe being doing something in medicine would be at least useful to somebody I said I suppose he realized that you have to do prelim science first and John 22 and you didn't ever you don't know correct me if I'm wrong anything about botany zoology physics or chemistry and he was right i sat down and over the years did the prelim science and went through preclinical medical school and by the time I graduated in anatomy and physiology from I yes I was then rather busy spare time hobby doing genetics experiments with with pet mice and this grew and in fact by that time I already had the loan of a Hut in the zoology department to follow up this research and the reader in genetics EB Ford died recently persuaded me I didn't need much persuasion to do my dear fellows we call it my doctorate in genetics at Oxford so that and that led to a 15 year career as a biologist I was a geneticist at Oxford no no no immediately after my PhD or DPhil I joined Medawar department in London and with a McLaren we worked there and elsewhere in London for six years before I what we both accepted different posts in Edinburgh and I was in the department of surgical science in Edinburgh until 1960 or there but when a learning machine that I had been building arising out of a bet with the colleague attracted some attention and coming out of that I was asked over to Stanford in 1961 and what I saw there completely assured me that the light that had always been at the end of the tunnel of maybe getting back to the topics that we've been discussing during the war it was now a practical proposition that there were actual computing facilities to do work in artificial intelligence can we come back to that a little bit later focus on Jack's post-war experience what did you do after the war well I joined max Newman's mathematics department of Manchester he had recently become a full professor in Manchester having previously been in Cambridge and he was now building up his mathematics department so he hardly as a lecturer and also with minor or subsidiary interests in a computer project that he hoped he would get doing in Manchester he obviously was influenced by his previous experience during the World War in wanting to have a computing Center and he predicted that computing would be a lot a multi-million dollar industry he was one of the few people who realized that was so I don't think he guessed that he would real multi-billionaire billion-dollar industry but certainly he didn't fall into the trap that many of us did of thinking that one computer would be sufficient for all the needs in the United Kingdom were you skeptical of his prediction at that time then I was skeptical because I knew that one computer could do all the computing that was then being done but of course the tools developed needs as GB stibitz pointed out in one of his papers early on from direct experience so Newman's prediction was right but he wanted the computer to deal with non numerical work in fact with pure mathematics and I was a bit skeptical about that but he did have a shot at tars some problems on group theory with David Rees whom he also brought over from Bletchley as an assistant lecturer was it around that time that you became interested in the idea of intelligent machines and thinking machines um yes I consecrate mine Lee of conversations was touring during the war I thought I was also fascinated though not the extent that Turing was obsessed with the notion of producing thinking machines so I was quite interested in them and in fact in 1958 in the I think December I was invited by IBM to evaluate the perceptron which had just been started by Frank Rosenblatt and I thought it was rather good idea but I was unimpressed by his mathematical proofs but I was I think first person to do an evaluation of the perceptron was that the Watson Research Center No more or less it was my honzik laboratory which is close to Yorktown Heights and in fact I was there just for a few weeks and wrote two papers for them one was on the kind of mathematics that might come into information retrieval these were half DeBakey ideas you know and I like speculating about science and especially in those days in fact I edited a book called the scientists that speculates but you said a moment ago that you were dubious about the perceptron is it can you elaborate on that I was somewhat dubious about newman's ambition at that time at least over the short term of devoting all that the Machine effort in Manchester to Mathematica pure mathematics I thought that the time wasn't ripe and I was probably right in that in that judgment it was used in fact for number crunching for quite a time as its main activity there was such a demand for number crunching as I say once people discovered that they could do number crunching most people wanted to do more on more and more of it and do you have any predictions that you made that you'd like to recount to us that you found work correct and/or your speculations then as a scientist that we're correct about how things unfolded in computer science and artificial intelligence I can't remember all my speculations but well I was interested in in chess as a field as as Donald was in fact he owned a Sean Riley produced a little program very early on just a bran move analyzer and in one of my letters to Turing I said that where I could easily beat this psychological play that is by complicating the game if it was only looking run move ahead and I also in a postscript to this to brother my letters in 1948 to him pointed out that a machine should not just look a certain number of moves ahead but should look ahead until the position became in modern terminology quiescent and I think that although of course that wasn't published I think it may have been an early written statement to that effect that some kind of a tree that was not just cut off uniformly like a prunes tree but went forward wherever it was found necessary when the position was not quiescent until it became quiescent was one of the early ideas which occurs in in Shannon's paper on chess as an example I think perhaps I might have influenced touring for a change I mean Turing certainly influenced me in many ways but I think perhaps I might have influenced him on that in that PostScript in are more informal conversations you had sounded out a warning about couldn't control being handed over to intelligent machines maybe you could comment on this yes well I wrote a paper in 1965 called speculations concerning the first ultra intelligent machine and I started off that letter Bert banging a gong by saying the survival of humanity depends on the early construction of an ultra-intelligent machine but I also realized that this was all so dangerous that such a machine if we do if we weren't careful would take everything over an idea that's been taken up in science fiction of course and the question for example would arise of the social repercussions of having such a machine it would certainly produce social repercussions unemployment perhaps because the machine would construct robots and so on for doing manual work in addition to all the scientists becoming unemployed and the question then that I would address to the ultra intelligent machine was how to kill these difficulties the machine might say well you better turn me off I don't know what it would say but but you're not sure you would get a direct or and even honest answer well is that danger too especially if an idea which came up sometime ago was put into effect of letting the machine so to speak fight it out for a natural selection amongst themselves because if they got into the habit of track of self-preservation then they would become even more dangerous than if they were regarded as tools of humanity you may be mentioned science fiction a moment ago are you an avid reader of science I used to read a lot of science fiction at one time I had one idea which I discovered later that had been anticipated by Isaac Asimov which one is that well I thought of all the alt intelligent entities in outer space it seemed to me there would be good reason to believe they existed if there is life in outer space and I think that's you know I felt that was probable and that they would be hundreds of millions of years ahead of us in all probability because a hundred million years is not much compared to the age of the universe and that therefore they would be so incredibly advanced that we could hardly imagine what they would be like and but at that time I also thought that telepathy might be possible and so I made the speculation that all these ultra intelligent machines in telepathic communication would be like the neurons of a vast brain which I called G OD D because it would resemble God so much you've never I know you've written hundreds of research papers and reviews but you've never have you ever ventured into writing science fiction yourself not seriously I once wrote a short story just for typing practice I never miss it for publication well what what do you think motivated you to write papers that range from reviews of philosophers of science like Karl Popper to fractals to almost anything I guess one can think of in computer science what motivates you to just have wide interests as is all I can say really and I have a speculative turn of mind I like pub I have published as I say the scientist speculates in which I put forward a number of our own ideas as well as those of some quite our eminent our scientists and I just like speculation I called them partly baked ideas to generalize the notion of a half-baked idea I had a formula for the lengths that a pea baked idea should be as it should not exceed what what was that if it was negative for example if the formula was such that a negatively baked idea it did not deserve even one word the formula I think was 10 raised to the power 9 9 PX / - was the index where X was the importance of the idea and P with the P measured the bacon as' and well we are parameters for X and P had it well they could only they couldn't exceed 1 so a perfectly baked idea that was also the greatest importance did not deserve more than 30,000 words and how did you and the judgment of the bacon Asst and of the importance were a strictly subjective on your part or did you apply certain tests to it there were no tests except consistency if you thought one idea was better than our then it should have had a a larger value of P or or of X and also the formula itself was based on examples I can't quite remember I think that an idea that was only a half-baked idea deserved only I think 200 words and that formula was put forth in the scientists speculate yes it's in the introduction I think the preface is much shorter the preface is only one sentence the preface was the purpose of this volume is to raise more questions than it answers maybe you could give us a few examples of those questions well they weren't all questions there was one anonymous idea which was due to Donald was that half that half-baked ideas from people are better than ideas from half-baked people and we're back to baking you remember that Don yes I do that I'd like to ask coming out again of your World War 2 experience whether this experience which was kept secret could have had an effect on the development of computers and if so why and how I asked you that jack but I wonder whether Donald has some kind of response to that as Jack has mentioned max Newman provided a pretty effective transmission belt between the concepts which were normal in the wartime environment and the post-war development of computers in Britain which has often puzzled outside observers as having got off the ground with a mysterious promptness and speed Americans have sometimes commented on this but there isn't now any mystery at all I think because of this direct transmission and Max took very active steps immediately to get I think about 40,000 pounds which in the money of that day was quite a lot of money to make a start to recruit people including Tom Kilburn who hadn't been a Bletchley person but also people like David Rees Jack good Alan Turing who had been alleged so I think that the impact was pretty direct and how did this compare with the American experience because the Americans were using some kind of code breaking machines as well I think a simpler kind perhaps than the Colossus nevertheless they were in this to a certain extent and didn't they begin to work with the ENIAC and other machines like this for calculating trajectories and so forth so why would the British so far ahead in those early years well from the standpoint of history of computing and computers the ENIAC machine of 1946 of the Moore School at Philadelphia would stand on a par with the Colossus machines of 1944 only two years between them and obviously with great relative strengths and weaknesses if you stack him up off against each other but both were high speed electronic computers the fact that one of them was under secrecy for another few decades means that the history books are only now beginning to be corrected as to which ISM was historically the first but in a longer perspective neither of them could claim in the full sense to be a stored-program digital computer in the sense of that's demanded by the theory of the universal Turing machine capable of being programmed compute any computable function I think most historians would say that in an empirical style Babbage's creation could be said to fit that logical specification but neither Colossus nor ENIAC meet that spec but they are of course both of them very important stepping stones what what would you call the first true computer then you just call these two electronic computers but then you're sort of I got the impression you sort of backtracked and I wonder how you would classify them as true computers or if not what would the real first true computer deep well they're certainly true computers but they're not either of them Universal machines which is a somewhat technical concept I would defer to others here generally for example but the first such machine in my understanding in the modern era is the edvac with an American machine which followed on that sink went operational a year or two later and certainly by 1950 there were three machines operational in Britain at least that met that super demanding spec of which the very first would have been probably the Ferranti mark 1 but then pilot ice and the EDSAC in Cambridge and what and what on what do you base that definition the quicker the shorter answer is whether the machine can be given a program which will cause it to compute any mathematical function whatsoever essentially a period in the definition whereas the Colossus or the ENIAC could only be set to solve problems which fell into certain defined logical categories I see a universal machine as was proved in a purely formal mathematical sense in Turing into 6 paper is a machine which given sufficient time given sufficient resources there is nothing there no purely symbolically describable task that it cannot compute announcer - I see and where do we go from these machines to artificial intelligence what's the link there in technical or human terms humans in human terms the outstanding link was Alan Turing himself I believe that it was during that infected both Jack and me and during the war I was something of his own obsessional interest mm-hmm in the use of these machines not for the immediate numerical computation or cryptographic computation that might occur to people but in simulating and outstripping the capabilities of human thought reason knowledge learning all those aptitudes and during saw that very clearly and communicated it extremely clearly in numerous discussions over those years with a small interest group that tended to be magnetized around him mm-hmm Jack and myself I would say in particular and where we how would you characterize the technical term the technical aspect of that that is quite a complex question because it hangs on the definition of intelligence whether we are speaking of the kind of thought processes which are officially recognized by professional intellectuals as being kosher thinking which have in common that they can be formulated in these symbolic styles if we restrict it to that part of human mental life then the connection is absolutely direct because if you have a form of machine which can compute any specifiable function that can be specified symbolically and if we confine ourselves to those aspects of thought which can be specified in principle in such a logical language then it follows that we have the means that in principle can implement old human thought and I think that's the kind of rough caricature of the reasoning that lies below the surface of Alan Turing's 1950 paper The Fairly 50 or rather famous paper in which he proposed the Turing test what we might get on to later is what one says about the discoveries of more recent time that the part of human cognition of thought and knowledge which is intellectually respectable in the sense that I've described is only that tiny bit of the iceberg above the surface and that most of the brains work in solving problems goes on below the level of introspection and could be called not cognition but sub common which there are two extremely important subdivisions of this submerged continent one being to do with highly trained skills whether playing tennis or solving different equations doesn't matter and the other huge submerge continent is to do with forms of visual creative visual thought which are very important to mathematicians and creative thinkers generally spatial visual intuition about which in general can only with greatest difficulty if at all be translated into ordinary language and symbolic forms so that there is a big question which perhaps we won't get into now about how that side is being can be tackled by digital computers well is that I know you don't want to get into it but I if I may just ask would chess playing machines be an example of that in view of the spatial relationships amongst the pieces that's a very great deal of evidence that it is the most massive evidence is a large monograph by Adrian de grote which was his PhD thesis in 1944 but was subsequently published as a book called thought and choice in chess and there he studies chess cognition at every level from patzers all the way up to grandmasters and former world champions and certainly he establishes that chess at the top level is a very intricate tapestry which is woven from strands of logical analysis of the kind that Alan Turing was very explicitly aware and other strands which are of these rather more mysterious intuitive and spatial reasoning that you're referring to I'm curious to know about the transition that you made after the war from your work in biology and I believe it was genetics from from that work into what we now call AI or artificial intelligence how did that happen it happened partly because I'd never forgotten the wartime experience and regarded this is something to be doing until I could get back to what we now called AI am partly because I had an argument with a fellow biologist in a pub about whether it was possible for a machine to learn and rather than spend words on it I spent a lot of time at home in the evenings building a machine out of match boxes and glass beads which taught himself by trial and error to learn to play the game of tic-tac-toe and eventually learn to play excellent perfect tic-tac-toe this called the fancy of a visiting professor from Stanford who was doing a stint in Europe on behalf the Office of Naval Research the upshot was I spent a summer in Stanford University in 61 and we're so bowled over by the possibilities at last after all these years of doing serious experimental work experimental programming and tackling machine learning problems there when I came back to the United Kingdom I essentially walked out on the medical faculty who employed me with a few helpers and students we got some money from the government and found some disused buildings and we host our own experimental programming unit flag and started to do machine intelligence work it was another two years before we got official recognition from the university but behind the scenes the vice-chancellor so Edward Appleton who you may know of because of the optimal air very great physicist was really protecting and aiding-and-abetting us he was very insightful and kind about the future prospects so that's how a small group got going in the early sixties in the early sixties yes was it at that time that you also had developed an interest in chess and speculated on whether or not a machine could beat a human chess player well there are two episodes there one during the war a lot of my discussions with Turing took place once a week when we used to play chess in the pub Wolverton and the whole idea of automating game playing as a testing ground for artificial intelligence in the small circle was very central incidentally that was basically how I got to know tearing so well in that place either there were chess masters who had been hard for the code-breaking where because they were chess masters that was one of the categories for the recruiters or else typically they couldn't play chess at all whereas Turing and I were about the only two who came in the in-between category of playing very bad yes so we gave each other an equal game but then very much later after the war he and David champion and Cambridge destitution developed the first chess machine which was really a paper machine almost like my matchbox contraption it had to be operated and scored by hand and Sean Wiley former Bletchley colleague and I developed a challenger for it and the two of us tried to play the two machines off against each other by correspondence chess and tearing was responsible for calculating and mailing the moves for their side and I was from our side and both of us were very inefficient and forgetful and somehow the game didn't get on fast and he tried to play the two against each other by programming the Manchester mark one machine which had then become operational he had difficulties with Tom Kilburn the great british pioneer who max Newman had appointed in Manchester and took to trying to program the machine at night and sleep during the day for one thing another project never finished and a few years later he died as you know that you may be thinking about the bet that I and John McCarthy Seymour Papert and one other laid in 1970 in 1968 I think it was in Edinburgh at a party after a machine intelligence workshop against international master David Levy who wagered that no machine would beat him in the next ten years he won his bet stirred up a great deal of work and had a beneficial impact I think and ten years after that David finally went down against more than ten years against deep thought which is the current world champion chess machine as you know right and which is playing not a fairly strong Grandmaster level okay I guess going back into the the division or Institute you referred to earlier was it there that you began to work in robotics yes we developed into quite a large operation and by 1973 there were about 70 I suppose in the whole school of artificial intelligence and I made robotics one of the central projects for very similar reasons to those which led John McCarthy to set up a robotics group at Stanford and Charlie Rose and Niels Nielsen and others at s RI International and Marvin Minsky at MIT namely that the great thing about solving problems in the real world like how am I going to get that light blue book over here without upsetting the dark green or dark blue book and in a way it's kind of trivial but how do I work out that I better study it with one hand get this out with the other then I can put that down gently put this over here now the world is very the real world is very unforgiving there's no cheating and if a strategy which looks superficially correct actually doesn't work because you've taken left out of a calendar there's a high polish on that surface or something of that kind a simulation program is not properly tested it's part of the nature of intelligence as most people understand it to have a grasp of how the world in which you do problem-solving actually works what are the causal structure how can you exploit that structure to form plans by practicing those plans they can become automatic and so forth problems which quite small children solve and if you go out into the robotics industry today you'll find if that if tackled in the straight down the middle non-ai style those problems are very unforgiving very hard they can really only be beaten in the long run by genuine intelligence so it seemed a fair and adverse test do you believe or do you think that if a robot can be taught to pick up the book the proper sequence that you were demonstrating or blocks or anything else and that it can learn that but then we can make the leap to say that it thinks what is the what is the connection in an artificial intelligence between learning and thinking and regulating on that the words of that sort like think and intelligence are useful for one particular aspect of shall I say smart behavior after all a dog can exhibit smart behavior and maybe even solve that same problem effectively what the dog cannot do is to communicate to you in terms that you can understand some retrospective justification or reconstruction of how the dog did it or why the dog did it and we can and it's precisely that extra facility to explain ourselves to each other explain our strategies that artificial intelligence has as a wild card in the pack and the reason why I call it a wild card is because although we humans can do that quite well some of the time an awful lot of our goal seeking and quite smart activity is a bit of a mystery to us and if we explain we give a very patchy and inadequate explanation particularly if you're talking paradoxically to somebody who's operating at a very high level of trained skill and imagination he may have the greatest difficulty now this is where I think it's a mistake to see the goals of artificial intelligence as merely reproducing the properties of human thought we and problem-solving we are entitled to aim at something rather better than that something that I've called super articulate see that is to say how nice to have assistant machines intelligent assistants that can make a better job of explaining to us why they're doing what they're doing what their current goal is then the human can particularly nice in safety critical areas like what's going on in a power station that's computer controlled or flying an aircraft so you're making a distinction between kinds of intelligence and there could be a machine a kind of intelligence that is machine intelligence like the name of your book isn't that right and human intelligence and they may not be the same they they may have different characteristics and some like different more appropriate for others different profiles people though there's a certain common area which the machine intelligence engineers or knowledge engineers absolutely must master the the it's pointless to have a machine that is simply solving the problem very cleverly and very fast unless it is doing it in a sufficiently human-like style to be able to maintain rapport with the humans that it's assisting and to give human-like intelligible explanations some there has been some feeling in the computer science community and I I suppose even amongst laypeople who are who watch field or of artificial intelligence that that it was very at the goals of the field were very ambitious and that they that there's been some disappointment do you have any thoughts about if you if you feel that that AI has gone astray in some ways do you do you think you could suggest by well that might have happened or what direction perhaps the field should go in well I'd like to separate that into the two halves wanting to do with the disappointment and the other should do with current directions the disappointment is the disappointment felt by onlookers who have goals in their mind which they suppose that the scientists are attempting to achieve if you measure the field off against its own goals I don't think that that sense is valid the there's a distinction between the very highly colored images in some of the science press as opposed to the science journals and the scientists can hardly be blamed if they don't achieve those wonderful things because that wasn't what they were trying to do on the subject of the field being in danger of losing its way I think that this is a very real danger the pioneers of the field developed two massive struts at opposite ends you might say on two different sides of a river one being on the symbolic logic side people like John McCarthy J Robinson nils nilsen a tremendous background and basis for mechanizing the purely logical ultra rational types of symbolic thought which ordinarily are only used by humans in discourse among trained intellectuals does not say it's unimportant but that's the part of thought which was focused on and another stream of which Neil and Simon would be difficult and to some extent Feigenbaum who was influenced by them taking on board was known about cognitive psychology and how the brain really works so this is a sort of symbolic school and the cognitive neural type it's clear today that neither of these two struts of the bridge are going to carry any traffic until we have a trained profession corresponding to the bridge itself the mid piece that combined the necessary components and types of knowledge experimental discipline as well as theoretical approach from the logical side with the cognitive side of those I mean which do you find that you most naturally allow yourself with I think it's so essential now to make a an economical blend to create a new scientific technology that I find the question almost difficult to follow I've been with a number of colleagues mainly in Europe and not entirely preceding in a very low-profile way for the last 10 years we formed a consortium of about half a dozen laboratories this is not generally known it's not in the public domain we circulate our end transactions and we typically meet once a year because it became clear to all of us that by a particular blend of tools and approaches taken from these two different traditions it was possible in a fairly immediate timescale to demonstrate a new phenomenon are very very difficult to predict where it will lead namely the automatic synthesis from scratch of brand-new human knowledge human type knowledge I should say human type knowledge I mean something which the professionals in the given field which might be in Pharmaceuticals or in cardiology or whatever could read understand apply pass on to their students but which was entirely synthesized by machine which sounds a little bit science fiction or might have sounded like that ten years ago well the good news is that it's now fully established we have three massive worked examples we've used outside judges to come in and help us assess before we certificate it any demonstration as being acceptable and logically I think we pretty well could just as well wind up now because it's been proved and some of us are busy now to try and capitalize on this in ways which would be of commercial or rather applied benefit but the key moves which are not to be found in the two original struts of the field one is the creation using a branch of logic programming of complete logical models or simulations of how the particular domain works secondly the use of such a model to generate exhaustive lookup dictionaries of basic facts of how to do it like how to interpret electrocardiograms and the final stage and this is key is the use of modern machine learning methods to compress this huge encyclopedia of low-level facts which are indigestion you know completing correct into compact reasonably you user-friendly and what I would call knowledge expressions or rules and that final product is what I mean by automatic synthesis of knowledge from initial logical specifications now that's been done three times now once in the chest end game once in the neck in electrocardiogram once in troubleshooting electronics on a satellite which was done as a strict commercial job so I believe that the doors open now and the next step is how do we train people to be knowledge synthesis specialists mm-hmm but if this is not picked up then it a difficulty about artificial intelligence is that it hasn't otherwise yet defined itself and people looking at it from the outside with the cool high of a physicist or a chemist or a traditional computer scientist are justified in saying who are these people they talk about knowledge they haven't got any logical or numerical way of defining and measuring it would I listen to a mechanical engineer who couldn't define work or mass or acceleration but the point about the knowledge since this discipline is to make it work you have to define all these constituent parts how do you regard the expert systems of today that have become the commercial side of AI are they a small step along the way to your idea of automatic synthesis of knowledge yes the first step which was rules obtained by a dialogue acquisition from experts which is a marginal technology in commercial terms for the reasons I mentioned that in so many domains the experts are not to articulate as a source of rules the next step from that is to don't ask the expert any questions just observe what he does and get him to give examples of his decisions and machine his machine learning to convert automatically that material into rules and that is much more cost-effective by a factor of more than 10 and is a large part of the revenue that my own Institute actually pays its rent with the third step is where you don't have the expert who can give you rules at all which I've just been describing which we've been using up 10 years in the margins to try and validate as a possibility and the answer is yes it can be done if you don't have an expert even who can show you how he does it never mind you can build a logical simulation that will substitute build up a huge dictionary for each case how to do it for each question what's the answer and then out of that extract a logical theory which I've described as human type knowledge mm-hmm so that's definitely on the road now and I believe by the turn of the century will be a major industry in its own right mm-hmm you mentioned your current Institute that's the turing institute what is the Mandate of the turing institute the Mandate of the Institute is to work in all or any of the fields cheering himself contributed to to test the theory that the technology is self viable that is that a laboratory where the scientists own all the only laboratory can keep alive from yet here by supplying artificial intelligence services and and products to industry and government but we mainly sell outside the United Kingdom America Europe and Japan mm-hmm so but I imagine that some of the work that you outlined will be going on there on automatic synthesis that's true in fact the contract done for the European Space Agency was exactly a case in point and that is now operational it is under operational tests well we have to wrap up so I just thought I would ask you for maybe a statement a broad statement about what you think the greatest contributions have been on the part of artificial intelligence to computer science I think that the biggest one of all is one that is awareness is spreading very rapidly that is a recent that it will be another few years before everybody's heard of it and it's called inductive logic programming and it would take more time than I suspect we have a way to plunge into it but inductive logic programming of whom the number one exponent now is Steven Michelsen who very shortly will be moving from Glasgow to join Tony Hoare in Oxford so coming soon inductive logic programming I think so yes absolutely central thank you for your time

Channel: Computer History Museum

Views: 80,409

Keywords: Computer, History, Museum, Bletchley, Park, Enigma, Colossus, cryptanalysis, WWII, Max, Newman, Alan, Turing, Tommy, Flowers

Id: 6p3mhkNgRXs

Length: 119min 15sec (7155 seconds)

Published: Fri Jun 22 2012