John Mccarthy: Well I Played Go up to the Level of First Q, Then After Some Years I Demoted
John McCarthy: Well I played Go up to the level of first Q, then after some years I demoted myself to second Q and now I think I've demoted myself to third Q.
Q: Yeah you've gotta keep up with it, but I remember they had all these various <inaudible> memorized, I couldn't be bothered to do all that.
John McCarthy: I'm afraid I discouraged my son by giving in recently when he wanted to start learning Go by giving him a 9 stone handicap. But there are a lot of levels in chess too.
Q: One of the things that Gasparof said I forget it was in the match or the rematch was that quantity had become quality; can you share your thoughts on that?
John McCarthy: Well of course that phrase is a bit of Marxist jargon, transformation of quantity into quality.
Q: In the sense of the Soviet Union producing multiples of something?
John McCarthy: No it goes all the way back to Marx that is the philosophy of dialectical materialism and of the various things or principles of dialectics that's the most comprehensible. The other ones are totally mysterious, they only give examples like you'll increase the temperature of water and then all of a sudden it boils and then they wanted to apply it to politics, you oppress the working class and eventually it revolts. That was the-- but anyway so precisely what did Gasparof say?
Q: He said quantity had become quality so I guess if you threw enough computing power at something it actually became a good chess player, it seemed to.
John McCarthy: Ok well I don't have any special comment on that comment because that principal is only compared with illustrations of it, it doesn't predict anything in particular for example.
Q: What's your prediction for chess, you said that, you know, Deep Blue was a lot of brute force but there were still <inaudible> do you know whether these heuristics are just things like evaluation of board position to some fancier ways of proving a game tree or has this basically now run it's course that we've got the best evaluation functions we're gonna get so we found all ways of pruning the game tree and things like that?
John McCarthy: No we certainly haven't found all ways of pruning the game tree, to give an example.
Q: <Inaudible>.
John McCarthy: Well Deep Blue will consider a move say pawn to Kings brook three, many millions of times and almost all the time it will reject it as not worth following up and for the same reason so a human chess player will reject the move and then revive it only under specific conditions. Now are you familiar at all with the alpha beta, you're familiar with the killer heuristic?
Q: No.
John McCarthy: Killer Heuristic is my invention but I think other people have reinvented it and used it, it's a way of making alpha beta work better and what it consists of is let's suppose that you consider a certain move and then you discover or the program discovers that in that position the opponent can catch-- take your queen. And let us to make matters really simple that your move had just simply left the queen on priest, then now suppose you consider another move for yourself then what you should do immediately is check whether in this new position your queen is on priest. Rather than discover it, ok so that's the killer heuristics says that if you find a good move, either for yourself or your opponent and some place up under the tree, put it on the killer list. Now if it is not a good move all you waste is a little time, you examine it first rather than later or you examine the moves on the killer list. I believe Deep Blue uses it and I think the, well I'm not sure ______didn't mention it, Cotalk chess program uses it, I suggested it to them.
Q: He didn't mention it to us, we did an oral interview with him and we also sent him a bunch of stuff.
John McCarthy: Ok but now consider bad moves, you discover a move that is good in one position and now looking at the thing mathematically if you look at the game tree just as an abstract tree there's no identification of two moves in the same position, and in different positions I mean but in real chess there is-- the tree is not an abstract tree, it's a tree in which the edges are marked. Now consider a bad move like ______pawn to rooks 3, then if you put it on a bad move list then you're risking more than time, you're risking not making this move at in some position in which it is important. Now the only program that I know of that actually used the bad move heuristic with a way of reviving a move, for example if you put pawn to rooks 3 on your bad move list then an enemy piece on your knights 4 would revive this pawn to rooks 3 and furthermore it would revive it not merely in the position where the piece was actually there but in previous positions because it could prevent the knight from going there. So the only program that used it was the stronger version of the Russian program.
Q: That's the one that you played 7094?
John McCarthy: 7090.
Q: 7090. The ITEP program?
John McCarthy: Yeah I've got it here but wait a minute, this was a previous program, it wasn't Kaiser, Kaiser was their later program and I think maybe that Kaiser didn't use it because it's risky. If your criteria for reviving a move are inadequate then you can fail to revive a move when it is important.
Q: I saw a group, Addleson, Velski, Arlisoff, Donscoy and others?
John McCarthy: What was the last one?
Q: Donscoy.
John McCarthy: Yeah ok now Donscoy was not part of the original group that played against us but he was part and even one of the leaders of the group that developed Kaiser.
Q: That's a good point because I've got him down here as being part of that group <inaudible>.
John McCarthy: No but the leader of the group that played us was Kronrod, now I can tell you a little story, Kronrod stuck his neck out well he stuck his neck out in two ways, one at ITEP, once they got going playing with the Americans then they got to use whatever computer time was required to the annoyance of the physicist but then Kronrod a little later did something else, namely he signed a petition against putting the mathematician ______in a psychiatric institution.
Q: That was dangerous in those days?
John McCarthy: Oh boy was it dangerous and according to my Russian informant the party demanded that Kronrod be punished for this but they didn't demand that he be fired. But the physicists were also angry with him about the computer time so they fired him at which point he was unemployed for a while and then he ended up in a job in the patent office which of course caused all his friends to compare him with Einstein.
Q: Yes of course.
John McCarthy: And his friends all quit and went to a different institution, that is in other words his group left ITEP, went to a different institution and of course some of em have immigrated
Q: That's Kronrod?
John McCarthy: Yep.
Q: Do you remember what his first name was?
John McCarthy: Alexander, Alexander Zergavich if I remember correctly.
Q: Ok that's one.
John McCarthy: To tell another irrelevant story after he had been canned and was working at the patent office, his friends thought it would be a good idea if this well known foreign scientist namely me were to want to talk to him. So it was the most Russian I ever had to speak, I had to make 17 phone calls in order-- because they didn't want to arrange it themselves because they-- it would be more effective if I did it. So I eventually got to meet with him, was several people sitting around to make sure that he didn't reveal any secrets.
Q: You actually went over there?
John McCarthy: No I was in the Soviet Union at the time; I didn't make a special trip to see him.
Q: Now the way we work in the museum is to try and get some idea about the whole area, which none of us know about of course other than, you know, I know David Levy 'cause we were graduate students together in Glasgow. But that's the sum total I know about chess and then we start hacking around, we write little two page items to ourselves and then we start talking to people like yourself and Cotalk and Greenblat and anybody else we can get our hands. Then we have to boil this down to almost zero text.
John McCarthy: Well I understand that but at least the almost zero shouldn't be actually wrong.
Q: That's right and that's why basically we're here, you know, the first thing you commented on was that Claude Shannon, what did I say, conjectured about the number of moves in a game of chess, we've changed that to estimated because we actually looked at his paper and that's the word he used. I think he said 10 to the 47, which is a good, you know, estimate. I wonder if I can get you to look over some of this and just a quick read to see if you can find anything obvious.
John McCarthy: Sure.
Q: Actually that's the one I'm keeping notes on; let me give you that one. And of course like any other historian we find people don't remember things correctly, Cotalk certainly didn't remember things correctly about what he had done and what other people had done.
John McCarthy: I thought the email he sent you was correct about what he had done; there was one mistake in it.
Q: About the size of the chess boards in previous games?
John McCarthy: Yeah. The Los Alamos one was the only one that was 6 x 6; the other predecessors were the full game.
Q: We certainly checked that out as soon as I soon as I saw that and we corrected that.
John McCarthy: Have you spelled wiener correctly?
Q: Perhaps not.
John McCarthy: I think it's wie.
Q: Yeah good point.
John McCarthy: Now I don't think that-- I might be wrong but I don't think wiener did anything in that direction, not even as much as Shannon or even a single article or anything.
Q: He just wrote about this in his cybernetics, he mentions chess that's all; we never actually <inaudible>.
John McCarthy: I wouldn't put him on the same level with regard to chess as Tiering and Shannon.
Q: No certainly not.
John McCarthy: And I don't know about De Souza.
Q: De Souza actually wrote a bit about it, yeah he never actually did anything but he wrote about it.
John McCarthy: Ok now Tiering only hand simulated his program and Shannon only wrote a theoretical article. You're not willing to say 10 to the 47?
Q: Well what we do is we get every different kind of person coming through that <inaudible>.
John McCarthy: Yeah but you could do both.
Q: Oh I suppose we could except we're limited in this thing really to 175 words and I'm trying to keep it down.
John McCarthy: Ok that only gives you one more word and a lot of people.
Q: A lot of people <inaudible>.
John McCarthy: Will understand 10 to the 47 a lot better than they understand.
Q: <Inaudible> I'll certainly put that down here.
John McCarthy: Now this statement that the computers had no near the storage capacity or speed required to play anything except the subset, I think that's a mistake. They had the storage capacity and speed to play; they would have played very slowly.
Q: And a limited game too.
John McCarthy: No they didn't, I think the Los Alamos program used a limited game more for programming convenience then for.
Q: One of their write ups said if they've actually included the bishops it would have been three hours between moves rather than 12 minutes or something.
John McCarthy: Uh huh, ok, alright, now maybe you wanna say didn't have the speed to play it at a reasonable rate because-- no the other thing of course is that at the time that program was written that was Los Alamos's main computer and spending 3 hours making a move in a chess game was not something reasonable and I think the ITEP people were quite annoyed because I suspect that some of the moves that they played in their match against us took 3 hours of computer time. I don't know that for a fact though but anyway.
Q: Rather than saying accept a subset of a chess game, perhaps some comment could only play in a rudimentary fashion or play very simple chess or something.
John McCarthy: Well could only examine a very small number of possibilities in a reasonable time.
Q: Ok <inaudible>.
John McCarthy: To be even more specific than that then the amount of time they were willing to allow for it to devise a move. You know the 1957 you're referring I suppose to Alex Bernstein?
Q: It's either that or Cotalk and have to see what one.
John McCarthy: Cotalk wasn't then.
Q: It must have been the first one.
John McCarthy: There's a story about Bernstein about which I don't know the whole thing but anyway when he came to the Dartmouth Conference in 1956 then he was already working on his chess program but he was an employee of IBM and in 1959 IBM had a purge of artificial intelligence, there was some guy in their lab an engineer who was against it who had high position but the basic reason apparently was they felt it would frighten their customers.
Q: Right this humanoid thing.
John McCarthy: Yeah and I don't recall exactly when it was but at some point IBM switched saying no we don't make computers, we make data processing machines.
Q: Yeah 'cause computers were getting a bad name, that's right.
John McCarthy: Well I think it wasn't so much a bad name is that from IBM's point of view to persuade someone to pay a large rental for a data processing machine was a more straightforward thing than to get a business corporation to pay a large sum of money to rent a computer.
Q: Ah yes that's possible too. I've heard there's a perceptual issue with putting people out of work, computers <inaudible>.
John McCarthy: Well that's of course always been with any kind of technology but after all a company that starts paying what was a very large sum in those days, several thousand dollars a month for a data processing machine is expecting to replace a lot of human labor. So it certainly doesn't frighten the people who were buying it, it might frighten various deep thinkers who would think they ought not.
Q: But they weren't customers. Back in that 68 ITEP in Edinburgh, it was maybe a year before that, that I had to try and automate the Glasgow University Registrar's office and they were still using ledgers in which they would write the student's name and the courses they taken, like being long gaps between em and I had to try and bring them to the computer age. Now I remember the first time I walked up there to have a look, the woman at the desk said yes, I said I was here to see Mr. McCargo the registrar and she looked up and so oh you're the one who's going to put us all out of work. I remember thinking at the time well a hell of a lot of cooperation I'm gonna get here, you know.
John McCarthy: I can tell an interesting story that's relevant to that, one the first computers was the PDP1, which was bought by <inaudible> Newman, they bought the prototype and then they bought one of the first ones. And Ed Fredkin, have you heard of him, who worked for them at the time said well maybe we can put our business computing on this machine. So we had an interview with them, you know, with the business, with accounting people, and they said well it has to do this and it has to do that and it has to do this other thing and eventually Fredkin was discouraged. And then in came the IBM salesman and the IBM salesman said you have to change this and you have to change that and you have to change this other thing and Fredkin was astonished on how this salesman was able to bully them into changing their procedures in order to fit what IBM was selling and it wasn't even a computer at the time, it was the accounting machines.
Q: And they probably did change it for him too.
John McCarthy: I guess it was after all for a long time when I was a consultant for IBM, you know, people would say to me well, you know, this computer stuff, all very well but IBM is still getting most of it's revenue from the accounting machines. I don't know when they changed but maybe along when the 360 came along is when the-- when the 1401 came along all of a sudden, all those accounting machines became obsolete.
Q: Gonna flip back to Bernstein for a minute, do you know what happened to him?
John McCarthy: No I don't. I'm sure that he continued to work for IBM a while after 56 but I don't know what happened to him.
Q: Ok thank you.
John McCarthy: And I don't recall that he ever wrote a paper about it, which maybe-- so I don't know whether the program was completed but I suspect it was completed enough to play.