Richard Greenblatt，Lisp Machine的发明人。
Richard Greenblatt and June Malone黑客曾是一种荣耀，一种美好的传统。它代表着60—70年代反权威却奉公守法的计算机英雄。这群电气工程师和计算机革新者，才华横溢，行为孤僻，也有个性缺陷。他们沉湎于技术和计算机，视工作为一种艺术。Steven Levy的《黑客：计算机革命的英雄》(1984)一书将我们带回了那段令人怀念的美好岁月。本书叙述了50年代后期至80年代中期的黑客史，尤其详述了黑客发祥地-----麻省理工学院的“原型火车俱乐部”(Model Railroad Club)成员的活动情况。他们早期秘密穿梭于笨拙的穿孔大型机，并从Xerox PARC偷取大量技术，开启了计算机革命的历程，促成了PC的诞生，使计算机真正走向大众。这群人中有Richard Greenblatt、Bill Gasper、Captain Crunch、Steve Woz甚至Bill Gates。他们自由不羁，反抗既有体制，却严格遵循黑客道德准则：自由使用，信息免费，打破权威，推动分权。他们认为计算机既可创造艺术与美，也能使生活更美好。Levy追溯了黑客真正的历史传统和严谨祖训，使本书为后来的许多黑客带来了积极的影响。
Richard D. Greenblatt is an American programmer. He was born in Portland, Oregon. His family moved to Philadelphia, Pennsylvania when he was a child. He later moved to Columbia, Missouri with his mother and sister when his parents divorced.[1] Along with Bill Gosper, he may be considered to have founded the hacker community, and holds a place of distinction in the Lisp and the MIT AI Lab communities.
Becoming a hacker
Greenblatt enrolled in MIT in the fall of 1962, and around his second term as an undergraduate student, he found his way to MIT's famous Tech Model Railroad Club. At that time, Peter Samson had written a program in Fortran to automate the tedious business of writing the intricate timetables for the Railroad Club's vast model train layout; for some reason, Greenblatt felt compelled to re-implement the program on the PDP-1. This feat of necessity led him to the AI Lab, where he proceeded to become a "hacker's hacker" noted for his programming acumen as described in Steven Levy's Hackers: Heroes of the Computer Revolution, and as acknowledged by Gerald Jay Sussman and Harold Abelson when they said they were fortunate to have been apprentice programmers at the feet of Bill Gosper and Richard Greenblatt
Richard GreenblattIndeed, he spent so much time programming the PDP machines there he failed out of MIT as a sophomore and had to take a job at the Charles Adams Associates firm until the AI Lab re-hired him.
He was the main implementor of Maclisp on the PDP-6. He wrote MacHack, in response to the claim by AI skeptic Hubert Dreyfus that computers would not be able to play chess. Dreyfus was beaten by the program, marking the beginning of computer chess. He also wrote, with Tom Knight and Stewart Nelson, the Incompatible Timesharing System, a highly influential timesharing operating system for the PDP-6 and PDP-10 used at MIT.
Later, he was the main designer of the MIT Lisp machine along with Tom Knight. He founded Lisp Machines, Inc. (which later became Gigamos Systems), according to his vision of an ideal hacker-friendly computer company, as opposed to the more commercial ideals of Symbolics.
Linux文本编辑器 Emacs
Richard GreenblattEmacs是一种强大的文本编辑器，在程序员和其他以技术工作为主的计算机用户中广受欢迎。EMACS，即Editor MACroS（宏编辑器）的缩写，最初由Richard Stallman于1975年在MIT协同Guy Steele共同完成。这一创意的灵感来源于TECMAC和TMACS，它们是由Guy Steele、Dave Moon、Richard Greenblatt、Charles Frankston等人编写的宏文本编辑器。自诞生以来，Emacs演化出了众多分支，其中使用最广泛的两种分别是：1984年由Richard Stallman发起并由他维护至今的GNU Emacs，以及1991年发起的XEmacs。XEmacs是GNU Emacs的分支，至今仍保持着相当的兼容性。它们都使用了Emacs Lisp这种有着极强扩展性的编程语言，从而实现了包括编程、编译乃至网络浏览等等功能的扩展。
Christopher Strachev在伦敦编写出了第一个西洋跳棋程序。1952年莫里斯·威尔克斯(Maurice Wilkes)和剑桥的科学家小组编写了能玩tic—tac—toe游戏的程序——EDSAC。1956年IBM的亚瑟·塞缪尔(Arthur Samue1)写出了第一个能够“学习”的西洋跳棋程序。1962年美国麻省理工学院的科学家阿兰·科托克(AIan Kotok)编写了一个能够击败国际象棋业余爱好者的程序。1963年亚瑟·塞缪尔的西洋跳棋程序击败了关i国康涅狄格州的跳棋大师罗伯特·尼尔利(Robert Nealey) 1966年麻省理工学院的理查德·格林布赖特(Richard Greenblatt)开发出了MacHack VI型机，它成为第一台有实力参与人类主办的国际象棋锦标赛的计算机。1970年美国威斯康星一麦迪逊大学的艾伯特·左瑞斯特(Albert Zorist)编写了第一个会玩围棋的程序1977年由美......
Richard Greenblatt口述历史编辑本段回目录
Oral History of Richard Greenblatt
Interviewed by: Gardner Hendrie
Recorded: January 12, 2005
Boston, MA
Total Running Time: 2:33:00
CHM Reference number: X3131.2005
?2005 Computer History Museum
Q: We have today with us Richard Greenblatt who has graciously agreed to do an oral history for the computer history museum. Tell us a little bit about your family background, your parents, where you were born and brought up.
Richard Greenblatt: Well, I was born in Portland, Oregon, but- and my family then moved to Philadelphia, Pennsylvania, where- for a couple years but I actually grew up almost entirely in Columbia, Missouri, which is a university town halfway between St. Louis and Kansas City. It抯-- At the time it was about 30,000 people or so and the University of Missouri had 15 or so thousand university students, two other universities in town so it抯 much smaller than the University of Missouri but anyway it was a rather ideal place as I look back on it now
Q: What did your parents?
Richard Greenblatt: Well, they were divorced. My father was a dentist and had a practice in Philadelphia and my mother, when they divorced she moved back with her family. She had grown up in Columbia, Missouri, her family lived there, so we moved back with my grandparents and we had another house and I grew up there. I spent a lot of time at the University of Missouri. There was a student union there and so when I was very young, in the third grade and stuff, I would go to the university student union at University of Missouri and play chess with the university kids and so forth and basically I was pretty good. I would hold my own with them. It was before the days of ratings and sort of chess organization really penetrating out as far as Columbia, Missouri so I didn抰 have an official rating and so forth. Occasionally, I visited my father and whatever and we would visit the Franklin Chess Club in Philadelphia or something like that so?
Q: Do you remember when you learned to play chess and how that happened?
Richard Greenblatt: Not exactly. My father I抦 sure taught me. He was fairly good. He was-- He had been a chess champion of-- Well, I think it was actually later. I think he had finished second in the city of Portland perhaps or something I think and later I think he won the city title in Eugene or something. He was kind of at that level. He wasn抰 a really serious player but he was pretty good?
Q: And he enjoyed that. That was a vocation of his?
Richard Greenblatt Yeah, and he occasionally visited in Columbia and so he initially took me to the University of Missouri Student Union and I- they would play chess and bridge there so I learned both of those games and I played quite a bit and so forth. So it was a rather unusual bringing up in that way. I was quite good in high school and I won a bunch of awards and did this and that and I probably-- There抯 no use going into too much of that. Anyway, I got admitted to MIT and I arrived at MIT in the fall of ?2.
Q: Did you have any brothers or sisters and what did?
Richard Greenblatt: I have one sister and she had epilepsy and so she lives in a small town in Missouri--
Q: It was just the two of you?
Richard Greenblatt: Just the two of us and really just me in some sense although she was there too but not?
Q: You were obviously playing chess at a very early age. What did you think at that time you might be when you grew up, that you wanted to do when you--
Richard Greenblatt: Well, I was into a number of things. I-- Through the university I met some engineering students, I was involved in ham radio. There was a radio shop in the area and a man ran the radio shop and I hung out at this radio shop and read- he had a lot of literature of various kinds and I read a lot of that. I was-- I worked after school as a Western Union messenger and I was around the Western Union office a certain amount and- but I was sort of science oriented, engineering oriented and mathematically oriented. I was-- They-- There were a few computers, very few computers in Columbia, Missouri in those days. Our high-school mathematics club visited the State Farm Insurance Company which had a regional office, a fairly large regional office, in the town and they had a card thing with a card sorter and trick decks and you could watch the cards make pretty patterns and so forth and then at the University of Missouri in the business school they had a real computer which was a vacuum tube machine and we visited and again the high-school class looked it over. However, I didn抰 actually get involved in computers at all at high school or really know what a computer was. However, I did-- My uncle was a professor of surgery at the University of Missouri and had a very long career there and so forth and so I also was involved in- around the hospital and around the animal labs in the hospital and quite a few different experimental things going on in the hospital I had some exposure to. So anyway--
Q: When you graduated from high school, where did you think of going to college or?
Richard Greenblatt: Well, I applied to two places, MIT and University of Missouri
Q: Why did you pick MIT? Where did that come from--
Richard Greenblatt: Yeah. Well, I-- That抯 a
Q: With the kinds of things you were interested in, you just thought?
Richard Greenblatt: Yeah. It was rather chancy. I could have missed it. Looking back on it, I don抰 know exactly how it would have happened but anyway I- but-- Yeah. So in the fall of ?2 I arrived in Cambridge at MIT and it was something they called Orientation Week which was where you had fraternities and fraternities sent you an incredible amount of material trying to get you to join their fraternity and so forth and so I arrived on campus and got picked up by a fraternity member or something and it basically took me one day to decide I didn抰 want to have anything to do with fraternities at which point I then had
Q: --do some other things a little bit.
Richard Greenblatt: Just-- Well, not much. For example, in those days all the students, freshmen, took exactly the same courses. You had only two small electives and other than that all your courses were completely regimented and in large classes and then recitations, smaller breakouts and one thing or another, and your quizzes were organized. You had a quiz every Friday morning so your week was completely organized. You抎 have a quiz in one subject one week, a quiz in the next subject the next week and so forth, rotating, and they were every Friday and so on. So it was still quite, quite regimented, there were- it wasn抰-, but anyway I did get involved in computers. There was a-- In addition to the PDP-1, there was-- One of the-- One of my roommates was taking a course in civil engineering. There was a 1620 with load-and-go FORTRAN and a Calcomp plotter and I was involved in helping him do with that with load-and-go FORTRAN and then I took a- I took two electives. One of them was called 641. In those days there were only two computer courses in the entirety of MIT. It was two real computer courses plus this other thing in the civil-engineering department but 641 was a freshman elective. One other course called 6251 was kind of a real programming course. That was it?
Q: 641 was about what?
Richard Greenblatt: 641 was a freshman elective and it was computer programming and it was actually quite a good course. It involved-- It was a-- The computer involved was the 7090, the campus-wide-- You programmed it batch programming with punch cards so in the building- the basement of building 26 there was all full of key punches and 407s and various card-handling things. You submitted your debt and you got your problem back and your grades and so forth and so I did that. I also-- There was a good library there. It was-- There was a reading room, it was- and then there was also the PDP-1 and there was a bunch of stuff around- connected with that and then there was the Tech Model Railroad Club which also had an interesting library of- and a lot of stuff and quite a few people who were involved in things in various ways including Alan Kotok[ph?] and a guy named Robert Saunders Raz.[ph?] He抯 still sort of around. You see him at the hacker conventions and so forth. Anyway, so I was in that-- Another person who was a big person in my life in those days in a certain way is a guy named R.L. Nelson. He flunked out of MIT and left MIT at the end of my freshman year but basically he and I, mainly him with me looking over his shoulder, designed a bunch of telephone equipment. The Tech Model Railroad Club had elaborate relay equipment including telephone systems and something called The System which involved connecting the trains to the operating positions through telephone equipment like crossbars and so on. So basically, I designed it and helped design with these guys looking over their shoulders quite a few different pieces of equipment, something called the standard connector which is a Stroger telephone that was switching over. It goes up and goes across with a two-dimensional ___________ switch and then we designed a-- The whole system was replaced. This-- The system at that time was located under the layout, the layout was plaster, the plaster was falling down into the relays and making it unreliable and one thing and another. So there- it was perceived that there needed to be a new system and this R.L. Nelson guy sort of designed the essence of it and that was actually built, it was- and it was built out away from- so it wasn抰 underneath the layout but the cab equipment was still controversial and wasn抰 built and I basically in that summer after my first year and so forth essentially generated a prototype cab equipment and finalized the design and then some other people kind of doubled over to it but Kotok in particular. There was a power structure at the club and that抯 another whole story and I was a little bit of a- I抦 a little bit of a countercultural type I guess you抎 have to say but?
Q: What抯 cab equipment when you say?
Richard Greenblatt: Well, cab equipment is the set of relays that are associated with an operating position. In other words, you have the block equipment which is associated with a piece of track, you have a crossbar switch which is a complicated electromechanical, two-dimensional switch as used in the most advanced forms of mechanical telephone exchanges, and then you have the cab equipment which is associated with the operating position and basically has to detect when it抯 time to advance, when your train is moving forward and sort of the-- Well, it抯 the most elaborate-- There-- In this case, there was only two relays with- or actually, initially there were three relays associated with each block which was a big reduction from the old system that did nine and so forth. So this resulted in a lot of innovative design and optimizing and eliminate this con-
Q: You were doing relay logic design.
Richard Greenblatt: Relay logic design and then so this cab equipment wound up it was more complicated. I guess it had pretty close to-- I think it had eight or nine relays in it per cab and there were five cab positions and so forth. So basically five different people could be running trains on this large layout, they could all- the trains could go wherever they went, the system would connect the train to the operating position and as the train went away- went around it would remain connected to the one operating position. Moments later another train could pass over the same track under the control of someone else and it would be signified by a different connection on the crossbar switch and that would be maintained and so on so it was a rather elaborate system. It-- Another friend of mine in those years, my best friend actually in my freshman year, was a guy named-- No. Actually, no, it was the sophomore year. My-- Okay. So anyway, the summer of my freshman year I went back to Missouri and I- and in those days- at that point time on the PDP-1 was incredibly tight. They-- There wasn抰 any timesharing system, it was just a single-user machine. There was a signup list that would go up on Friday mornings I think it was and basically within a couple hours that signup list would be signed up 24 hours a day for the following week. So basically, to get time on the machine was something that graduate students and a few others did. As far as legitimately signing up for it, I was too low on the totem pole to be able to really sign up for any legitimate time. However, I hung around the machine some and got to know quite a bit and occasionally would have a few minutes when someone didn抰 show up for their time or they left early or one thing or another so but- so then-- However, I didn抰 actually get on the machine to really actually do anything my freshman year really but I did learn quite a lot about it and I started writing a FORTRAN compiler and I just wrote it out on a piece of paper and it was- it抯- never really ran it. So the- one of the people who was there who was more or less the-- I guess the PDP-1-- The faculty adviser at that time was a guy named Jack Dennis, Professor Jack Dennis who has been around- is still around and there was a guy named Peter Deutsch who was the son of an MIT professor and he was in high school at the time and he knew probably more about the computer software than anyone else at the time, a frequent thing of course, of young people
Q: What did you do over the summer? Do you remember?
Richard Greenblatt: Well, I think I mostly just did the things I did in Columbia. I played chess at the student union and stuff like that I guess. I don抰 recall anything too much. However, then when I arrived back the fall of the following year, the place was in a tizzy and basically a guy named Stuart Nelson had shown up and again more or less doing the same thing I did and he showed up for the Orientation Week and after one day stopped that and he was a phone hacker and he knew a lot about phone companies and operators and so forth and MIT had a bunch of things called tie lines in those days and basically a number of them went to places like Lincoln Lab and Haystack Observatory and one thing or another and a number of those places had ways that you could get outside, just simply dial 9 or 1 in some cases or do other things and so basically it wasn抰- it didn抰 take him very long to discover the fact that you could make free phone calls and so forth and of course in those days phone calls were extremely expensive and the idea of making free phone calls was an incredible thing. So anyway, he had also hooked up some kind- attempted to hook up some kind of a dialing device to the PDP-1 and gotten the campus patrol and of course it was basically a major tizzy. So anyway, I became friendly with Stu Nelson, he became my best friend and we were friends the following year and so forth, and he had quite a bit of experience. He-- His father had been on the National Television Standards Committee, the company- the committee that actually defined the color standard that抯 used to this day. His father had then passed away or something, I don抰 know exactly the timing of it, but he had- he grew up in Poughkeepsie, New York, and had quite a bit of experience with the telephone company. He also had some experience with IBM and with certain IBM systems and so forth so?
Richard Greenblatt: So anyway, we came back and I knew more of what was going on. By this time, there began to appear a few more PDP-1s. The AI Project led by Professor Minsky had submitted a proposal to ARPA and they got- the guy who actually submitted this proposal was a member of the model railroad club, a guy named Peter Sampson, also a- an old hacker of some note. Anyway, they- another PDP-1 showed up in building 26 and then not too long after that moved to Project MAC which was in Tech Square and basically Nelson and I, having some experience on the PDP-1, kind of gravitated over there. Two other guys were the actual student employees but they weren抰 too terribly good and one thing and another so basically Stu and I were more with it I guess you would say on some level. Anyway, we sort of gradually started helping various people who were trying to use those machines and this and that and then we wound up getting involved there and becoming student employees. It was before the days of UROP but it was more or less like that and so Stu and I and another fellow named Jack Holloway did a number of things. The-- Project MAC which was in Tech Square had another 7090 computer that they had gotten and their whole business was to make a timesharing system called CTSS so they had a timesharing system. There were two other timesharing systems that were about contemporaneous and the three of these were just- can be viewed as the first timesharing systems. Of course, that抯 something that was going on in a lot of places and so forth but there was on this PDP-1 at RLE, Dennis and Deutsch were developing a timesharing system for that machine. That was a machine that had 4K of 18-bit memory and it had a fully elaborate timesharing system and a bunch of things again by-- It was a five-microsecond cycle so it- 10 microseconds per instruction in those cases and just incredibly- so we- by today抯 standards but actually we did it every?
Q: --tour de force.
Richard Greenblatt: Yes, and it had a display and all kinds of good things could be done with that display and so forth. So anyway and then there was another fellow named Bill Mann who worked at BBN who was also around the model railroad club and Bill- and BBN had a contract with Mass General Hospital and had a- something called the hospital system and so this was actually sort of a little bit commercial in the- attempt or it could pass for commercial in those days and they had a rather interesting system and I got to know quite a few people at BBN. Many of them had been at MIT shortly before and had left by the time I got there but- recently but a guy named Dan Barbro,[ph?] another guy named Dan Murphy, for example. Anyway, then the-- Then I-- After my second year?
Q: Had you done anything with chess while you were here particularly?
Richard Greenblatt: No. No, I had not done anything with chess at all and in fact I- although I knew Kotok. Of course, Kotok had done things with chess. However?
Q: When did he graduate?
Richard Greenblatt: He-- It took him a couple years to officially graduate but his normal time to have graduated would have been I think the year before I got there. However, his adviser, thesis adviser, essentially had been McCarthy. McCarthy had left MIT and gone to Stanford and so I guess the way you抎 have to say it is that a lot of the momentum connected with that project had essentially gone with him and there would have- if there had been people around playing chess and doing chess and so forth, I probably- I抦 sure I would have picked it up. However, it just wasn抰 happening. Kotok was out at DEC and nobody knew anything about chess really and so I didn抰 either. I didn抰 do anything with chess. That抯 right. So when I got in- I-- The-- In those days there were two major conferences, computer conferences, each year. There was the fall joint computer conference on the West Coast and the spring joint computer conference on the East Coast and basically as a student employee or something, I don抰 know exactly, the- I had basically got sent to these conferences amazingly enough
Q: Maybe what we ought to do is follow that path and then we can pick up on the conference.
Richard Greenblatt: Oh, okay. Okay. Well, let抯 see. I was so interested--
Q: After your sophomore year. Okay?
Richard Greenblatt: Yeah. When I first came to MIT, I was on the dean抯 list. I-- It turns out that I was very well prepared and amazingly enough?
Q: Even from high school?
Richard Greenblatt: Even from high school in Missouri and so forth and I had to work fairly hard but on the other hand it really wasn抰 that difficult and I was on the dean抯 list. Okay. Well, however, I was getting interested in computers and I was getting less interested in school so my grades just went more or less in a straight line down and my-- So basically, by the end of my second year I was on probation or something, I had-, and I think that year with- if we get the years right, that was when the AI Lab or the predecessor of it, it was part of Project MAC, but anyway they traded in their PDP-1 to DEC for a PDP-6, the new PDP-6, and the state of the software on the PDP-6 of course was quite primitive. So initially, DEC allowed us to keep the PDP-1 even after the PDP-6 was delivered and so we used the PDP-1 to edit on paper tape and so forth like that. It was quite-- Well, there came a day where they said okay, on Friday- they said on Monday DEC is coming and they抮e going to take their PDP-1 and we go ohhh because well, this- the- it- we-- If the-- If they抎 really done that--
Q: --had not all moved to the PDP-6--
Richard Greenblatt: Well, not the system software in particular if we would have had to prepare programs on a typewriter- on a teletype machine which was even harder than a Flexowriter. On the PDP-1 they had something called Flexowriters which was this kind of a mechanical?
Q: A predecessor of the teletype.
Richard Greenblatt: --and so on. Anyway, so basically, given that, Nelson Holloway and I went into crash mode for this weekend and we essentially re-implemented- and of course in those days a lot of the software existing on the PDP-1, the general idea was to implement improved versions for the PDP-6. So there was a program called TIKO that had written by Dan Murphy on the PDP-1. He was associated with the spark chamber group which was another group that had a PDP-1 and we had become users of that. We had made some patches to it and made it so it could use micro tapes somewhat and different things.
Q: TIKO is an operating?
Richard Greenblatt: --editor, an editor, and it-- Well, it was? I can go into some stories about that if we want. I don抰 know how much we want to push down on this. Anyway, we decided we had to write PD- TIKO for the PDP-6 over- on this weekend. So, to make a long story short, we- just as they came on Monday morning to take the PDP-1
Q: With a light pan and a bunch?
Richard Greenblatt: Oh, it had light pan and it also had a- an increment mode and a character generator and some other things that were?
Q: When you moved the software, did you write it in PDP-6 assembly language?
Richard Greenblatt: Yes. We wrote it in PDP-6 assembly language. We wrote it from scratch. We didn抰 consult the-- Actually, there was a separate story with each piece of software but with this TIKO we just started out and just wrote it from scratch. Earlier we had worked on the PDP-6 LISP. Okay. Yeah. Maybe I should go into that because I don抰 think we had this tape running when we discussed that so-- Well, somewhat earlier at the model railroad club there was a time when they were working on the prototype PDP-6 at DEC and so what Kotok-- There was a lot of discussion going on about the instruction set of the PDP-6 and how to do this and that and then the PDP-6 had a- quite a nice instruction set and quite a few different things and such. So anyway, we got into a mode where Kotok would bring a group of us out to the old mill in the evening and we would--
#### End of Tape 102630620_1greenblatt.mp3 ####
Q: Story about that.
Richard Greenblatt: It was great. I just remembered that my car was called the Blue Bird.
Q: O.k. Good.
Richard Greenblatt: A ?4 Chevy convertible. And what I did in those days, it was before there were Chinese restaurants in Cambridge much. So essentially we all got in the car and we came to eat in Chinatown, right here. And, of course, the central artery, we抮e right now looking at the remains of the central artery, which is almost entirely gone. But I spent many, many hours basically battling with traffic which, of course, was no where near as bad then as it is now, but it was still fairly bad. We went to Chinatown. We ate dinner in Chinatown. Then we went back and we hacked, and then at about two or three in the morning, we would very frequently come to Chinatown again and eat dinner.
Q: Get some more food.
Richard Greenblatt: Get some more food. And sometimes we would even do it for lunch. So basically I would make either two or three trips to Chinatown just about every day with a carload of people.
Q: That抯 funny. That抯 great.
Richard Greenblatt: And that was a hassle. And then at the very peak of it, when I was really in my most voracious consumption of food, I would actually eat four Chinese meals a day. By that time there was some Chinese restaurants in Cambridge, so maybe I would only go eat once or twice in Cambridge, and once or twice in Chinatown.
Q: Wow. That抯 funny.
Richard Greenblatt: Anyway, yeah, so we抮e.
Q: Great computer companies, right?
Richard Greenblatt: Right, right. So _____ 39.43 the computer conference of ?5, it was basically the first time I had been to California, and California is a fantastic place, particularly in those days. I mean, the place was a garden spot, and you抎 go into Palo Alto and you抎 drive in on that Oregon expressway and there抯 just plants everywhere and just completely unlike anything I had ever seen. We went out to what was called the D.C. Power Building. The Stanford AI lab was located in the foothills, sort of past the Stanford campus. It抯 on Page Mill and Aristedero [ph?] Road, and it was an idyllic place. I mean, they had a really big building. It was built California style with outside walkways. Well, it was just an amazing place.
Q: O.k. And this is where Minski was now.
Richard Greenblatt: No. McCarthy.
Q: Oh, McCarthy. Yes, I抦 sorry.
Richard Greenblatt: Minski stayed at Massachusetts.
Q: At MIT, yes. I抦 sorry.
Richard Greenblatt: That抯 right. That抯 right. This is where McCarthy went. That抯 right. So that抯 right that I went out there and you抮e on the top of the hill, and you could see the lights of San Francisco and other things. Amazing. O.k. In McCarthy抯 office I pawed through various things. And at that time, McCarthy抯 Kotox program, which had kind of moved with McCarthy to Stanford, was playing the Russian chess program. Well Russia had an interesting story, too. There抯 a guy named Mikhail Botvinnik, and he was the world champion, and he was one of the strongest players ever in the history up to that time. Well, of course, there was a whole thing about chess in Russia. The Russians set this up as a test of their society versus ours. There had been a lead-up into World War II, where Russia had kind of had some good results just before World War II. Then there was World War II, of course, which was a tremendous connection for Russia. Then just after World War II, there was some famous encounters between the U.S. and Russia in chess, and won by the Russians. And the Russians had a group of grand masters that came along, led by this Botvinnik guy, who pretty well decimated the U.S. for many years until the era of Bobby Fisher. So basically Botvinnik had gotten involved in the chess program. And nothing too much came out of that, but they did have another program that somebody in Russia had written, and it played the McCarthy program, the Kotok program.
Q: Now, it ran on a Russian computer?
Richard Greenblatt: Besam 6, Besam 12, or something. That抯 right. It ran on a
Q: Oh, one of the Besam.
Richard Greenblatt: That抯 right.
Q: Machines.
Richard Greenblatt: That抯 right. And so, anyway, I was just perusing various offices, all of which were open and so forth, out there. And I saw some listings connected with this match. And so I set up my chess thing and kind of looked over and I was just horrified by the play.
Q: By the play of both computers.
Richard Greenblatt: Yeah. Well, both computers and particularly the quality of the analysis. Most of this printout was analysis from the Kotok program. And I also saw some kind of a textual thing, which I don抰 believe was Kotok抯 thesis, but which had some of the same information as Kotok抯 thesis. It was probably some kind of a technical report, or something, that was anticipatory to Kotok抯 thesis. Anyway, one of the things I remembered, and which I just talked with Kotok, as a matter of fact, a few days ago, was the detail that they had is Alpha Beta, and so forth, and they had these whips, and the whips were set at 4, 4, 3, 3, 2, 2, 1, 1. In other words, that was how many. It would first look at the top ply. It would look at the four best moves. The next plys, it would look at the three best. Next ply, two best, next ply, one best. Well, I just recognized immediately that that was incredibly wrong. You see, basically looking at only one wide, you just have no signals or noise function. In other words, you look at one move, which you think is the best, but there抯 a tremendous amount of noise. Well, you look at some more moves, and if you find that one of those are better, you抳e effectively rejected some noise. Well, essentially the thing that I knew that they did, they were very weak chess players, both McCarthy and Kotok. And basically they had a very romanticized view of chess. And so I knew, however, that chess is a very, very precise game. And you really- the name of the game is take the other guy抯 pieces, and you don抰 just go along. In any kind of a strong game, you don抰 just lose pieces, win pieces, lose pieces, win pieces. I mean, if you lose even a single pawn without compensation, then you may have drawing chances, if you抮e lucky. Otherwise, the game is lost. Losing more than one pawn almost invariably results in loss of the game, period.
Q: O.k. And these would be things that you, as an experienced chess player.
Richard Greenblatt: Yeah.
Q: At a reasonable level would understand.
Richard Greenblatt: That抯 right.
Q: And have experienced. And have scar tissue about.
Richard Greenblatt: In some sense, that抯 right. Although I did not play in real tournaments, you know. I just played over the board chess with these university kids.
Q: Yeah, o.k.
Richard Greenblatt: And I had not played any chess at all since I抎 been at MIT, so that was at that point probably three, three and a half years, or something. Anyway, I looked at this thing and I could see that the quality of the analysis was not good. And I said, gee, I can do better than that. And so I immediately set to it, and within just a few weeks, after I got back, I had the thing playing chess.
Q: Now, what the thing be?
Richard Greenblatt: The PDP6.
Q: The PDP6.
Richard Greenblatt: Yes.
Q: Now this was you had written your own.
Richard Greenblatt: Yes. I had written my own.
Q: From scratch?
Richard Greenblatt: From scratch.
Q: O.k.
Richard Greenblatt: And so then as word got around- Well, there was a guy a MIT in those days named Hubert Dreyfuss, who was a prominent critic of artificial intelligent, and made some statements of the form, you know, computers will never be any good for chess, and so forth. And, of course, he was, again, very romanticized. He was not a strong chess player. However, he thought he was, or I guess he knew was wasn抰 world class, but he thought he was a lot better than he was. So anyway, I had this chess program and basically Jerry Sussman, who抯 a professor at MIT now, and who was also a member of our group, had played. It was around and it was available on the machine. People played it, and so forth. And basically Sussman brought over Dreyfuss and said, well, how would you like to have a friendly game or something. Dreyfuss said, oh, sure. And sure enough, Dreyfuss sat down and got beat. So this immediately got quite a bit of publicity.
Q: Yes. Word of mouth.
Richard Greenblatt: Word of mouth and eventually later more than- other forms of communication, as well. Anyway, so at that point, some other real chess players who also happened to be students at MIT, began showing up at the AI lab. And it so happened that there were three national masters who were students at MIT at that time.
Q: Oh, my goodness.
Richard Greenblatt: And so one of them, a guy named Larry Kaufman. He was probably the strongest one. He抯 a chess journalist of a sort to the present day. He lives in the D.C. area. He抯 an international master. He抯 not super strong, but he抯 pretty good. And then there was another guy named Alan Baisley, who which a long story connected with. And a third guy named Carl Wagner, who just played a couple of games. I really didn抰 get to know him that well, but he was also an MIT student. Anyway, all three of these guys came in and they started playing the machine. And I started getting in friendly with them. I started finding out something about what chess tournaments, and so forth, and what chess books were, and so forth. So, I had the idea. I don抰 know where it came from, but the way to gauge this thing was to play in a human chess tournament. So it turned out that in those days there was something called the Boylston YMCU, not too far from here, actually. It抯 just off of what used to be the combat zone. It抯 on Boylston Street between the Public Garden, Boston Common and the combat zone. It抯 a little thing, a little bit like the YMCA, except that it was different, called the YMCU. Young Men抯 Christian Association. And it basically provided services to servicemen mostly, who were docked in Boston, or whatever they were. Anyway, they had a chess club, and they had a fairly large space, multiple floors, and a gymnasium, and quite a few different things in downtown Boston. So at some point, I found out about this and we decided, o.k., let抯 play our computer in a chess tournament. So this was in the days before portable terminals of any kind. So for the first chess tournament, we got one of these big Checker cabs, and we put in a model 35 teletype, if you know that is. That抯 a big one, weighs-
Q: Yes. Much bigger than the 33.
Richard Greenblatt: Much bigger than the 33, weighs several hundred pounds, well, maybe a hundred pounds, and is basically a full load for the large trunk of a Checker taxicab. We took that downstairs. By this time there was quite a few people that were helping. This was a sort of a lab-wide activity, so I didn抰 do all this myself. But anyway, they hauled it up a couple flights of stairs and installed it in this chess club in downtown Boston. This YMCU was an old, stodgy thing, and had a funny phone system, so Nelson hacked their phone system. Again, these were old people that were very oh, I don抰 know, they抎 be horrified by the fact that their phone system was hacked, but we managed to do something there, clip in a modem. One of my friends who had a shop, the shop was at the AI lab, and he was quite a machinist and had nails and so forth. He made a device that clamped onto a chess clock so that when you pushed the buttons on the chess clock, this also activated a switch and then somebody did some electronics so that this switch then controlled something called the restrain tone, which was something that was a part of very early teletypes. It involved slowing down the paper tape reader in case the machine couldn抰 keep up with it. So we had a detector for this.
Q: For that tone back in MIT.
Richard Greenblatt: Back in MIT set up. And so the computer could basically sense its time, and sense whether its clock was on or not. And we played. These chess tournaments are rather serious. In other words, it抯 held over a weekend. Each game was- the basic time control, I believe, was 40 moves in- maybe 45 moves in two hours for each side. So the game would then take four hours. And there was five rounds scheduled, so there was three rounds on Saturday, and two on Sunday, or something like that. So that was sort of all day, night to evening, on Saturday, and then almost all day on Sunday. And so it was quite a big operation to do that in those days with the computer. We had people tending the machine, and typing in the moves, and getting things to eat, and dealing with problems, and putting paper in the printer. There was just a zillion little things that you had to worry about.
Q: Was there any problem in just getting that amount of time on the machine, or people were really supportive?
Richard Greenblatt: Yeah. There was no timesharing on the machine in those early days, so it was just.
Q: So it was just hold the sign up for a complete two days.
Richard Greenblatt: And this one machine was available for the whole lab, but on the other hand, everybody was sort of involved in looking at the game and whatever. So it wasn抰 really any work going to get done that weekend anyway, anyhow.
Q: Exactly.
Richard Greenblatt: So we first played in a tournament and it was just about a month and a half or so after I came back from that.
Q: That trip to California.
Richard Greenblatt: So that抯 right. So I think it was in December or maybe January.
Q: Wow. Now how was this. You had a teletype at the thing.
Richard Greenblatt: At the conference.
Q: At the conference. Now was that actually connected to the computer.
Richard Greenblatt: Yeah.
Q: So if you put in- you would type in the move the other person made.
Richard Greenblatt: That抯 right. The operator would.
Q: An operator there.
Richard Greenblatt: That抯 right.
Q: And then the computer would figure out what the move was and print it out, type out what the move was. And then
Richard Greenblatt: Yeah.
Q: The operator would move the piece on the board.
Richard Greenblatt: Yeah. And press the clock. That抯 right.
Q: And press the clock. So there wasn抰 any manual transcription at the PDP6.
Richard Greenblatt: No.
Q: O.k.
Richard Greenblatt: No. It was automatic at that point. By that time we did have a line printer
Q: So you didn抰 have to hack.
Richard Greenblatt: So we didn抰 have to hack the thing. That抯 right.
Q: The telephone. Tell me about do you remember what happened at that tournament?
Richard Greenblatt: Yes. In the first tournament, the computer drew one game. It played five rounds and lost four, one of which was to Wagner, who is this master who, of course, didn抰 have any chance against him. But anyway, it did draw a game against a guy named Ben Landy. Ben Landy was a local legend who was an older guy who was involved in sort of setting up the chess tournaments and kind of being some kind of a thing for the chess club, and so on. And he was. His rating was, I think, about 1440, which is not too tremendously wonderful. Anyway, he drew. That was the first tournament. Then about a month later, we played another tournament and again another five games. This time the computer won one, and beat Landy, the same guy; beat Landy, and lost the other four. So then the listing I have here is for the third tournament.
Q: These are all at the same place.
Richard Greenblatt: Yeah. I think all of these were at the YMCU. Later there were a couple also at the Prudential Center.
Q: All right.
Richard Greenblatt: Somebody had a thing inside the Prudential Center.
Q: But they抮e all local in Boston.
Richard Greenblatt: Both were in Boston. That抯 right. And so this one I have here is the first one where the computer really had a positive score. I just looked over it last night. It looks like we got a by in the first round. For some reason, we didn抰 play a game in the first round, so there抯 four games and it looks like the computer won three and drew one.
Q: Very good.
Richard Greenblatt: Some of these players weren抰 rated, so computing a performance rating wasn抰 too easy for that. But we then played in several more tournaments. The best single result, in my opinion, was a draw of an 1880 player. The way these ratings work, 1880 was what was called an 揂?player. And he抯 really a pretty good guy. I mean, he抯 still not a couple levels below. There抯 sort of roughly standard deviations, sort of have international master. It抯 a different scale, but you can kind of view it as above 2400. Then 22 to 24, was national master. That was kind of in the range where Baisley and Kaufman and Wagner were. Then 2000 to 2200 was called 揺xpert? And, again, pretty strong player. And then the next range below that, 2000 to 1800, was called 揂? And the median player, I believe, was in the lower A range. And then it went on down by 200 points, and increasing levels of the alphabet. So I think by the time you got down to Landy at 1440, you were down to a 揇?player, or something like that.
Q: I see. O.k. Just out of curiosity, how are the numeric ratings established through?
Richard Greenblatt: Well, there抯 an elaborate amount of stuff that抯 going, and the essence of the present system was in effect then, at that time. There was a guy named Ilo [ph?], Professor Ilo, who worked this out, and who has adjusted it over the years, and it抯 now become internationally standardized and so forth. So these are called Ilo Ratings. And an early form of his system was in use at that time. There抯 been a little bit of rating inflation, I think, but aside from that, these ratings are more or less comparable with ratings from the USCF today. The computer, by the way, did become an honorary member of the Massachusetts State Chess Association and the United States Chess Federation, which was necessary in order to play into these tournaments.
Q: These ratings, could you just explain just briefly the theory, how a rating is computed, or just the general idea behind it.
Richard Greenblatt: Well, basically, there抯 approximately speaking, if you take the difference in rating between two players, then the probability of winning is some kind of an S curve. At a certain level, the lower rated player has no chance of winning at all. Then it basically kind of comes down as an S, and then at some point it goes the other way. So this interval is approximately 400 points. In other words, if a guy is 400 points above you, then statistically speaking, you have nearly no chance at all of winning. The way it works is that, in those days, I think it抯 a little more elaborate now, but basically given the two guys play, you sort of subtract their ratings and you kind of get what the probability- what the stakes are. So if you lose to a guy, there might be say ten points at risk. And if you are a 100 points ahead of him, then if you win, you get only three points, say. And if you lose, you get minus seven, because you were expected to win and you lost, so it takes off more.
Q: I see.
Richard Greenblatt: But if you抮e way above him, if you抮e 400 above him, then you get practically nothing. You might get just one or two points, because you were expected to win and you won, and so who cares? So that抯 the kind of way it is. And in those days, it was whatever you win, the other guy loses, so you抮e actually sort of playing for a stake and to keep the system inflated, there抯 sort of activity points. Just for entering the tournament at all, you get a couple of points, or something like that.
Q: O.k. And somebody worked out the math and it all appears to actually sort of be a stable system.
Richard Greenblatt: That抯 right. And particularly now with more refinements than were in place in those days.
Q: O.k. Good. Thank you.
Richard Greenblatt: So, that抯 right. So the computer抯 performance rating for that tournament- given any tournament one of the things you can compute is a performance rating, which is what your rating would be. And I believe it was 1200 or so. In other words, a draw with a 1440 player, and then four losses. The best the computer ever did, which I believe was after this one, the performance rating, I believe, was 1820, or something like that. So the computer did get up into an A level. I was working on this, but it should be emphasized, of course, that this was kind of a sideline. I mean, so I didn抰- I was also writing system software and managing and doing numerous other things.
Q: You had a real job, too.
Richard Greenblatt: A real job, and doing research and a number of things. And it was an awful lot of work to play in these tournaments. So I would guess we probably played altogether, probably in half a dozen major.
Q: Now did you work on the program in between each tournament, in terms of trying to improve it?
Richard Greenblatt: Oh, yeah. Well, particularly during those days, yes.
Q: Yeah, during those days.
Richard Greenblatt: Sure. People would play the program, and then it would make bad moves in certain states, you know, in the game. And so then we would come in and try to figure out why, and put in things to improve it. And that抯, of course, one of the central things about chess, as a matter fact. It isn抰 really appreciated by very few people to the present day, which is that it is so terribly precise, so it is so important to have the program debugged. So that is really the most important thing about the playing strength of a program. And so in particular, these computer versus computer matches, which they sometimes have, well, those are likely to be just random noise, at least until it gets to quite a high level, because the programs just simply aren抰 well enough debugged and if you get to a situation where, for one reason or another, one program does something bad, well that.
Q: Yeah. Something stupid that wasn抰 in the- that抯 literally the result of a software bug as opposed to.
Richard Greenblatt: And that抯 the reason why I never wanted to play in these computer versus computer matches. And never played in any of these world championship computer chess championship, or any of those matches, because it was just too random in the sense that the computer makes one bad move and then you抮e done for a whole year. You lose and you don抰 get a good result that tournament, you don抰 play again until the following year. So I thought it was much better to play against people, which No. 1, was sort of the goal from an artificial intelligence point of view. And No. 2, it抯 more available in the sense that anytime you want, you can go down to your local chess club and play in principal.
Q: Could we roll back just a little bit, back to could you tell me about some of the things you did when you wrote your program after looking at Alan抯.
Richard Greenblatt: Well, see, I didn抰 look at his.
Q: You didn抰 really.
Richard Greenblatt: I looked at the printout of the chess analysis, but I did not actually look at his program.
Q: O.k. O.k.
Richard Greenblatt: And I didn抰 discuss chess with Alan, and so forth. Again, even though I knew him well, the whole chess thing just wasn抰 happening at MIT at that point.
Q: Exactly. And he wasn抰 involved in it anymore, either, at that time.
Richard Greenblatt: That抯 right.
Q: So talk to me about what you did in your program that you think improved over.
Richard Greenblatt: Well, I did have quite a bit of stuff. I mean, I had a plausible move generator. It had quite a lot of databases of square control and pins, and attacks and masking attacks, and ability to estimate the swap out that would occur on a particular square. How many pieces from the two sides are attacking that square, and so forth. And it also made- this is quite a lot of databases, and so a lot of this was maintained incrementally so that, you know, as you make a move, you have to subtract one from all those squares, which you抮e no longer attacking; add one to all those squares which you are attacking, and do that for several other databases involved in that, too. So that was quite a bit of it. But it had a plausible move generator. It had a quiescence detector.
Q: Now what was that?
Richard Greenblatt: Well, that is basically where you consider all captures. I mean, the basic idea is that you consider the best so many moves as a function of depth in the tree down to a certain depth. And then at that point, what you抮e trying to basically do is come up with a representative value to return and to return into your minimax alpha beta, you know.
Q: Yeah. This is the evaluation problem.
Richard Greenblatt: That抯 right.
Q: How good is this position really?
Richard Greenblatt: That抯 right. And however, the problem is, of course, that if you抮e in the middle of a capture, or in a situation where one piece is hopelessly trapped and can抰 get lost, can抰 be saved, then you抮e at great risk of returning a incorrect value.
Q: Oh. Because you really needed to look just one more ply ahead, or maybe two, to understand what抯 really going on.
Richard Greenblatt: That抯 right. And in particular, if it抯 a swap out, then it抯 just absolutely necessary.
Q: By a swap out you mean your trade.
Richard Greenblatt: Queen takes queen, pawn takes queen. I mean, if you look at queen takes queen, oh, I抦 a queen up. Well, of course, you抮e not. I mean, pawn takes queen is coming right back. So therefore, if you抮e in the middle of a swap out, you absolutely must look at all the captures and recaptures and get to a quiescence state there. And even if it抯 not in the middle of a swap out, if you have, for example, two different pieces being attacked, then there抯 considerable danger that you might lose one, because, you know, you only have one move. You move one, and then pawn takes the other one, perhaps, or something. So I had things like that that basically did look for possible advantageous swap outs, and if there抯 more than one of those, then it says, o.k., well, we better look deeper, because we might be able to save both of these pieces.
Q: So you would get down to however many plys you were going to, in terms of looking at all the possible moves, and then rather than just evaluating at that point, the quiescent goes.
Richard Greenblatt: That抯 right.
Q: And looks and sees- looks for these situations that are going to make the evaluation criteria bad, or lied to you, and figures- and goes a little bit further.
Richard Greenblatt: That抯 right. And then there抯 also some technique involved in assuring that the position, the values are as comparable as possible. In other words, you are comparing positions where it抯 your turn to move versus positions where it抯 his turn to move. Well, you see, that抯 a problem because, on the one hand, if you just compare it as a static position, you won抰 get a comparable value because he gets a move, and he moves a piece to some advantageous square, and then now controls more squares, and so forth and so on. So that抯 good for him. So you抳e got to try to estimate what the amplitude of the oscillation. We抮e talking about a minimax. One player is trying to maximize. The other player is trying to minimize. So therefore, the general value of the function tends to go kind of like so. You need to try to estimate what the amplitude of that oscillation is and kind of average it together so that you get a value that抯- from the odd plys, that is reasonably comparable with the value from the even plys.
Q: Ah. O.k.
Richard Greenblatt: And so on like that. And oh, quite a few other things, actually. Sometimes you have a long sequence of moves which however are forced. In other words, the fundamental thing you have is an exponential growth. So, you know, if you just- you do have a depth抯 parameter. You can set it to higher values, but if you do, things will slow down generally at around a factor of six or so, per depth. In other words, there抯 around 36 or so average moves, and the minimax thing more or less cuts it down as a square root, so more or less it抯 comparable to a exponential tree growing a factor of six at each ply. So there抯 quite a few actual techniques of various sorts that are involved in that.
Q: Any others that just come to mind?
Richard Greenblatt: Yeah. Well, this thing where if you have a forcing thing, then what you sometimes do is play out to the end of that variation, and then do another short search from there.
Q: O.k.
Richard Greenblatt: So this allows you to just sample a small part of the tree which, however, is quite a bit deeper than what you were going to be able to look at otherwise. And this is a good thing to do.
Q: Yes. So the fact not just having a single layer, you just go down certain number, but in certain cases you go deeper.
Richard Greenblatt: Now, just to talk about what our hardware was at this time, we were running on the PDP6 computer. It was approximately ten microseconds per instruction. We did have the so-called Moby memory. The original configuration of this machine delivers from Dec, had 16k of 36-bit words. This memory was two relay racks big. It cost 0,000, two microseconds. However, in a famous interlude, at some point ARPA said, o.k., you AI labs doing work, mainly us at Stanford, and maybe Carnegie, a couple of others, mostly us at Stanford. They said, o.k., we抳e got an extra 400 grand, so you can buy a big, new toy. So basically we bought something called the Fabritech Moby Memory. This was a 256k memory, 40 bits. It had two wires going through each core. The normal core memory had four wires going through each core. And it was two big bays, so it was a little bit larger, but it was a little bit larger physically, but 16 times larger logically. And at the time, was one of the very largest memories in the world. Who knows if there was some other one comparable some way. But it was vastly cheaper, so we had a special deal with Fabritech to buy it. It cost us just under 0,000. Stanford took their money and they bought something called the Librascope disc, which was a huge.
Q: Huge. Yes. I think we have one of those at the museum.
Richard Greenblatt: And so on.
Q: It抯 amazing.
Richard Greenblatt: So that was a whole story.
Q: But was this still a two microsecond cycle time memory?
Richard Greenblatt: Yes.
Q: This Fabritech?
Richard Greenblatt: That was a little slower. It was a little slower.
Q: I was going to say, you had to give up something when you went to a four-wire scheme.
Richard Greenblatt: It was 2.8. Also, these memories were on a memory bus and they were asynchronous, so when configuring this, you know, you were sometimes concerned by the access time. It was a certain length of time when you got the data back. Then there was another time where the memory is cycling, writing the data back and so forth, before it抯 ready to accept another. So the Fabritech was a little bit worse in that. Not only was it 2.8 microseconds, but also the data came a little bit later in that 2.8 microseconds.
Q: All right. But the PDP6 had the ability to deal with an asynchronous.
Richard Greenblatt: That抯 right.
Q: Main memory.
Richard Greenblatt: It had this memory bus.
Q: Yep. O.k.
Richard Greenblatt: Yep. So it was about 1/8th of a mip, is another way of looking at this. And what our search depths were, as I was just looking on this listing, for tournament play, our basic depth was 15, 15, 9, 9, 7. So at the first two plys, we would look at the top 15 moves, plus captures. 15, 9, 9, 7. And then 7, however deep you have to go due to these quiescent searches, you keep it at 7, but you never go below 7, so the thing with the Kotok program originally had with the 4, 3, 2, 1. I mean, of course, their searching was much slower, so they couldn抰 set it much wider than that, apparently, but just having it at any ply, a level of 1, you抮e just introducing almost pure noise.
Q: Yeah. Because there抯 no decision.
Richard Greenblatt: That抯 right.
Q: What抯 it tell you?
Richard Greenblatt: You抮e just taking what appears to be the best move and that抯 a very fallible estimation.
Q: Yes.
Richard Greenblatt: And then you抮e not looking at anything else. So that抯 what I was just telling Kotok the other day, and he pulled out his thesis and sure enough there that was. And I told him I thought that his program would have played much better had he simply set it to 7, 7, and set the depth to 2, instead of 4, or whatever. And he said, well, they basically were not sufficiently good chess players to really understand that that was required.
#### End of Tape 102634500_2greenblatt.mp3 ####
Q: Okay, one thing I had wanted to ask was the ?Ken Thompson, you know.
Richard Greenblatt: Yeah.
Q: If ?quite a long time later became pretty interested in ?
Richard Greenblatt: Yeah.
Q: ?in chess and he ?we made a lot of use of ?he actually made these ?uses of databases of, you know historical classic opening moves.
Richard Greenblatt: Oh yeah, no we did too.
Q: Okay. Could you maybe talk a little bit about what you did there?
Richard Greenblatt: Yeah, we had that right from the beginning, or nearly the beginning. This guy Larry Kaufman generated our opening book. And I had a thing that encoded the opening moves and you know in a small amount of memory so that it fit in the computer memory and so forth. And he ?so he did that.
Q: And you put in main memory?
Richard Greenblatt: Yeah.
Q: Okay. So it was loaded in main memory?
Richard Greenblatt: That's right.
Q: Alright.
Richard Greenblatt: And it was compacted pretty efficiently. And you know, there was a ?by later standards it wasn't so big, but at the time it was pretty good sized. I don't know I think it was probably 8,000 or 10,000 moves in there.
Q: Wow.
Richard Greenblatt: He ?Kaufman spent quite a while doing it. Of course, a number of the moves, you know, would ?the game would just play it. There was very little chance of it actually staying in the book. You know, the ?but I think Kaufman did a good job and he did ?we did perceive some of the ?well of course, one of the basic things about computer programs, really to the present day, is that they're very tactical. It's much easier for them to see that the tactics than the strategy. So what's called a closed position is hard for a computer. That's where all the pieces are blocking each other and it's very long maneuvering. Whereas, an open position which is sort of tactical combinations is much more to the computer's liking. So therefore, when designing the opening book, you want to kind of play offbeat and kind of unbalanced type openings that tend to lead to these closed position ?open positions, which are then good for the computer. And so we realized that and you know, I think Kaufman did a fairly good job of ?
Q: So there was ?you did some strategic thinking in figuring out what to ?
Richard Greenblatt: That's right. But it ?
Q: ?what is the opening positions are?
Richard Greenblatt: That's right. It's not based just merely on the pure logic of chess of what is objectively the best move. What you're trying to do, of course is to get into positions where the computer will play its ?will understand "the position" and play its type of a game to the best advantage.
Q: Alright. About how deep do the opening, you know ?and if I give an opening ?
Richard Greenblatt: Well it varies tremendously. I mean in many cases ?
Q: Would you ?
Richard Greenblatt: ?of course there's a certain game that is played and frequently it's just easier to just type in the whole game or pretty deep into the game, rather than cut off someplace. But that is just, you know, that is just sort of a gratuitous at that point. It ?but in most positions, as a practical matter, it would stay in the opening for opening book for maybe six, seven moves.
Q: Oh alright.
Richard Greenblatt: Not too many.
Q: Yeah, but still.
Richard Greenblatt: Sometimes it's ?
Q: ______ number.
Richard Greenblatt: Yeah. Sometimes, and of course it also ?
Q: It depends on the opponent? The opponent goes out of the ?
Richard Greenblatt: Absolutely.
Q: But then ?
Richard Greenblatt: And the opponent plays pawn the queen, rook three then you're immediately out of the book.
Q: Yeah.
Richard Greenblatt: Right, first move.
Q: Yeah. Okay.
Richard Greenblatt: So.
Q: Okay.
Richard Greenblatt: It also has something to do with white versus black. Of course, you know with white, you know, the white is trying to get a position and get an advantage. The only thing that sort of makes opening books for humans finite at all is the fact that white is trying to get an advantage and black is trying to equalize. Once black is equalized then that's the end of it. So as far as the opening theory is "concerned" for humans.
Q: Okay.
Richard Greenblatt: So then in the ?you know, if the computer is playing white, well again it can just play a completely random move such as pawn the queen three, it will immediately be out of the book ?actually pawn the queen rook three let's say.
Q: Yeah.
Richard Greenblatt: And you know it will still be an even position. So you know, with white it's very easy to get out of the book in an even position. With black, on the other hand, you know black is kind of the underdog trying to creep up. So if black just plays a random move then there's more of a danger of white getting a significant advantage out of this.
Q: Yes.
Richard Greenblatt: And actually, you know having ?maybe a tangible advantage. So ?
Q: Okay. So what about "N" game, when it comes to relatively few people have done a fair amount of work?
Richard Greenblatt: Yeah. I didn't ?
Q: In analyzing those even using computer analysis.
Richard Greenblatt: Well I just did the bare minimum that was necessary to know about pass pawns and, you know, that if somebody has a pawn that's running on, you know, you don抰 want to stop without giving that due course and so on and so forth. But I never did anything serious with "N" games. They're ?some work had been done. A gal named Barbara Huberman who later married, I forgot what her name is, so ?anyways she was around for many years. I never really met her. I heard about her, but she had done some kind of a classical "N" game thing. But these come up so rarely in practice that from a practical point of view you have to be at a very, very high level of play before there's really any significant chance that that will improve the strength of the program.
Q: Okay. Alright. Good. What else should I ask you about, you know, you're ?
Richard Greenblatt: Well ?
Q: ?design of the program and the structure? What other things?
Richard Greenblatt: Well let's see. I ?what we could say is that after that first year then I returned in '66 at the Fall Joint Computer Conference and I generated a paper. So in the fall ?in the proceedings of the Fall Joint Computer Conference of 1966 there is the paper that I wrote about this thing, and I wrote it with ?
Q: Oh really, okay.
Richard Greenblatt: ?a guy named Steve Crocker who was a graduate student at the time. Later went on to become a _________ and various things. And ?but then I ?you know after the early 60's, I mean mid-60's, say '67 or '68 I ?the chess program was there and I would kind of toy with it occasionally. But I really didn't ?it wasn't really that large a fraction of my overall effort, because I was doing so many other things, involving system software and later Lisp machines and so on.
Q: Okay.
Richard Greenblatt: However, we did play, occasionally, you know in a chess tournament or occasionally somebody would visit or occasionally there would be a match set up. We did play Ken Thompson once, he won.
Q: With his bell?
Richard Greenblatt: With bell.
Q: Okay.
Richard Greenblatt: That was just a ?we played a couple of games by ham radio with a program in Switzerland, we won that. There ?when the Papert bet up after ten years, you know, essentially Papert and this guy David Levy had made a bet that a computer program wouldn't beat him in ten years. And ?
Q: David ?it wouldn't beat David Levy?
Richard Greenblatt: Yeah. And so David Levy won that bet, however he was only ahead by about three or four years, you know. That ?David Levy was an international master, so he was you know a strong player but not ?on the world stage he was not that strong.
Q: Mm-hmm.
Richard Greenblatt: And you know there's ?I would guess there was probably something like at least a 100, maybe a couple 100 grand masters who, you know, the next ?
Q: The whole next ?
Richard Greenblatt: ?that's right.
Q: Two hundred points.
Richard Greenblatt: So basically I think that bet came due in approximately 1980 or so and probably six or seven years after that was when chess programs kind of passed him by and he would ?I don't know if he ever actually ?played a match with a computer program that he lost. But he would be a definite underdog of course these days.
Q: Yes. Okay.
Richard Greenblatt: But anyway, we played that and we did a few things. Then ?well another major chapter in this thing ?I guess we can skip forward a few years.
Q: Okay.
Richard Greenblatt: And in the early 70's we had gotten to the point where the PDP 10, PDP 6 and then the PDP 10 address space, which was 256K, was no longer enough. And, you know, but programs were trying to expand and there wasn't enough room because you're running out of address space. And so we realized also that, you know, there's a lot of things can be done. So that ?we ?it became clear that we were going to have to do something else. And we ?I started the Lisp machine project in 1973. And for about a year I worked just by myself on a ?an interpreter that ran on the PDP 10; it was a software emulator; emulated a macro code. And the macro code was designed especially for Lisp, the tag machine and so forth. During the ?approximately the same time frame, Licklighter, you know the ?there had been a split of the AI Lab and the so called laboratory for computer science into two laboratories. And Professor Licklighter, who had sort of been head of the LCS decided he didn't want to do it anymore. He was getting old and he wanted to do his research and so forth. So it came up that a guy named Ed Fredkin, who had been around the AI Lab ?
Q: Ah, yes.
Richard Greenblatt: ?and Fred Miminsky and so forth was kind of standing there and he had been ?had had a successful company and I had known him, and my friend ?a student that had also worked for him and so forth ?and information international. Anyway, he was sort of tapped to be head of LCS.
Q: Okay.
Richard Greenblatt: And he was a capable guy. However, he was ?he wasn't ?it wasn't a very good fit actually. And he has his island in the Caribbean that he likes to go to and he disappears so forth and so on.
Q: Yes.
Richard Greenblatt: So, but at some point he also had a interesting line of research having to do with so called digital physics. The idea that physics is explained by a ?some kind of an automanon machine and some sort of time. So he ?a particular guy named John Masuris was a road scholar at Oxford. And he was doing his PhD and he kind of bogged and he decided to ?that this digital physics was interesting and he would come to MIT ?
#### End of 102630621_3greenblatt ####
Q: Okay, so. Let's see, where did we sort of stop with the ?I guess probably after the ?you know you're first sort of work on chess. Maybe you could just sort of highlight what things you worked on up to the ?you know up to the ?you're involvement in Lisp and Lisp machines and that sort of thing.
Richard Greenblatt: Well it was all going on in parallel. So you know ?
Q: Okay.
Richard Greenblatt: ?I worked on all these things. And Lisp actually ?well this whole group at the Mala River Club of course had done this original Lisp on the PDP 6 and the prototype at Deck. And then after we got a machine, you know, we kind of picked it up and Stu Nelson and I did a Lisp compiler and that was a big deal in those days.
Q: Okay. The first system had been just an interpretive system?
Richard Greenblatt: Yeah, yeah. There had been a compiler before on the 7090 system but ?and some other ones, not local, but not ?but ours was quite a bit fancier.
Q: Okay.
Richard Greenblatt: In various ways. So anyway, yeah, that was a major project and that of course was our primary resource ?just about all the programming in the lab and in maxima and so forth was done in Lisp by using that compiler. And then I sort of passed on and various people took it over. I just poked at it a little bit, but too much for a while. But I guess really the thing I did mostly for quite a while was the time sharing system and you know it was just a lot of things; different IO devices and different developments, because a lot of things had to do with time sharing system.
Q: Now how did that time sharing system get started sort of in parallel with the project Mac 1, sort of ?
Richard Greenblatt: Well the one that was really in parallel with was the Deck one. I mean you know we got our machine and Deck, you know was developing a system but it was very primitive and had a lot of things we didn't like about it. So ?
Q: If it had been really cool you might have just said, oh why do we need to do this?
Richard Greenblatt: Yeah.
Q: But it wasn't?
Richard Greenblatt: But it wasn't. Actually, initially, we didn't use a time sharing system. Initially, it was just single user and then ?and of course at that time there was ?it was only 16K of memory on the machine. What could you do? But then we did get the Moby memory, the 256K, and of course that was when it became possible to do serious time sharing and stuff. And ?
Q: Now were there drums on this machine? Was there backings ?you know ?
Richard Greenblatt: Well that's another ?
Q: Random access backup store ?backing store?
Richard Greenblatt: That is a long story. And we can go into it. It's fun. But it will take us a while. Well, the short answer, is initially no. And it was a real problem over the years. The way it developed was that there was a company called Analex here in Boston. And they had made printers and at some point they decided to make ?try their hand at disk drives. And it was the days of the IBM?311, so ?but these were very expensive. They were only available on rentals. The interface wasn't available; you know there's no controllers. So you know, we were hesitant to do that so ?but Deck didn't have anything that was reasonable that you could buy from Deck at all. So we decided to get this thing from Analex ?the thing from Analex was really quite a kludge to this day. It was based on a mechanical D to A. I don't know if you seen these things ?sort of a thing on the wall. It consisted of a shaft that went down this way, with clutches and gears and little wheels that had two D10's. So the thing had a clutch and it could either go to stop in one of two stable positions. And then sort of a like a locomotive, like a steam locomotive thing, there was shafts and balance beams, so forth connected to these in such a way that it affected a mechanical digital to analog converter. And there were ten of these wheels, so each of one could be in either of two positions. So basically, you know, it had the unusual characteristic, which they thought was a great thing, was that the seek time was independent. You know most disk drives you seek from one track to the next and then you get to the near one ?
Q: A lot faster than the far one?
Richard Greenblatt: That's right. Well this idea was interesting ?but that was not the case. That it took a certain time for these wheels to make a half revolution. But once the wheels had made a half revolution then the balance beams did their work and the heads were then in the right position. So the seek time from any track to any other track was a constant.
Q: Okay.
Richard Greenblatt: Well so we got the raw disk drive from Analex and we built our own controller for it; at Holloway ?meaning they built it. And we didn't know anything. The machine, it turned out, was just outside the door of this machine shop. This guy had dropped out of MIT and went to work at Lockheed as a machinist. And you know, he would just ?had this milling machine and he would just really feed the metal in there and put in the heavy oil, you know, and the smoke would come out of it. This poor disk drive was just sitting just outside the door of that machine shop.
Q: Oh.
Richard Greenblatt: We didn't have the faintest idea of what was involved as far as ?
Q: Contaminants or anything?
Richard Greenblatt: ?contaminations or ?and neither did the Analex people, really. I mean they had a little ?some air cleaner but it just was totally inadequate. Needless to say we disk crashes. Just then, Analex had been kind of getting pretty old in the tooth as far as their printer business and they hadn't been keeping that up too much. And they ?another company in particular called Data Products had come out with a better printer.
Q: Yes.
Richard Greenblatt: So basically we didn't accept it and Analex decided that they were going to get out of the disk business and so we sent it back and we didn't have disks for ?well we didn't have any disk at all for a year after that. We just had micro tape, the device for ?you know it was 65K of 36-bit words on a little tape, yea big and it took a couple of minutes to wind.
Q: Right.
Richard Greenblatt: So ?
Q: Could not make a paging story even if you had the idea.
Richard Greenblatt: Yeah.
Q: Out of that.
Richard Greenblatt: We actually did just sort of think about that just a tiny bit. But of course it was totally impractical.
Q: Yeah.
Richard Greenblatt: So then there was another company called Data Disk. And they had a kind of a player type disk that's kind of a large vitrola desktop unit and it recorded on some hard surface. Or it was actually a contact recording thing. Basically, we made another controller and software to interface that thing and we used it for a number of years. But that thing had the problem that it was very small. They ?two of them together ?we had two units ?the two of them together could dump the Moby memory once. So each ?the entire capacity of the unit was 128K, 36-bit words.
Q: Yes, alright.
Richard Greenblatt: So again, it was ?and it was quite slow. But at least you could have the temporary stuff with the data that you were working on while you were on the machine. At the beginning of your session and the end of your session, you read ?wrote for micro tapes. But in the middle of your session you could use that and then we did the ?this 256K main memory at that time. So that was you know pretty big and allowed you to. And then finally we decided well we really are going to have to go this 2311. By then IBM had kind of loosened up their policies on 2311's a little bit. And so Stu Nelson, who by this time, was in the Systems Concepts Company, but was still good friends, he and me ?sort of mostly him, made a 2311 controller. And we actually got some 2311's and used those. And then finally, Systems Concepts, developing some of this hardware technology I was ?we were talking about decided to develop a sort of a more of a heavy weight IBM controller. It turns out that this was a endemic problem. I mean Deck had no peripherals. Anyone who wanted to use Deck machines for serious work was just at a great loss for peripherals because the tape drives and the disks and so forth were just not commercial quality compared to the IBM ones of the day. So anyway they made this disk control and we used that. And then finally, they made another product called and SA10 which was a IBM channel interface. And actually interfaced not directly to the disk drive itself but to a actual IBM channel which then meant it could interface to most any IBM device. And that turned out to be in incredible product for them. They sold those things for ten years or something. You know, and then they sold a great number to a lot of Deck customers and so forth.
Q: And that allowed you then to ?if you had the money you could buy ?
Richard Greenblatt: Buy the ?
Q: ?commercial grade ?
Richard Greenblatt: Yeah. We however didn't do it.
Q: Of course.
Richard Greenblatt: We had this one that was interfaced just directly to the ?to what by that time was called 2314's. And then they ?there was ?for a little while there, there was actually a competitive start up company where a bunch of different companies started making their own 2314's and going into commercial ?
Q: Yeah.
Richard Greenblatt: ?competition and so forth. So we ?there was a company called Calcom that made ?and so for ?we got from them and basically we wound up getting double density 2314's ?
Q: Okay.
Richard Greenblatt: ?upstairs and downstairs and we had a full eight of those. And that was a fairly decent ?
Q: A road bust amount of secondary storage ?
Richard Greenblatt: ?of secondary storage.
Q: ?for a time sharing system.
Richard Greenblatt: Yeah. And that's right, so.
Q: Good.
Richard Greenblatt: But ?
Q: Well I interrupted you.
Richard Greenblatt: That was over a number of years. I mean each of those cycles of designing are hard ?a piece of hardware and building and writing software for it and so on, took at least six months and more like a year or sometimes almost a year and a half or two years. So you know, just years went by and you know progress was gradually made. Also on the display front, you know there's likewise a story of ?we ?you know we had teletypes and Tom Knight made a teletype multiplexer and then there was a totally funny GE system that we got and we interfaced and then several things. And then finally one time McCarthy came to visit on sabbatical or something and we got into ?
#### End of 102630622_4greenblatt ####
Richard Greenblatt: ?that抯 of interest is the software. There had been two main traditions, shall we say, in Lisp, for some years. One was MACLISP, which was basically our stuff. The other was called Inner Lisp, and it had originally started at BBN, with Bob Rowe and Tidelmann [ph?], and then, at some point, Bob Rowe and Tidelmann moved to Xerox Park, when they started having Xerox Park. And they called it Inner Lisp, and developed it at Xerox Park. And at some point, ARPA put their foot down and they said, okay, let抯 make things common. And they convened a meeting at SRI. And they sort of knocked heads together, and we somewhat grudgingly- but, at that point, things had worked out that the two Lisp systems had pretty much evolved to the point where they had solved the major problems. In some cases, though, there were trivial differences in how they- certain functions were implemented or certain properties or this or that. Anyway, there came a standardization effort, and it went through committees. There抯 several different umbrella organizations. Of course, if you抮e a company, you抳e got to be very careful because you can go to jail for anti-trust violations, if you抮e not careful with it- talking with your competitors, and one thing and another. So, we wound up forming this thing called the X-3 J-11, Standards Committee. And it was chaired initially by a guy named Bob Mathis who was from the ARPA program office or something like that. And basically it met a number of times over a number of years, and considered, in detail, zillions of different issues, and eventually hammered out something called Common Lisp, which is sort of the LISP standard, to this day. Some comments I can make on that, eventually, if we get a chance. But, anyway, essentially we had already had the early stages of Lisp machine project fairly advanced, by this time. And, so, Common Lisp wound up being Lisp Machine Lisp, I would say, about 85% or 90%. There were just a few situations where we made some accommodation to the Inner Lisp people, or something like that, or did something that was an improvement on both. But anyway, that committee started and kind of went on, with this bureaucracy and so on, for many years- quite a few years. And so, basically, we made these caters, and it really got to be quite a hardware making operation. It was actually more hardware than any project at MIT had ever made, up to that time, and I wouldn抰 be surprised if even to this time because, in total, we wound up making about 35 caters. And we had an outstanding offer, at this time, that if you had ,000.00 in your research contract, we would make you a cater- which was a fantastic buy, at that time. And so there was quite a few groups around campus that took up on it. There was this plasma physics group and there was several groups over on the main campus, and so on. And there was a- particularly, a guy named Tom Callaghan. Tom Callaghan was just really the greatest guy. He was a little bit older. He was sort of in charge of our shop and did a lot of that, and stuff. He had grown up in South Boston. He knew Whitey Vulgers [ph?], and so forth, from having grown up with him. And he was just the best guy. Anyway, he sort of handled our construction operation. The actual construction was outsourced, but we actually, of course, had the parts and had to order the parts and assemble them, kit them, whatever, and send them to someplace where they got?
Q: Wire wrapped.
Richard Greenblatt: Well, wire wrap was one thing, and then stuff- boards and so forth. Then I developed a machine which buzzed out these processor panels. So, you got a backdoor of a cater, which had six of these panels on it. And this thing had two XY probes, with probes, and it used just a plastic- a gear- plastic chains, and people said, oh, it would never- it wasn抰 particularly accurate. But it had a rather cute thing that calibrated itself, and it worked quite well. Anyway, so the thing could basically just sit there and go bung, bung, bung, bung, and buzz out the entire cater processor and find any missing wires and stuff like that.
Q: Wow, that抯 great. Okay.
Richard Greenblatt: And so that was a good hack. And we used that and that greatly accelerated check-out of machines. After things had got well sorted, there was-it got to the point where there was about a 25% chance or so of the machine just working the first time. Of course, sometimes it didn抰, and sometimes it could be very frustrating to find- to check it out.
Q: Sometimes, it can take a wile to find it.
Richard Greenblatt: Yes. And we gradually got better diagnostics, and so forth, that would accelerate that process. Anyway, so, yes. So we made quite a pile of these machines. Now, one particular contract, which was of note, and maybe- and, again, we抣l put in some- close some things here. But, see, in those days, and for many years, MIT had an incredibly poor computer situation, as regards its students. In other words, if you were part of a research project, then you had access to lots of stuff, where there were all kinds of research projects, et cetera. But, if you were just part of the main university, then what you had available to you was some old IBM 360 card batch thing- number one. And number two, this project called Multix, which grew out of CTSS, number one. It had a very long history, and whatever. But, basically, it was based on GE, and later Honeywell, equipment. And this was very expensive and very limited, in capability, compared to the Deck equipment. At some point, there had been a bid, and they tried to evaluate that. But Deck, it really wasn抰 a big enough company to be viewed as a serious competitor. And so Deck lost the bid and it went in with GE. Anyway, to make a long story short, this was a whole develop of what was supposed to be a ultility. It had a few good ideas, but it had a tremendous number- it had an incredible bureaucracy. It was based on the idea that bureaucracy is good, and what we really need to make this a business is bureaucracy. So, in order to do anything, required incredible numbers of meetings of groups and sign offs and just? It wasn抰 based on displays at all. So?
Q: Was it based on teletypes- ASR 33抯?
Richard Greenblatt: Actually, it was originally based on 37抯, I think it was. The 35抯 had only uppercase. And they realized they needed lower case, as well. So there was a model of teletype called a 37, I believe- it might be a 39- that did have upper and lower case. It turned out to be a dog. So, although the system was sort of designed for that, they wound up really using Selectrics, various versions of IBM Selectrics terminals. But, the system had long names- mumbo, underline, mumbo, underline, mumbo, underline- which was actually a good thing, and innovative at the time, from a system engineering point of view. However, they didn抰 do anything to try to ameliorate the typing, from the user point of view. So you just had to type incredible amounts of stuff to get anything done- long lines of- and if one character was wrong, it would just barf at you and you抎 have to type the whole thing over again. Make a long story short, the thing was a collosaful [ph?] white elephant, which MIT more or less forced through. Corbito and so forth, being powers that be, they managed to get some kind of an institutional commitment from MIT, which says, come what may, MIT will support this system, as a institute resource, for five years, or something like that. And so they did. But the resource was essentially very, very limited. The whole system had like 2 or 3 boxes, each of which had 128k words of memory in it. This system had a very, very large virtual address space- another one of its primary features- which was, in essence good, in some ways, in a lot of ways, but yet it was kind of ahead of its time and they didn抰 really work it very well. The thing is, the overhead was just very high and it was very hard to get anything done. So, this was, however, was an institute resource. It was extremely expensive. And these people had the idea that the way to make everything commercial- they had a form of the neocon- economics idea, that if everything is accounted for, and charge what it costs, then the forces of the market will optimize things and the system will be configured with the right amount of the various pieces, depending upon the price. And so they, of course, had to dilute this eventually. But the original idea was that you would pay for residency time, of core memory, and for transfer rates- every little operation that went on during the?
Q: You抎 pay for.
Richard Greenblatt: You would individually pay for- that抯 right- and you would thereby optimize it by reducing your?
Q: By natural economic.
Richard Greenblatt: Yes, that抯 right.
Q: Yes. As opposed to just getting upset.
Richard Greenblatt: Or trying to get something done. That抯 right.
Q: Yes, exactly.
Richard Greenblatt: So, anyway, this project was going on this whole time. And, in fact, it was going on in our building, and much of the time the machine was on the 9th floor- they went through several generations of machines, and one thing and another. And, at some point, Honeywell bought out GE, or something, or GEI, and they moved Honeywell- yes. At some point, Bell Labs dropped out of the project and they did Unix- there抯 another story. And Honeywell moved in, in Text Square, in the area, actually, where this Adams thing had been partially. Anyway, so that was mostly a big rat hole, where the institutes always were going down, and very, very, very little useable resource was being devoted to students, or anything like that. So, meanwhile, however, out on the west coast, McCarthy had this long-term commitment to time-sharing for the masses, and he was pushing that all along. And so, at some point, he evolved something called LOTS- low overhead timesharing. And this consisted of buying a 2060- off the shelf, by that time- Deck machine, and plunking it down in the middle of a hall in the main Stanford campus. This is right next to the Stanford bookstore, just- there抯 some architecture, with a big internal atrium. And they just opened it up to students, on a low overhead basis. They ran the thing with just a handful of people- six or eight people which, for those days, was small numbers to run such a thing. And this was low overhead timesharing. And it was open to the Stanford students, and they got a bunch of cheap terminals and they did this and that. And it was really quite great, for its time. And MIT was being left seriously behind, in my view- and no doubt other people抯. And that actually persisted, for a couple of years, that Stanford had LOTS- MIT had nothing.
Q: And the name was maybe chosen with some malice.
Richard Greenblatt: It could be. Anyway, at some point, the powers that be- very likely, Joel Moses, and I抦 not sure who else- passed the hat to MIT alums, with money- namely Tectronics and E&GG folks. And they raised a fairly sizable pile of cash. And they built some new buildings- Building 36 and 38, along Vassar Street there. And, to go into the new buildings, they had some money, and so they split their money in two and they bought one of Deck- a timesharing system- and they also bought what was originally going to be half a dozen List machines, at this- and we gave them an even considerably better deal. I抳e forgot what it was, but it抯 considerably less than ,000.00 per machine.
Q: These are for caters? Yes.
Richard Greenblatt: Yes, these are for caters. And there抯 a little story connected with that. It was sort of even a cheapened up cater. We tried to save some money on the disc drive and then those disc drives turned out to losers. We had to send them back. So we would up equipping the machines with five machines, with T-80 disc drives- and one T-300- instead of all the?
Q: Okay.
Richard Greenblatt: It抯 a set of 6- it originally would have been six machines, one with a T-300 and five with a data disc mumble, that turned out to be a loser. Now, it turned out the List machine, although it was used some, it turned out was not a great success as an instructional tool, and actually the timesharing system- by this time, Deck had Tops 20.
Q: Had a real timesharing system, that worked.
Richard Greenblatt: Had a real timesharing system. Well, they also, over the years, eventually got Tops 10 to a fairly presentable state.
Q: To a pretty good shape too. Okay.
Richard Greenblatt: Although it still had a number of fundamental problems. But that抯 right. At some point, BBN had developed a timesharing system called 10-X. Both
Q: Built your own paging boxes.
Richard Greenblatt: We built our own paging box and did brain surgery on the inside of the PDB-10 to install it. So, we wound up with three KA-10抯, that had ITS pagers. And BBN- and made, I抦 not sure how many things. They were a commercial company. So they were a little bit at an advantage to us. At some point, Deck realized that they were losing, sort of, and they said, okay, let抯 make a deal with BBN. And they made some kind of a deal with BBN where they took a bunch of BBN抯 people, to Deck- including this guy, Dan Murphy, who was- and they basically then worked on it for another year or so, and came out with TOPS 20.
Q: Okay. Which was based on 10-X.
Richard Greenblatt: Which was based on 10-X. And so that also had paging, and that was a good thing. Meanwhile, we also- the KL-10, however, had microcode. And so we succeeded- and I succeeded, in part, on one particular occasion, I actually- this guy, Dave Moon, did the main thing. But, in fact, it抯 an interesting story, which you probably don抰 have time to go into to- I went out to Deck one time and convinced them that they should give us the source of the microcode, which was incredibly proprietary, at that time, but- so that we could implement- hack the paging and implement an ITS pager, in KL-10 microcode. And so we succeeded in doing that. And so our MC machine had an ITS pager- and it was a Deck machine with no hardware change, but just microcode change.
Q: Very good. Very elegant. And ITS was still being used in the?
Richard Greenblatt: Oh, yes.
Q: In your lab?
Richard Greenblatt: Yes.
Q: In the- what do you call it?
Richard Greenblatt: Yes, so anyway we got the microcode. And Dave Moon eventually took that over, and Dave Moon was the main honcho for the MC machine. I just did incidental things after some time. I was involved, of course, in List machine and other things. But, so, then-- we were talking about low overhead timesharing. Anyway, MIT raised this money. MIT built their buildings. MIT- the Tectronics people, in EG&G, funded this ECS computer system, which is the List machines, plus a Deck 20. And then finally, MIT also had the equivalent of low overhead timesharing, and a computer for EECS students. Prior to that time, they just absolutely had nothing.
Q: Wow. The Multix machines were still running at this time?
Richard Greenblatt: The Multix machines were still running. And their prices gradually went down a little bit. And there was something called the student information processing board- or SIPB- that became quite active and that some good people were involved in and did deliver some services to students. But still, the amount of resources that students could get and what they could really do, and so forth and so on, was very minimal.
Q: Nothing like the opportunities you had, even when you first got there with?
Richard Greenblatt: Well, because I got tied up with research.
Q: Because you got tied up with research.
Richard Greenblatt: And, of course, that continued to be the case. Many students did get tied up with research groups, and they could do things. But, if they didn抰- if they were just in the regular main floor of MIT courses, then they could?
Q: They were really stuck.
Richard Greenblatt: They were really out of it. And, actually, just to touch on a thing we discussed earlier, I dropped by Tom Knight抯 office, just a few days ago. And one of the conversations came up is that these days it抯 actually illegal- the Federal government being concerned about use of research money to subsidize teaching or something, has now decreed that absolutely no classrooms can go on in research buildings. So, for example, that-
not only does that Dreyfus building cost a fortune, but the classrooms that it has, if any- I think there抯 one or something- cannot be used for an MIT class, because that would be using research money to subsidize education, or something. So it抯 an incredible story. And it comes up now with this biology thing. He has this thing, BioBricks, and he抯 looking for MIT to try to teach his BioBricks thing. And he抯 at a loss, sort of, because he抯 not in the biology department, with access to the right kind of classrooms, really, via the normal chain of events, sort of. So here he wants to teach a biology class, but he doesn抰 have any place to teach it because he can抰 do it in the Dreyfus building, because it抯 illegal, and he can抰 do it anywhere else either.
Q: He can抰 find another place to?
Richard Greenblatt: This is not totally insoluble. I think eventually they made me- but it抯 the kind of thing that goes on. Anyway, so, that was a fairly major thing, finally, when MIT did have some sort of a student _________.
Q: So now there are a bunch of these List machines, out there, and students can get at them. So what happens next, in the story?
Richard Greenblatt: Well, what happens next is, having made 30 of them, and having a pretty good sized demand, from people that are at a number of different places across the country, and it being sort of ARPA having kind of pumped up to their- to everyone抯 eventual regret, I guess you would say, what AI was going to do for people. It became obvious that there was going to have to be some companies. And so, I was in favor of basically spinning off a company, and having it self-financed. In other words, that we actually had quite a number of people who were willing to pay, up front, 80 grand, for a cater, essentially. And we had a cater production facility, involving this contracting thing and so forth.
Q: Yes. You actually had a fairly good system, all tuned up.
Richard Greenblatt: That抯 right. We had a system tuned up. And, so essentially we did do that. LMI initially- I initially started with just a incidental ,000.00 of my own money, and that was just printing brochures and stuff- not really very much. But then I hooked up with this Steve Wiley guy- the son of Frank Wiley, of Wiley Laboratories. And basically his father kicked in a hundred grand, I think it was. And essentially that was enough. We got a space on Blackstone Street. However, basically, Nosker, and eventually other people, decided this wasn抰 enough, and that they really should get regular venture capitalism. And so they got a guy involved, this Robert Adams, who had been at SDS, working with Max Pilafski [ph?], or whatever, and basically started throwing money around- what at the time was probably fairly small amounts, in some sense. But, in the particular situation, at that time, it was fairly large amounts. So, ultimately, we had a split. By this time, Nosker- I had lost confidence in Nosker.
Q: Now was Nosker originally involved in LMI at all?
Richard Greenblatt: No. Nosker actually had been at the AI lab and had had a falling out at the AI lab, and had left the AI lab, and had been moved to the West coast and involved in another company for 6 years or so. But he would occasionally?
Q: What was that company? Do you remember?
Richard Greenblatt: Yes, I do. It抯 a company called Perceptron, or something like that. Apparently they had some kind of a manufacturing thing. And it was, quote, a success. In terms of companies, it managed to achieve income and to sort of be self-sustaining, at some level, in some ways. Anyway, I wanted to do this bootstrap thing. They didn抰. And there was a number of other differences, shall we say. So, what ultimately wound up happening was, the two companies split out. One was Symbolics, the other was LMI. Both licensed the software from MIT. The majority of the original team, I抦 afraid, went with Symbolics. A few people stayed at MIT, and the few people that stayed at MIT, sort of actively helped LMI- in particular RMS- Richard Stallman, who you may have heard of. Anyway, also a number of the people who were actually involved in this production of caters. That was one of the things was Symbolics was planning to not deliver the cater- essentially what Symbolics did was to take the exact design of the cater and to turn it into printed circuit cards.
Q: Redesign it for higher volume commercial production.
Richard Greenblatt: Well, supposedly.
Q: Supposedly. That was their- that was what they said.
Richard Greenblatt: That抯 right. And so the machine remained compatible. You could actually take a disc pack off of one and put it on the other and they were?
Q: So they really were the identical design?
Richard Greenblatt: There was a identical design, but it was- this is the LM-2 now.
Q: Yes.
Richard Greenblatt: So, that抯 right. So LMI made exactly the same cater, precisely as at MIT. And we ultimately made about 35 of them, I think. We got a number of orders, on this money in advance basis.
Q: Wow. I just need to interrupt, just for a second. You said an LM-2. What changes did you make between the original cater, and when did it sort of transition?
Richard Greenblatt: Okay. MIT had the cater. Symbolics, when they spun off, called theirs the LM-2. This is this ________ second version of the same exact logical design.
Q: Okay, I抦 sorry.
Richard Greenblatt: LMI did the same physical design, as well as logical design, and they called theirs the Lambda- let抯 see, the Lambda was later.
Q: I misunderstood. I thought the LM-2 was a second version cater that LMI.
Richard Greenblatt: No.
Q: That you were building at MIT, at some point.
Richard Greenblatt: No.
Q: No. Okay, I got it.
Richard Greenblatt: So the LM-2 versus cater was 100% compatible. So both companies spun off and both companies initially had some success. But then- and we did bootstrap the company. We essentially had a positive cash flow, for a period, even without any more capital. But, at some point, it became clear that we were going to need more capital. And Texas Instruments had bought some machines. And we didn抰 want to be building so much stuff anyway. So, it turned out that there抯 a company in Irvine, California, called Western Digital. And there抯 a guy named George White, who was a MIT engineer who was working there. He had been involved in various standardization activities, particularly something called the new bus, which later formed the basis of Apple machines, for a period. That project actually had had its genesis also at MIT, through another long history that was going on in parallel. There抯 Steve Ward- another group at MIT, in the same building, on the 5th floor- had decided you should have a workstation with a 68000. And so they designed such a thing and they attempted to get it built, by subcontracting it out. They did it a couple of times and it flopped, because the execution of the company was bad. On about the third try, they subcontracted with this Western Digital Company- a group at Western Digital. And there was a long period where the processor of a 68000 was available, but there was no- you couldn抰 buy a workstation with a 68000 in it. It was quite a- anyway, a frustrating period, in a lot of ways, for a lot of people. Anyway, at some point, these guys developed this standard bus, which was a fairly innovative bus for its time, called the new bus. At LMI, we were saying, okay, we don抰 want to be in the hardware business, we don抰 want to make all this hardware. What would really be good is if we can buy these new bus machines from somebody, and just make our processor- or maybe not even make our processor, after awhile- and do software, and so on. So, we investigated that, and I got involved. This new bus also had a standardization committee, and I went to a number of those meetings. And, however, although Western Digital抯 execution was a little better, it was still not sufficient to make the grade, and they sold a few machines, but practically none. So, they were about ready to lose. Meanwhile, here was Texas Instruments. So, essentially, what wound up happening was a 3-way deal between us, Texas Instruments, and Western Digital, with the money essentially coming from Texas Instruments. And, basically, that kind of put money into this group at Western-- the group was actually transferred to Texas Instruments, from Western Digital. But initially they just moved across the parking lot, to another building, in Irvine, where they started making these new bus machines, with an additional trial, and with us as a potential customer, as well as, hopefully, other people. But, as it worked out, there weren抰 any other people, really, or the machines weren抰 good enough to really make the grade that way. So, somewhat later, there was another round of- another deal making- whereby we raised considerably more money
Q: In the same box.
Richard Greenblatt: In the same box.
Q: And they could talk to each other too.
Richard Greenblatt: They could talk to each other but they really didn抰, except that they- all the List machines talked to each other over a local network. They all had Chaos Net or Ethernet variety, so they could talk that way. And that was good enough for the talking they needed to do, essentially. But they did share a disc drive. They shared sort of the overhead of the box. And, by this time, it became possible to make a 64 meg memory board, all in one memory board, PC, and it had SIMMs kind of soldered in. They weren抰 plug-in SIMMs, but they were kind of solder in SIMMs. And that enabled you to have 8 megabytes of memory for each of your two machines.
Q: Yes. Very nice.
Richard Greenblatt: And that worked well, and that was popular, and we sold quite a few.
Q: Yes, I would think so.
Richard Greenblatt: So, that was the Lambda, and that went on. Essentially, as part of that deal, we licensed Texas Instrument to a bunch of the software. And I went down to Texas Instrument several times, and basically brought them up to speed on a number of things. And they eventually went in, in competition, and came out with something called The Explorer Series. And we didn抰 really want to be in the hardware business. But, unfortunately, by this time, the company had been lost control of. In other words, the infrastructure to build this thing was this whole building full of people. Meanwhile, the control of the company was in the hands of these venture capitalists and their new presidents, who wanted to go in the hardware business- and basically he was out of control.
Q: And now they had two competitors- Symbolics and Texas Instruments.
Richard Greenblatt: That抯 right. So, at some point, the company got into trouble and it eventually declared bankruptcy. It also had a major other product, that it had invested in all this time and brought along, which was a process control product called PICON. And that was viewed as being a very valuable thing. And, right at the critical moment, unfortunately we got stabbed in the back. The guy who was involved in this decided to steal it- it was a guy named Lowell Hawkinson- generated a company called Gensym. Lowell Hawkinson had worked for Fredkin, actually, many years before. So, anyway, we later sued them. There was a settlement and there was a nominal payment. But, of course, by that time, there wasn抰 anything anybody could do, besides pay lawyers, so that wasn抰? But, anyway, that was- I don抰 know that the deal necessarily would have happened anyway. But there was various deals that were supposedly going to happen. And the last straw, shall we say, was that these Gensym guys decided to steal PICON. And so that caused the remaining things. There were several other players involved in the picture, at this time. There was something called MCC, which is the microcomputer consortium in Austin, who was a big user of List machines, and to whom we had sold a big bunch of List machines. Anyway, what happened, as far as I was concerned, by this time, was that I had pretty well- I was pretty well tired of it anyway. But I had an employment contract. And Ward McKenzie- the guy who was the president- the first president of LMI, brought in by Venture Capitals, from TI, was a guy named Frank Spitznoggle [ph?]. The second one, who?
#### End of Tape 102630623_5greenblatt.mp3 ####
Richard Greenblatt: 卼o a situation where LMI had a large manufacturing plant in Andover in a new building. They had a research and development facility in a fancy building called a Thousand Mass Avenue in Cambridge, which is where I was basically working and the software, this Pikon Group, and other software people. And the company was headed for the rocks. So?but I had an employment contract so I wanted to quit. However, what I was going to do was consult for MCC and by this point I knew the MCC people and they had a bunch of LAMDES [ph?] and they wanted LAM to support and they also?there might have been some artificial intelligence. I know there was?what抯 his- the Stanford guy抯 name, I forgot for a moment I抣l remember, anyway he was there doing his supposedly AI system and so on. So McKenzie decided that he didn抰 want that, he didn抰 want me consulting at MCC in competition with them. So we reached an agreement that well, okay if they wanted employment contracts they could just pay me. So basically for about six or eight months, maybe even a little longer, I was in a state of suspended animation where I was getting my salary but I was not actually doing anything for LMI. I had another couple of guys that I was going to form this consulting company with so I decided since I wasn抰 really planning to get this money I would just support them out of that salary. I had a pretty high salary. And also at a Thousand Mass Ave there was some space available. So?and at this point since the company was contracting so much they didn抰 have room for all their machines so they physically had excess LAMDES that they?and furniture and stuff and I- it actually was quite a big space and I just had it pretty well stuffed with LAMDES and amazingly enough there was electrical service and was able to plug them in and use them. So basically I had that space and we sort of had our little company but we couldn抰 consult for our major customer. But we just sort of did that. Well, the meanwhile the company went ahead and filed for bankruptcy and a receiver got a point and one thing another and then it turned out that a friend of mine, a guy named Guy MonPetit [ph?], he was a Canadian, and he was a?an associate of Seymour Papper [ph?] and involved in the logo children program and so forth. And basically he had become our Canadian distributor and was pursuing various schemes in Canada and so forth. Anyway he suddenly appeared on the scene with money and, to make a long story short, bought the assets from the bankruptcy court and installed me as president of a reconstituted company called Gigamoss [ph?] systems and his- it was called Gigamoss. Now, but I was?you know, I was pretty mixed feelings about this whole thing but- and I decided to go along with it just feeling I- sort of a duty or something I don抰 know what. Anyway by this time there had?another processor- a new processor had been designed, not by me, called the K-Machine. It was designed primarily by a guy named Pace Willerson [ph?], who抯 a good engineer, and a few other people. Well, they had been working on it during this inneregment [ph?] when this company was going through this thing and so forth...
Q: They抎 worked on it at LMI?
Richard Greenblatt: Yeah. LMI, right during this thing. And we- I wasn抰 particularly interested in it but then it got- it kind of landed in my lap after this buyout and this thing. So it wound up that we moved the company down the street to 675 Mass Avenue from a Thousand Mass Avenue that we kind of reconstituted a smaller company. Meanwhile the manufacturing and another large piece of company had moved from Andover to the Lowell Mill- Mill and Lowell and much lower rent space but still quite a lot of it. And there- just an incredible amount of stuff. They, you know, disposed of equipment, it was just totally incredible how much furniture and, you know, there was just long lines, you know, they?this mill had, you know, long bays and wired cages and these bays were just filled with expensive furniture, you know, that had been bought by this company and so forth and so on.
Q: With venture capital money?
Richard Greenblatt: With venture capital money.
Q: Cause it never made a profit did it?
Richard Greenblatt: No. And so they sold all that and they sold this and that and so?
Q: Hey, million will buy a lot of furniture.
Richard Greenblatt: That抯 right. They also bought- they bought a testing machine, a PC circuit testing machine thing for I think it was at least three quarters of a million or something. And they bought several other major things and so on. Anyway I was not only president of the entire company including that piece. The Pikon guys having left we had these customers and we still had Pikon and so some other people that had been kind of on the border of the Pikon thing, including a guy named George Garet [ph?], and so forth sort of stepped forward and offered to take?to sort of operate that part of the company. So that抯 right. So basically that happened out of this Lowell facility and meanwhile in Cambridge I worked on trying to get this K-Machine processor put together. Well, basically eventually I sort of did but eventually, you know, several of the people?well the primary person was Pace Willerson just- with good reason he just said this company doesn抰 have a good environment, I抦 leaving. And he left pretty early in the game. Another couple of people hung around for a while and?but it was a terrible working relationship and they eventually left too. And so I just kind of slogged forward and eventually did some- more or less get this K-Machine thing going. However, at that point?
Q: At what year are we maybe?
Richard Greenblatt: Let抯 see, this whole LMI bankruptcy thing was in ?6 I believe and?
Q: So ?7 or ?8?
Richard Greenblatt: Yeah and then in ?8 another stroke from above, incredible thing happened, and this happened in Canada. Turns out that this Guy Monpetit had made a deal with a Japanese venture capitalist- I- let抯 see, remember his name. Anyway he was very rich from money in Japan and he was interested in logo and children抯 computer language. And Monpetit had been involved in that. I had been tangentially. I mean I had been a consultant to the logo group and involved in it but as all this other stuff happened I kind of faded off to that to a pretty low degree. But anyway, turns out that Monpetit had made this arrangement with him to- involving buying a major property in Canada, near- outside of Montreal that involved essentially a- Hoffman-LaRoche had developed a major drug research campus outside of Montreal and then Canada changed their laws or taxes or something and Hoffman-LaRoche decided not to through with it. So here was this humungous eight story, fancy building, beautiful sitting in St. Catherine Eglise [ph?], or something a few- 20 miles or so outside of Montreal and being completely unused. And so this deal from- Suru [ph?] was the Japanese guy抯 name. So anyway Monpetit also had this small silicone design company, IC design, called Silicard [ph?] that consisted of a few hackers and they had some list machines and they were based in downtown Montreal. Well, by means of this deal Suru put through a fair pile of money such as I think million or something. Monpetit used that money to number one, buy this unused Hoffman-LaRoche property; number two, buy the assets of LMI out of bankruptcy court; and number three, also take control of a company called LCSI, Logo Computer Systems, which was a Canadian company that had?that he had been involved in earlier and that had had its own long story and was involved in this children抯 language logo thing. Well, so that was a deal and that went forward. But then at a totally incredible thing happened in Canada wherein some Canadian lawyers managed to get Suru抯 ear and to feed him incredibly false stories that he was being taken advantage of and basically it started a big lawsuit up there and fundamentally caused the whole thing to collapse. So we then- and they were full of stories about all this money having been taken to the US and so forth. So here was Gigamoss Systems of which I was president but which I didn抰 own. You know, it was subsidiary of Gigamoss Canada and there was essentially this seizer action that went on there. So basically the lawyers of Suru, who were tapped into incredible amounts of money, generated some kind of an international guarantee that they抎 say okay, here抯 this LM- Gigamoss Systems we抮e taking control of it and we own it and we抣l pay whatever expenses are involved in doing this. So it was sort of like a bankruptcy except instead of being in bankruptcy court it was actually in federal district court in Boston and it was an international thing. You know, it involved the international guarantee that they made in order to persuade the US court to allow them to do this in effect. So you then got auditors here in Boston who had no motivation to do anything except run up a completely blank check for- and they basically did and they?and anyway so they went through the motions of trying to sell the company and doing this and that but, of course, they were just charging time at 0 and up an hour and then, you know, just incredible amounts of this and that. And no matter what they did they were tapped into this unlimited supply of money in Canada essentially as a result of that. So essentially that was the end of Gigamoss and that was about ?8. And- so then well I mean there抯 more story, I don抰 know if you want to go further with this. But just to make a short thing we can say that so then what I did was I sort of went back around MIT and I saw that the fact that MIT抯 phone system was incredibly screwed up and I generated some technology called?eventually called Omni Phone Technology. I applied for an SBIR, I got?small business innovative research grant, got that and then there抯 a deal whereby that funds you for six months and then you apply for phase 2, got phase 2, so I got just less than a half million dollars and proceeded to develop this technology. Unfortunately again it involved standards- international standards, it involved modem things and one thing and another and ultimately it was not really successful in doing that up. But it抯 still something which is sort of necessary to this day and hasn抰 really been achieved to this day. But I got some patents on it. But anyway finally approximately at the end of that my mother who had moved to Boston in ?8, much earlier, well in ?0 she fell and she broke her hip and so I essentially became responsible for her care at some point about then. And then she began to have a degenerative brain disease and so she required more care. So essentially I worked in the independent researcher mode and also kind of taking care of my mother for a number of years. And then finally my mother went to a nursing home and I抦 continuing to be an independent researcher.
Q: Independent researcher and her caregiver in a sense too.
Richard Greenblatt: Well, I visit her everyday in the nursing home so I抦 not an official caregiver but I抦 a something or other. But yeah, so that抯 that. And I抦 working on interesting projects which is a whole other story.
Q: I had a couple of questions. When LMI spun off and you started that was your partner the CEO? You clearly were the lead technical person.
Richard Greenblatt: Yes. That抯 right, he was the?Steve Wiley [ph?] was the CEO, that抯 right. I was chief technical officer.
Q: And then when venture capital arrived?
Richard Greenblatt: I continued to be chief technical officer and member of the board but, you know, the control of the company just progressively obviously went away.
Q: Where did you get your initial venture capital?
Richard Greenblatt: Well I guess there were three major rounds altogether. First is Steve Wiley, which is sort of not- that was just initial. Then there was Steve Wiley抯 father plus a company called Genesis Capital involving one guy from Silicon Valley and one guy from Washington. I may have their names if they would mean anything to you although I抦 not sure if I can find them right now. Let抯 see it was Jerry somebody. Anyway they- we raised I think a fairly small amount; I think it was 0,000 or something like that. Then there was another round where we raised from a bunch more venture capital where including new enterprise associates and a fairly long list of venture capital funds. And as part of that a guy named Frank Spitsnaugel [ph?] was brought in who had been a right hand man to Jerry Judkins [ph?] or something at Texas Instruments- who was at Texas- uh?
Q: The NEA were the lead people?
Richard Greenblatt: They were pretty prominent.
Q: You remember who represented them?
Richard Greenblatt: Yeah, I do actually. One guy is on Louis Rukeyser occasionally. Of course Louis Rukeyser is off- his name is Brian Rogers. And he kind of came in later. The other guy Art something- Art Marks right. Art Marks was the first guy
Q: Okay, I was just curious.
Richard Greenblatt: So yeah. At that time I think that loss was probably the biggest loss that New Enterprise Associates had had up to that point when they eventually lost their money.
Q: And then you said there was one more after a while.
Richard Greenblatt: And then- yeah, then there was a mezzanine thing and most of those same investors plus a bunch more. And at that point they brought in this Ward McKenzie guy who at the time was very sought after and, you know, for whatever reason people thought that just because he抎 almost replaced Ken Olson he was- somehow knew something or whatever it was that抯 right. And then after- actually then another round was that just as LMI was getting in trouble and so forth it turned out that Data General wanted Ward McKenzie and so there actually was a transaction where Data General actually paid some money into LMI for Ward McKenzie抯 contract and Ward McKenzie went off and was- I抦 not sure he was, you know, chief operating officer or something. He was essentially a senior guy at Data General.
Q: I sort of remember that now but I think I was long gone from Data General. I left in 1980 so it was many years later.
Richard Greenblatt: But- and he made some attempt to hire me at that time but I was not interested in being hired. So I didn抰 do that.
Q: And this is sort of now LMI抯 all ready in some trouble. Did they ever raise another?a last round or was the mezzanine sort of the last?
Richard Greenblatt: They tried to and it was pending when this Pikon/Jenson walkout occurred and that?and then that essentially sabotaged that.
Q: Some of the assets walked out the door and pretty hard to raise money if you?
Richard Greenblatt: Yeah. One of the main investors was a company called General Signal Corporation and- who was a primary player in some of this stuff. And the guy at General Signal, whose name I don抰 quite remember right now, was a senior guy there and their interest was Pikon. So they were interested in this Pikon as being a process control thing and so forth and that was the primary part of the deal that they were interested in and so when that went away?
Q: Well unless you have some more things you抎 like to?
Richard Greenblatt: I think we抳e pretty well covered it here and I think we抮e just about out of time too.
Q: Thank you very much for doing this, I really appreciate this. And the Computer History Museum thanks you too.
Richard Greenblatt: The tournament that was played April 22 to 23, 1967. So this is tournament number three.
Richard Greenblatt: And this was kind of- this was the first tournament at which a computer had a positive score and basically as I said it was the third one ?in the first one we drew one game and lost all the rest, in the second one we won one game and lost all the rest. In this one we played only four games but won three and drew one I believe. And so this then shows the games and the analysis and I just looked over it last night again to just see what went on and what was going on here and so I did kind of play over the games and just?
Q: Can I see how the analysis looks?
Richard Greenblatt: Yeah. Here you have?here抯 the start of the game when the parameters are set up. And here抯 the portion of the game is playing in the book. So it made?looks like about four or five moves out of the book. And then here it- once it抯 out of the book each page here shows the position, the list of legal moves and their plausibility, and then the results of the analysis. This number here is the evaluation. Zero is an even game, positive is good for white, negative is good for black. And so?
Q: These are evaluations?
Richard Greenblatt: Well those are cutoffs. This minus the thing?this- the machine is playing white so it抯 trying to find the most positive number here. And so in this case the most plausible move is the one it抯 going to play. And here is the analysis. You know, it抯 going to play bishop to queen night 5, it抯 expecting bishop to queen 1, bishop takes bishop, knight takes bishop. And then it- having looked at this line it then goes back to that position and searches a little bit deeper in this thing where it says okay, well after that if we do another small search from there we get pawn takes pawn, knight takes knight.
Q: Could you hold that? Could you just go over the first two pages so I can be point- I抦 just going to pick up the camera. Do you have time to do that, just for a second?
Richard Greenblatt: Well I have time but I don抰 know how much you抮e going- I mean we can get out a chessboard here and get into this position and stuff.
Q: I think it抯 interesting how you read this because I抦 going to ask you about whether you抎 be willing to?
Richard Greenblatt: Here抯 a list of legal moves with their plausibility. So these are moves which are scored and then this is the analysis. So it starts with the most plausible move, it expects this reply, this reply and this reply. That抯 step four. In this case there抯 no further search deeper necessary then. Then it comes back to the top level and says okay, well that抯 interesting, if we go to this end position and do another short search what do we get. And so it goes there and gets two more plies here. This number here is the time and seconds so this whole move?it will print here. Well it looks like it doesn抰. Sometimes it prints a total- yeah, there抯 the time, 54 seconds. So at this level I think you had about two minutes per move.
Q: Okay, you had two minutes to work on it.
Richard Greenblatt: Something at a fraction but something like that. So anyway that抯- that抯 one move. The machine plays bishop to queen knight 5 check and then the opponent replies bishop to queen 2 and here抯 the next move. Start all over again. The machine eventually won this game so let抯 see we get to the end here. Let抯 see we抮e playing white, move 15, okay, here抯 our tag. And yeah, you see here at the end the numbers are getting substantially positive for white which means that white抯 winning or seems to be winning so that抯 good. So that抯 a game and there抯 a total of four ga?
#### End of 102630624_6greenblatt.mp3 ####
CHM Ref: X3056.2005 ?2005 Computer History Museum Page 2 of 56
国际象棋程序设计：数据结构 编辑本段回目录
Fran?ois Dominic Laramée/文
上个月我简要介绍了象棋程序设计中所需要的知识，其他信息完全的双人游戏也是一样的。现在我们将讨论一些细节——棋盘的内部表示方法。
在棋盘表示方法这个理念上，近三十年内没有多大发展，你可能会觉得很吃惊。它的发展需要智慧的推动，很早就有人提出过绝妙的方案了，但同时也受到制约，因为这些方案需要数据结构的支持，某些数据结构至今还没有实现。
尽管如此，我还是会介绍三种数据结构，尽管它们并不是必需的，但是对于提高你的下棋水平是大有帮助的。其中两种可以加快思考速度(但是其中一种需要无限多的存储器)，另一种可以加快着法产生速度。【译注：分别指后面要提到的换位表、历史表和着法生成预处理的数据库。】
在我们继续讨论之前，我提一句格言：“无论是象棋还是其他游戏，你通常使用的数据结构，应该是能达到目的的最简单的数据结构。”然而象棋程序设计者提出了很多技巧，它们能让程序运行的更快，其中相当多的还适用于其他游戏。如果你对你要设计的游戏不很了解，而且手头的资料很有限，那么你应该好好掌握我所提到的这些技巧，你可以把这些技巧试验到你的程序上。
基本的棋盘表示
在上世纪70年代，个人电脑的存储器是稀有资源，所以棋盘表示得越紧凑越好。很多人会很自信地说：用一个64字节的数组，每个字节表示棋盘上的一个格子，用一个整数就可以表示格子的位置了。(任何棋盘的数据结构都必须用一些额外的字节，来记录吃过路兵的目标格、王车易位权利等信息，但是这里我们暂且忽略它，因为这些因素可以独立处理，而且处理方法大同小异。)
后来又流行一些更优化的算法：
1. 早期的SARGON【可能是一个象棋程序】扩展了64字节的数组，在它的外围加了两圈“虚格”，并在这些格子上作了非法标记。这一技巧能加快着法产生的速度，例如象在走棋时会延斜线滑行，直到走到虚格上才中止。这样就没有必要用复杂的运算来预先判断象到达的格子是否会超出存储器上的表示了。第二圈虚格是给马用的，例如位于盘角的马试图跳出棋盘，这就必须用两圈虚格来保护。
2. MYCHESS用了相反的方法，它只用32字节表示一个棋盘，每个字节代表一个棋子，例如代表白方王、黑方王翼马前兵【英文居然是black King's Knight's Pawn，一开始让我大惑不解】等，它存储的信息是棋盘上的位置，或者已经被吃的标记。这种技术有一个缺点，它无法描述由兵升变而来的其他棋子(同时这个棋子在棋盘上还有一个)。在后来的版本中，这个缺点被修正了。【其实这并不难办，一个字节取值从0到255，通常255表示该子已被吃，从0到63表示该子在棋盘上的位置，兵通常是升变成后的，那么从64到127可以表示该子已经升变为后，如果升变为车、马或象，则需要额外的字节来处理。】
如今，这种极端吝啬的数据结构只可能出现在掌上电脑、手机或电视机机顶盒上，它们的80~90%的资源都被操作系统占用，没有额外的空间给游戏使用。但是，某些游戏缺别无选择地使用这种方法【我也不知道什么游戏非得使用这种方法不可】。
想更多地了解以前的象棋程序，可以看一下David Welsh在1984年注写的《计算机象棋》(Computer Chess)一书。
位棋盘
用一个数组来表示棋盘，对于很多游戏来说，就找不到更好的办法了。但是对于像象棋、西洋跳棋之类在64格棋盘上的游戏来说，有一个高明的技巧——“位棋盘” (首先由苏联的KAISSA制作组提出)，在上世纪60年代末诞生了。
KAISSA是运行在64位处理器上的程序【我很怀疑当时就有64位处理器，或许当时处理器字长的概念和现在的有所不同】。“64”恰巧是象棋棋盘格子的数目，所以这就有可能让一个字来表示一个棋盘，以判断某个格子的状态是“是”或者“非”。例如，一个位棋盘可以回答棋盘的每个格子“是否有白子”。【把位棋盘运用到中国象棋上，这是我将要进行的一个课题，中国象棋的棋盘有90个格点，可以用3个32位的字来表示。】
因此，一个完整的局面需要用12个位棋盘表示：白兵、白车、黑兵等等。再加上两个用来表示“所有白子”和“所有黑子”的位棋盘，以加快运算速度。【其实只需要8个就可以了，同一兵种(不管黑白)用一个位棋盘，一共是6个，再加上代表“所有白子”和“所有黑子”的。做得更过分的是，有人把象和后并作一个位棋盘，车和后并作一个位棋盘，这样又可以减少一个。如果要表示白方的象，只要“所有白子”、“所有车或后”的补集(用“非”运算)、“所有象或后”这三个位棋盘作“与”运算就可以了。】或许你还想用一个位棋盘表示被某种子力攻击到的格子，诸如此类，这些位棋可以盘灵活运用在着法产生的运算过程中。
位棋盘之所以有效，是因为很多运算可以转化为处理器的一个逻辑指令。例如，假设你需要确认黑王被白后将军，如果用简单的数组来表示棋盘，那么你需要这样做：
1. 首先找到后的位置，这需要从64个字节中一个一个地找；
2. 然后在后所能走的八个方向找，直到找到王或者找到后走不到的格子为止。
这些运算是相当花费时间的，如果后碰巧是在数组的最后一格，更糟的是，将军只会发生在少数情况下【这种运算纯粹是白费时间！】。
如果用位棋盘，那你只要这样做：
1. 载入“白方后的位置”的位棋盘；
2. 根据这个位置，从索引数据库中找到被后攻击的位棋盘；
3. 用这个位棋盘和“黑方王的位置”的位棋盘作“与”运算。
如果结果不是零，则说明黑王被白后将军。假设被后攻击的位棋盘是储藏于存储器中的【这是上面提到的第二步的前提】，那么整个操作只需要3到4个时钟周期【通常计算机执行1条指令需要1(算术或逻辑运算)到2(寻址操作)个时钟周期】。
【这里允许我发挥一下，原作所说的“从索引的数据库中找到”(即上面提到的第二步)，其实并非是简单的一步，对于后的每个位置，都有一定的攻击格子(从边线到中心依次是21、23、25和27格)，但是要考虑被别的子阻挡的情况，程序无法为所有的情况都作索引，最多只能对某条线(横线、纵线或斜线)的棋子占有情况作索引，这也需要28 = 256 种情况，再加后本身有64种位置，所以即使这样，数据库中也至少要保存256x64 = 16384个位棋盘。另外，横线、纵线和两条斜线的处理方法各不相同，横线只要作简单的“移位运算”就可以了，而纵线和两条斜线都要用到“位棋盘旋转”的技术，为了提高运算效率，程序的复杂程度是惊人的，细节可以参考《对弈程序基本技术》专题之《数据结构——旋转的位棋盘》一文，文中作者多次提示读者用咖啡来提神，以完成烦琐的程序。】
再举一个例子，如果在当前的棋盘上，你要产生白马的所有着法，那么你只要找到当与前位置相关联的“马能走到的格子”的位棋盘，并“与”(AND)上“所有被白方占有的格子”的位棋盘的补集(就是对这个位棋盘作“非”(NOT)运算)，因为马的着法限制仅仅在于它不能去吃自己的子。【国际象棋比较简单，而中国象棋中还有“绊马腿”的限制(还有象也是)，这种情况该怎样使用位棋盘，也是我将要研究的课题。】
如果你想更详细地了解位棋盘(也只是稍微详细一点而已)，可以去看看描述CHESS 4.5 (它是由美国西北大学开发的)的文章——Peter Frey写的《人脑和电脑的象棋技巧》(Ches Skill in Man and Machine)，现在至少已经出了两版了，分别出版于1977年和1981年。
值得注意的事，到今天为止，几乎还没有真正意义上使用64位处理器的个人电脑，所以位棋盘的速度优势是要打折扣的。尽管如此，位棋盘的技术仍是行之有效的。【苹果公司的Macintosh图形工作站据说是64位处理器，但不知有没有针对这种电脑的象棋软件。时隔4年，到我翻译这篇文章时，还没有什么别的64位处理器用在个人电脑上。因为毕竟没这个必要，除非你专门用它来玩国际象棋。】
换位表
在象棋里，有很多着法可以到达相同的位置。例如，不管你先走 1. P-K4 ... 2. P-Q4或者1. P-Q4... 2.P-K4，【这里K4和Q4可能就代表e4和d4，在实战中有这样的例子，1. e4 e6 2. d4和1. d4 e6, 2. e4是形成法兰西防御一种变例的两种途径。】最终局面是一样的。有不同的路线可以达到同样位置的，这种现象称为“换位”(Transposing)。【在中国象棋中，换位现象更为普遍，通常用成语“殊途同归”来称呼这种现象。】
如果你的程序已经对1. P-Q4... 2.P-K4产生的结果竭尽全力作了搜索和评估，那么你最好记住这个结果，以避免碰到1. P-K4... 2.P-Q4时再作同样的运算。自上世纪60年代Richard Greenblatt的Mac Hack VI问世以来，所有的对弈程序都会吸纳“换位表”这一技术，这就是原因所在。
换位表存储的是已经搜索过的结果，它通常使用类似于散列表(Hash Dictionary)的数据结构来达到最快的查询速度。在搜索某个局面时，结果(包括局面分析的值、搜索深度、最佳着法等)就存储到换位表里。当搜索新的局面时，我们先从换位表里找，如果已经有这种局面，那么就可以利用它，省去重复的搜索。
这种处理有以下很多好处：
1. 加快速度。在换位情况发生得很多时(特别是残局局面里，棋盘上棋子很少时)，90%以上的局面可以在表中找到。【在中国象棋中，这优势时更为明显，因为它的子力密度小，在开局阶段就有很多“殊途同归”的现象。】
2. 任意深度。假设你需要对某个局面搜索到一个指定的深度，例如4步(也就是两个回合)，如果换位表里有这个局面而且已经搜索了6步，那么你不仅可以跳过这次搜索，还可以得到比预期更精确的结果。
3. 用途广泛。通常每个象棋程序都配有“开局库”(Opening Book)，即包含一些常见局面及其最好的着法，这些通常是从已有的记载中提炼的【例如大师们写的“象棋开局大全”或“象棋开局手册”之类的书，而我本人更倾向于从大量对局记录中提炼的结果】。有了开局库，程序就不必在开局阶段就做傻事了【因为前人已经做过无数次的计算了】。既然开局库的操作过程和换位表是一样的(即搜索局面)，那么为什么不在棋局一开始就把开局库装入我们的换位表里去呢？如果这样做，即使棋局暂时脱离了开局库，后来又回到开局库里的局面【要注意，这个局面可以是棋局过程中出现的局面，但更多的情况是搜索程序推演到的】，那么换位表里保留了这样的局面，我们仍旧有机会用到它。
换位表唯一的缺点就是它贪婪地占用着存储器，无论如何它至少需要存储成千上万个局面，百万甚至上亿就更好了。如果每个局面用16字节【用最紧凑的办法至少也需要32字节(用MYCHESS这种吝啬的办法)，但是这里可以存放“散列键值”，下面会提到这个技术】，那么在存储器紧缺的时候这将成为很严重的问题。
换位表还有其他用途。
CHESS 4.5还用散列表来存储其他的局面计算结果【指下面提到的兵型、子力平衡等】，这些计算结果在多数情况下是不会变动的，例如：
1. 兵型(Pawn Structure)。散列表只存储兵的位置，这需要较小的存储空间，由于兵的移动不会很频繁，所以99%的兵型局面可以在散列表中找到；【国际象棋的局面分析中，需要考虑连兵、叠兵、孤兵、通路兵等因素，这些计算是相当费时的，而中国象棋就没有这个必要了。】
2. 子力平衡(Material Balance)，即棋盘上子力的相对强弱，它只在吃子或兵升变时变动。
在CPU速度不快的今天，这些方法或许看来用处不是很大，但是它们还是有指导意义的，一些稍微花费存储器的预处理可以节省相当多的计算。【因为寻址操作(特别指在若干M的大范围区域内寻址)所占用的时间要远多于简单的运算指令，如果哪天寻址操作的速度接近于基本运算了，那么这种技术将会对程序运行速度有很大的提升。】如果你仔细研究你的程序，或许你会发现这方面是值得改进的。
为棋盘产生散列键值
上面提到的换位表的结构，是由散列表来实现的，由此引发了以下课题：你如何快速而有效地从棋盘来产生散列键值(Hash Key)？
以下是1970年Zobrist的方案。
1. 产生12x64个N位的随机数(如果换位表能储存2N个局面)，并把他们做成一张表。每个随机数只跟某个位置上的某种棋子有关(例如H4格的黑车)，空的格子用零表示；【因为棋子总共有12种，棋盘总共有64格，所以是12x64个数，空的格子不用随机数而用0表示。】
2. 先把散列键值设为零；
3. 搜索整个棋盘，每遇到一个子，就在原来的散列键值上“异或”(XOR)它所对应的随机数，重复这个过程直到整个棋盘被检查一遍。
这个方案有趣的一面在于，每走一步，散列键值的更新都非常容易，不需要重新搜索整个棋盘。你知道“XOR图形处理”(XOR-Graphics)吗？某个图形用XOR运算作用到背景上，然后再作同样一次运算，不就又回到背景了吗？这里也同样这么做。【如果你熟悉图形操作中的技术，就不难理解了，原文把这个过程比作“位图”(Bitmap)操作，12x64个棋子的状态就对应12x64张位图，原先的散列键值被比作“背景”(Background)，只要在背景上作位图的粘贴操作(用的是XOR处理)就可以了。】举个H1格的白车吃掉了H4格的黑兵的例子，要产生这个散列键值，先XOR“在H1格的白车”这个随机数(把这个图形擦掉)，然后是“在H4格的黑兵”(把这个图形也擦掉)和“在H4格的白车”(粘贴一个新的图形，代表新的车的位置)【其实顺序是无关紧要的】。
用相同的方法，不同的随机数，可以产生第二个散列键值，或称“散列锁”(Hash Lock)【在英语中Lock(锁)和Key(钥匙)是对应的】，它是在换位表中存储的真正有用的信息。这个技术是用来检测冲突的，如果恰巧有两个局面具有相同的散列键值，那么他们具有同样的散列锁的几率是微乎其微的。
历史表
“历史启发”(History Heuristic)是“杀手着法”(Killer Move)【以后的连载会提到的，但为什么称“杀手着法”我也不清楚】技术的衍生技术。一篇研究性的文章是这么解释的，历史表用来保存这些着法，在过去的搜中非常有意义(因为使用高效搜索技术的而对它进行了很深的搜索)，这个着法今后还可能用到。历史表由一个64x64的整数数组构成【着法的起始格和到达格，共有64x64种组合】，记录每种着法的数值，当搜索算法认为某个着法很有用时，它会让历史表增加这步的数值。表中的数值是用来对着法排序的，“历史上占优势”的着法会优先考虑。
着法产生的预处理
着法的产生(即决定特定位置下那些着法是合理的)和局面的估计一样，是象棋程序设计中计算量最大的部分。因此，在这个方面的一点预处理会对提高速度大有帮助。
我个人喜欢Jean Goulet在1984年写的《象棋的数据结构》(McGill大学出版社)一书中提到的方案，它概括为：
1. 出于着法产生的目的，棋子的颜色是无关紧要的，除了兵以外，它只朝对面走；
2. 对于基本的子力和它的位置，有64x5=320种组合【指除了兵以外的5种棋子，根据上一条，这些棋子是不分黑白的】，黑兵有48个格子可以放(他们后面一行是走不到的，并且一到第八行就会变成别的棋子)，白兵也有48个格子；
3. 从某个格子沿某个方向，可以确定一条着法“射线”，例如，后从H3格朝北走，这就是一条“射线”；
4. 每个格子上的每个棋子，都有确定的几条射线可以走，例如位于中央的王可以朝8个方向走，而位于盘角的象却只有一条逃生的路；
5. 在开始游戏之前，要计算数据库里在所有格子的所有棋子的所有射线，假设棋盘是空的(即着法只受到棋盘边缘的限制，不受其他棋子的限制)；
6. 当你为某个格子的某个棋子产生着法时，朝每个方向搜索直到碰到棋子为止。如果是对方的棋子，最后一种着法就是吃子的着法，如果是本方的棋子，最后一种着法就是不合理的。
有了恰当的数据库，着法的产生就变成简单得接近于线性的寻找了，几乎用不着什么计算。整个事情就掌握在这么几千个字节里，只是换位表的一个零头。
以上提到的所有技术(位棋盘、换位表、历史表和预处理数据库)都会反映在我自己的程序中，当我写完这个连载以后就会发布出来。下个月我会详细介绍着法产生的方法。
Fran?ois Dominic Laramée，2000年6月
出处：www.gamedev.net
译者：黄晨 (auntyellow@citiz.net)
类型：全译
