Pensamientos profundos y aprendizaje autodidacta

por Frederic Friedel
14/12/2017 – La semana pasada, la empresa DeepMind volvió a salir en las noticias, afirmando que un sistema informático de nombre "AlphaZero" se había enseñado a si mismo a jugar al ajedrez a nivel magistral en plan autodicacta. La estrella de ajedrez Garry Kasparov este año ha publicado un libro que trata sobre el mismo tema y en el cual afirma que estaba convencido de que iba a haber sistemas que serían capaz de autoaprender. Frederic Friedel ha publicdo un artículo (en inglés) uniendo ambos temas en nuestra página de noticias en inglés. | Foto: cortesía de los autores, Garry Kasparov y Mig Greengard

Master Class Vol.7: Garry Kasparov Master Class Vol.7: Garry Kasparov

Inigualable compendio de la carrera ajedrecística del XIII Campeón del Mundo, con 9 horas de vídeo, entre otras muchas exquisiteces.

Más información...

On a personal note

Many years ago I was with Garry Kasparov for an event in London's Home House, and there we had dinner with a young lad, a former child prodigy in chess, one who had reached master level (Elo 2300+) at the age of 13 and captained a number English junior chess teams. It was an interesting encounter with the boy enthusiastically describing a computer game he was developing. After he left I said to Garry: "That's a cocky young fellow!" "But very smart," Garry replied. And we left it at that.

More than twenty years later I had occasion to contact him again. Demis Hassabis sold his software company DeepMind to Google, for half a billion dollars, and had subsequently produced a self-learning system that was mastering the very difficult game of Go. While searching for material I wrote to Demis: "To my chagrin I cannot find any pictures of you as a kid with Garry Kasparov, during our London meeting. Probably cameras had not been invented at the time. Do you by chance have pictures? Or recollections? You were 14 or 15 at the time, had just signed a contract with a company to produce a game." To which he replied: "I don't think I have pictures either, unfortunately, but yes I do remember it well. I was actually around 18 years old..." (he definitely looked younger). The Guardian mentioned this encounter in a report.

Last week DeepMind was back in the news — in thousands of mainstream news outlets — for having developed a system that learned to play super-human chess from scratch. Kasparov's book puts this into acute perspective, actually presaging the development of a self-learning chess system. If AlplaGo turns its attention to chess, Kasparov asks, "how long will it take before it can crush the strongest chess engines." Four hours apparently, as we now know.

Actually I should be reviewing Kasparov's book — I was at his side for most of the journey he describes. And I will come back to it in the future. But for today we have a special kind of review: Demis Hassabis discussing Deep Learning (link at the bottom of this article) with the author, in a very illuminating Google Talk. I have transcribed key passages and provided links to the specific parts of the one-hour discussion. You can scroll forward to the points of interest or click on the time stamps, which will open a new YouTube player starting from there.

Hope you find time to watch this important discussion and that you will enjoy it.

01:00 min – About growing up as a chess champion in the Soviet Union.

03:05 min – The first match against Deep Blue in Philadelphia (03:35). The watershed moment (04:02): "That was in February 1996 when I lost the first game of the match." – It had a record following on the Internet (04:32).

05:00 min – The second match: "I made many mistakes in preparation. The biggest was that I did not treat IBM as an opponent like Karpov, Anand or Short. I still believed I was part of a great social, scientific experiment. For IBM it was just about winning or losing."

06:00 min – Did IBM cheat? No, they just bent the rules in their favour. "And my preparation was quite lousy. Only a week before the match I realized how difficult this challenge could be" (7:35).

"In game five in the endgame I was slightly better, and everyone believed at the time that it was a brilliant escape by Deep Blue. Today, in 30 seconds to one minute chess engines like Stockfish or Komodo will tell you that the endgame was a draw, that Deep Blue made a bad mistake, and I missed a win" (9:40)

10:26 min – "I played two more matches with Deep Fritz and Deep Junior, both ended in a draw."

10:50 min – How do you think chess computers have changed chess? Do you think that it is for better or worse?

11:10 min – "The technology is neither good nor bad. It's happening, and we just have to adjust. It is now different, because the young generation of chess players learn very differently from us. I remember I had many books, every book was cherished, I had my notebooks, I recorded my analyses and treasured them. In the eighties they were top secret and a powerful weapon, like the magic sword of Merlin. Now when you look at it with a computer you understand it was a broken knife."

12:15 min – "When you look at young chess players there is such a difference in the way they approach the game, the way they look at the pieces. They point out mistakes, give a long computer line. But when I ask them why a move is wrong they don't understand the question. Their answer: 'Because the machine said so.' Somehow their minds are being hijacked by the power of the machine."

13:07 min – One of the reasons why Magnus Carlsen was so successful and still is a dominant force — and I remember this from working with him in 2009 and 2010 — is that he never looked at the machine as an ultimate source of wisdom. For him it was more like a calculator to verify his own understanding and evaluation of the position.

14:27 min – "The game has not changed, just the way people are watching it. Twenty, thirty, forty years ago the World Championship was a match of absolute quality. When someone made a terrible blunder it took time in the press center for grandmasters to whisper 'Mistake!' Today when you are watching games, like Carlsen-Caruana, you have thousands of amateurs shouting 'Mistake, mistake!' because of the machine evaluation has dropped. Some kind of respect has disappeared. But it has also added interest, because they don't have to be strong players to understand what is happening."

16:00 min – The process for maturing as a chess player today is much shorter. You have grandmasters who are 14, 15, who know much more today than Bobby Fischer knew forty years ago. Chess is a perfect match for the Internet — you can watch, you can learn, you can analyse, and dramatically increase the pace of learning and getting to the top.

16:29 min – Advanced Chess: it sounds ironic but you don't need a very strong player the get the best result of human plus machine combination. It sounds like heresy, but I would say you don't want a strong player. You need a good operator, a decent player who will guide the machine — not use the machine to back up his or her own ideas, to maximize the effect of the machine's play.

18:29 min – Kasparov's Law: a human plus a machine will beat a super-computer quite handily. It's all about interface, it's about empowering machines with our creativity. The result could be phenomenal.

19:39 min – Is Advanced Chess a blueprint for how things will go forward in other areas of life? "I cannot stand the doom and gloom predictions, dystopian visions of the Terminator. But it reflects Kasparov's Law. The first was about humans against a machine, Terminator 2 and 3 were about human plus machine plus better process beating a super-computer. These things are going to happen anyway. What is the point of trying to slow down a natural cycle. We have technology replacing certain elements of human activity. That's absolutely normal, that's called progress. There are still many things humans can do. We need to look for new challenges, new frontiers.

22:20 min – How far does he think top chess computers are from optimal chess, what is the top Elo rating that it would be possible for them to play? "If you look at endgame databases — now we have 100 terabyte seven-piece endings — we have a position that says 'Mate in 492 moves." I bet you that in the first 450 moves you will not see the difference. I don't know what this says about the game we play. The game of chess is an ultimate endgame with 32 pieces. I don't see any future in which a machine will play 1.e2-e4 and announce mate in 16,750 moves. But it's not about solving the game, its about winning the game. Machines can get better and better, the sky's the limit. Today I still think that Magnus with white on a good day will probably secure a draw against the machine. But winning against the computer today — it's virtually impossible. The level of precision that is required, the level of vigilance, it's impossible.

24:16 min – "We see improvements all the time. When writing my books, some of the games I analysed a few years later, with a new version of the same engine, I could see that some of the moves I treated as great, I had my doubts.

Audience questions

25:14 min – During the match against Deep Blue did people want him to win? "Most people who wanted me to lose were in the world of chess. I had been the World Champion for twelve years, and a lot of people wanted me to lose one day. Since I was unbeatable in human chess they pinned their hopes on the machine. But the atmosphere in the match was phenomenal: Newsweek had a cover calling it 'The Brain's Last Stand." When I won the first game a CNN commentator said: 'It's a Russian playing against an American machine — and I am rooting for the Russian!'"

26:56 min – What are humans so good at in chess that they only need to examine so few variations as compared to computers? "We don't have to analyse millions of lines — we couldn't — so we look for one or two options. How do we know that they are the best? I don't know! I simply know what it is. There are many patterns, and bringing them together will give you a bigger picture. That's what humans are good at.

30:15 min – Moravec's Paradox: Machines are very good at what humans are not so good, and the other way around. For Western science chess was the ultimat test for artificial intelligence. The expectations of the founding fathers of computer science that a machine beating the World Champion would it, the moment for AI. I have to say they were wrong — Deep Blue was as intelligent as your alarm clock!

31:12 min – Demis Hassabis: "If you have hand-built systems like Deep Blue was, you have to understand clearly enough what you are trying to codify. But for many thing we take for granted as humans we don't explicitly understand well enough how we do those things, so we can't codify it. That is why learning systems, like AlphaGo, might be more powerful, because they can learn from experience how to do the things humans do. You say in your book that up to now anybody who attempted learning systems fell short against the hand-coded systems."

32:30 min – Kasparov: "In 1952 Turing wrote a chess program, but there was no computer to run it on. He had to calculate the moves on a piece of paper. When I spoke at the Centenary I asked my friends from Germany to reconstruct it and put it into a computer. So you can actually play the Turing machine. It's pretty weak, but it's from 1952. They believed that the way to make machines play chess is not brute force but understanding. This concept failed very quickly because brute force kept coming, like an avalanche. So all attempts to come up with the concept of 'learning' failed, and by the end of the sixties early seventies the story was over. But now it seems we are going back to this notion, and maybe it will prove to be superior."

33:52 min – For Go that needed to happen. What is the difference between Go and Chess? "Difficult question, because I have almost absolute knowledge of chess and almost zero knowledge of Go. If you compare the strengths I think chess playing engines are relatively stronger than AlphaGo, in absolute ratings."

35:23 min – Hassabis: "I guess we will have to put that to test by teaching AlphaGo to play chess, right?" Kasparov: "That will be interesting – how soon AlphaGo can crush the strongest chess engines."

36:22 min – Last question (by Victoria, 11, daughter of the AlphaGo research scientist Thore Graepel): "If you could, would you play Deep Blue again?" Kasparov: "Hmmm. There are a couple of problems. One: I'm retired and don't play professional chess; and two: Deep Blue is dead. I wanted to play in 1998, and wish I had had a chance. But that is old history, spilled milk, water under the bridge. I played other computers, and as long as I was an active chess player I never ducked a challenge. My book begins with the story of me playing 32 chess computer in 1985, a simultaneous exhibition, and I won all the games. The most amazing thing was that nobody was surprised. That was the golden age: machines were weak, my hair was strong.


Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins 1st Edition

By Garry Kasparov (Author), Mig Greengard (Contributor)

Garry Kasparov's 1997 chess match against the IBM supercomputer Deep Blue was a watershed moment in the history of technology. It was the dawn of a new era in artificial intelligence: a machine capable of beating the reigning human champion at this most cerebral game.

That moment was more than a century in the making, and in this breakthrough book, Kasparov reveals his astonishing side of the story for the first time. He describes how it felt to strategize against an implacable, untiring opponent with the whole world watching, and recounts the history of machine intelligence through the microcosm of chess, considered by generations of scientific pioneers to be a key to unlocking the secrets of human and machine cognition.

Kasparov uses his unrivalled experience to look into the future of intelligent machines and sees it bright with possibility. As many critics decry artificial intelligence as a menace, particularly to human jobs, Kasparov shows how humanity can rise to new heights with the help of our most extraordinary creations, rather than fear them. Deep Thinking is a tightly argued case for technological progress, from the man who stood at its precipice with his own career at stake.

Available from Amazon.com and other outlets. ISBN-13: 978-1610397865 ISBN-10: 161039786X. Hardcover $19.04; paperback $16.99; Kindle $11.60.

 

He has made the most popular and best-selling DVDs in our inventory.

With his dynamic style, Kasparov had an epoch-making influence on the development of tournament chess at the end of the 20th century. The basis of his exceptional position in chess was extraordinary talent combined with hard work, enormous will-power and a boundless memory. He himself once characterised his style as "a combination of Alekhine, Tal and Fischer".

Kasparov won most of the competitions in which he took part. With his series of victories in 1999 he built up a lead of 80 points in the Elo list; the rating of 2851 which, until the advent of Magnus Carlsen, had never been equalled, despite Elo inflation. Kasparov also set the standard as an author of chess books, with the greatest attention being earned by his series on the world chess champions "My Great Predecessors".

Order these very popular Kasparov's DVDs in the ChessBase Shop

How I became World Champion Vol.1 1973-1985

Autobiografía profesional del mayor genio del ajedrez de todos los tiempos a través del análisis de sus propias partidas. Desde sus más tiernos comienzos. Descubra el mito desde su origen.

Más información...



Editor jefe de la página de noticias de ChessBase en inglés. Estudió Filosofía y Lingüistica en las universidades de Hamburgo y Oxford. Del mundo académico pasó al periodismo científico, produciendo documentales para la televisión alemana. En 1986 fue uno de los fundadores de ChessBase.
Discussion and Feedback Join the public discussion or submit your feedback to the editors


Comentar

Normas sobre los comentarios

 
 

¿Aún no eres usuario? Registro