Antidepressants and liver cancer are today some of the leading fields of research thanks to what the company Deep Mind (Google) learned with its AlphaZero program in chess and Go (a very popular game in several Asian countries) in 2017. It then created AlphaFold, which achieved one of the greatest advances in the history of biology in 2021: understanding how proteins work. As early as 1947, Alan Turing and Claude Shannon, fathers of computer science, chose chess as a field of experimentation. They were right, but they were unable to prove it: Deep Blue (IBM) beat Kasparov half a century later and laid the foundations for enormous scientific advances.
To understand why chess and go have contributed so much to the development of science, we must look at three numbers that, for the normal human mind, are associated with infinity. The number of possible different games of chess is 1 followed by 123 zeros. The equivalent in go (a board of 19×19 squares) is much larger. And so is the number of combinations of amino acids in a protein (an essential element for life). Let us add a fourth number to better understand what we are talking about: the number of atoms in the known universe is 1 followed by 80 zeros.
Deep Blue stored in its memory millions of games played since the 16th century, when one of the best chess players was the Spanish priest Ruy López de Segura. Based on this database, the program could calculate up to 200 million moves… every second.
Just ten years ago, this enormity led even experts to think that science was very far from unraveling the structure of proteins. One of the great Spanish references in artificial intelligence (AI), Ramón López de Mántaras, confirms that “the experience with AlphaZero in chess and Go was very useful to develop AlphaFold”, although he also believes that “the same success could have been achieved with another way”.
Chess masterminds against the Third Reich
The key to why the company Deep Mind chose the path of more complex mental sports is probably in a historical connection: the British Demis Hassabis, its CEO and co-founder, was a child prodigy at chess, which he also liked very much since childhood. childhood to his compatriot Turing, born in 1912. It is no coincidence that the secret team led by Turing, and organized by Prime Minister Winston Churchill to unravel the secret code of the Nazis (Operation Enigma), included the three best British chess players of the time: Hugh Alexander, Harry Golombek and Stuart Milner-Barry.
It is estimated that this feat shortened the Second World War by several years and could have saved up to 14 million lives. It is therefore quite logical that, a few years later, in the late 1940s, Turing in the United Kingdom and the mathematician Shannon in the United States separately experimented with the 64-square game as a testing ground for artificial intelligence. Shannon, so passionate about chess that he met the world champion of the time, the Soviet Mikhail Botvinnik, was the first to calculate that the number of possible games is greater than the number of atoms. And Turing wrote the first chess program, Turochamp, which was lost in the 1960s and reconstructed in 2012 to play a game with Kasparov at the congress commemorating the centenary of his birth.
Turing and Shannon envisioned that, if a computer was capable of beating the world champion, what it learned in that process would be very useful in much more important fields of science. What they probably did not foresee is that it would take half a century to achieve it due to the enormous difficulty of expressing in binary language (zeros and ones) concepts that even non-chess players assimilate in half a minute. For example, a machine immediately understands that a queen is worth ten points; the tower, five; the bishop and the knight, three; and the pawn, one. The problem is in relative value: if a queen is trapped by her own pieces in a corner of the board she will not be worth ten points until she frees herself, because in that position she is nothing short of useless. It is impossible to play well without understanding that; Hence the first silicon chess players caused laughter among fans for their ridiculous way of thinking.
But then IBM came along, first with Deep Thought, which managed to beat a grandmaster (the Dane Bent Larsen) in 1988. And then with Deep Blue, Kasparov’s executioner in 1997 (New York, 3.5-2.5 in six games) after losing (2-4) the first duel between the two (Philadelphia, 1996). The Russian’s defeat was front-page news all over the world and collapsed the archaic internet lines of the time. IBM’s share price soared on Wall Street, and it was even said that the whole thing was a circus mounted by the American multinational for purely publicity purposes.
However, IBM was quick to announce that what it had learned from Deep Blue was very useful in a number of fields related to molecular calculations: the manufacture of complex medicines, agricultural planning, air traffic, weather forecasting, the stock market, etc. In other words, Turing and Shannon had been right, but they were prevented from enjoying their time by separate tragedies. Turing committed suicide in 1954 after accepting chemical castration instead of prison because he was homosexual. Shannon lived until 2001, but in 1997 he was suffering from Alzheimer’s.
The wonder that was inspired by its neural network
Hassabis, two years ahead of his age since his teenage years, followed all this closely while, also in 1997, he graduated from Cambridge University with a grade equivalent to outstanding. cum laude in computer science, at 21 years old. He studied neuroscience, and so, when he founded Deep Mind, the light bulb went on that has inscribed his name in history: he would take over from Deep Blue, but with a very different approach; The AlphaZero chess program would be based on neural networks, inspired by the structure of the human brain.
Deep Blue stored a database with millions of games played by humans since the 16th century, when the Spanish priest Ruy López de Segura, sponsored by Philip II, was the unofficial world champion. After that learning, IBM’s monster could calculate up to 200 million plays per second. That is to say, an enormous brute force, but based on a purely human style of playing chess.
The talent and chess passion of the best British players were decisive in deciphering the Nazi codes in Operation Enigma. It is estimated that this intellectual command contributed to shortening World War II by several years and saving some 14 million lives.
Hassabis’ team didn’t put that database into AlphaZero; He only programmed the basic rules of chess. And then he had the machine play millions of games against itself in a few hours. The result was an astonishing rout in 2017 against the best silicon chess player up to that point: AlphaZero 28 – Stockfisch 0 (and 72 draws in a 100-game duel). In parallel, Deep Mind created AlphaGo, which in 2016 and 2017 defeated the best human Go players.
It took Deep Mind four years to amaze the world, through AlphaFold, with one of the greatest advances in biology. Science is already taking advantage of this knowledge of protein structure to research in various fields. Studies have been published on liver cancer and antidepressants, but it can be assured that many more are being worked on because it is a fascinating and hopeful stage of knowledge that has just begun.
Meanwhile, the great paradox is that chess as a sport is in danger of extinction due to artificial intelligence. It is expected that in less than 10 years we will be able to have chips inserted into the brain, or connected to it by glasses or headbands. These gadgets may contain a program that plays chess perfectly, thanks to quantum computing. It seems like it will be very easy to cheat. But there is a solution: the referees will have a detector-disconnector of chips with which to scan players at the tournament gate.
Long live chess, even if in a way that even Turing and Shannon could not have imagined.
The victories of science
There were two man-machine encounters (on the left). The first is won by Garry Kasparov’s biological brain. But in the second, in 1997, Deep Blue’s electronic brain prevailed. With this milestone, the general public takes note that there is a primitive AI capable of complex cognitive tasks. Below, a game of Go against the AI program AlphaGo, and prediction of the structure of a protein generated by the AlphaFold model.
