The year is 1996. A piece of news quickly goes around the globe, arousing a stir and admiration: for the first time a computer manages to prevail against a world champion on the chessboard. It was February 10, 1996 when Deep Blue, a computer designed and built by IBM specifically for the game of chess, defeated the holder of the title, Garry Kasparov, in the opening match of the man-calculator challenge. It is a historic date, artificial intelligence achieves a goal considered by most to be out of reach, impossible to reach, barred to machines. Although the IBM supercomputer got the better of Kasparov in the first game, the final outcome of the clash sees the victory of the Russian champion. In subsequent matches Deep Blue is unable to repeat the initial success and in the end there are three victories for Kasparov and two draws. Man beats car 4-2.
Many breathe a sigh of relief, humanity preserves its primacy. The overtaking of computers to the detriment of man is perhaps no longer the exclusive subject of science fiction, but at least it is still to come. Nothing more wrong! The following year sees the revenge between the two contenders and this time Deep Blue prevails. Therefore in May 1997 computer beats man: 3,5-2,5. A limit is crossed, a wall shattered.
Garry Kasparov reconstructed this lost challenge in a recently published 300-page book entitled "Deep Thinking: Where Artificial Intelligence Ends and Human Creativity Begins".
However, the event is not without controversy. First of all, Deep Blue is not located inside the room in which the match is played, but a few kilometers away and, consequently, the data is not sent directly from the computer. This fact and other elements lead Kasparov to claim that Deep Blue was helped during the matches. An element that returns several times during the challenge and the refusal of IBM to Kasparov's request to provide the printouts concerning the activity of the machine, as per previous agreements, only reinforces doubts and fuels suspicions. The loser finally asks for a rematch, but IBM opposes an outright refusal, definitively withdrawing Deep Blue.
20 years later comes the Deep mind algorithm
In a 2003 documentary, Game Over: Kasparov and the Machine, interviews are offered with some witnesses who cast a heavy shadow on the supercomputer's victory. From these it would emerge that, ultimately, Deep Blue represented a ploy implemented by IBM to increase the value of its stock on the stock exchange. In the single day of the triumph over Kasparov, IBM shares rise by 15%.
So the victory of the machine, specifically, is perhaps much more humane than initially led to believe. Or the result of chance, as Nate Silver claims. In his book (The signal and the noise. The art and science of prediction) he ascribes the success of Deep Blue to an absolutely irrational move, the result of chance, made by the machine. A move that would have irreparably destabilized Kasparov's humanity, undermining his trust and rationality.
Twenty years later, AlphaGo breaks another limit. The Deep Mind algorithm in the first months of 2016 defeats, this time without controversy or tricks, first the European champion and then the world GO champion. The victory is clear and unequivocal. And that the machine is really capable of outclassing man is demonstrated by the fact that, in one week, between 29 December 2016 and 4 January 2017, AlphaGo, in disguise, managed to ring together sixty consecutive victories against the best players of the planet in the flesh.
Calculation capacity: 10170
Until recently it was believed that some particular areas would remain the exclusive domain of man or, in the worst case scenario, would remain so for a long time to come. Among these were both chess and Go. Both games, although they have meager rules and apparently simple dynamics, actually hide a high difficulty. A difficulty that is given by the number of combinations of moves and consequently of possible game configurations. Such a large number as to lead one to believe that no computer would have been able to master such complexity. Therefore it was common opinion that chess, and even more Go, would remain the prerogative of human intelligence and above all intuition. Still less was such a sudden capitulation expected.
But if Deep Blue managed to prevail through brute force, thanks to a formidable computing power (capable of calculating 220 million moves per second), for AlphaGo the matter is different. Go has a much higher number of combinations than chess. For each move, on average, the chances are ten times more. The calculation capacity alone, albeit enormous, would not be sufficient to consider about 10170 different situations and at the same time to be right about the pitfalls posed by the game.
So how did AlphaGo manage to beat the world champion? Simple, learning.
Deep learning, the machine's secret weapon
As Demis Hassabis, founder of Deep Mind says, there are two ways to make a computer smarter. A first method provides for programming a determined solution to a given problem and letting the machine execute it. The second, and this is the case of AlphaGo, consists in «… giving the computer the ability to learn from itself, to learn through experience. … I'll teach you how to learn, I won't give you the solution».
The secret behind AlphaGo's success is called Deep Learning. "The algorithm has three features, all taken from neurobiology: reinforcement learning, convolutional neural networks, and selective memory loops."
Another peculiarity that distinguishes AlphaGo from Deep Blue consists in the fact that the Deep Mind creature was not created specifically and exclusively for Go. The underlying algorithm has proven to be able to learn and adapt, over time, to different types of games. Its progress and its acquisitions are not in fact due to careful and meticulous planning, but are the result of experience. What it knows, AlphaGo has discovered and assimilated by playing, in perfect autonomy. “The hallmark of the new generation of algorithms is that, like people, they learn from their successes and failures.”
Trained in as many as 49 different Atari 2600 games, Deep Mind's algorithm has outshone not only other competing algorithms, but humans as well. "… in 29 of these games the algorithm outscored a human professional tester by 75 percent or more, sometimes beating them by a very large margin."
Cognitive Computing
AlphaGo demonstrates that we are now entering what IBM describes as the third era of data processing: that of Cognitive Computing. The first era, that of tabulating computing, from the very beginning until the 50s, saw computers capable of performing essential mathematical operations, using punched cards for both input and output. The second era, that of Programming Computing, from the 50s to the present day, is the era of electronic computers. Machines are also able to process commands (if/then) and logical structures, but they must be instructed to do so by following certain rules and therefore need to be programmed.
Cognitive Computing, explains Jonas Nwuke (IBM), «… was born to break down the rigidity inherent in the “if/then” logic … Cognitive Computing is a probabilistic elaboration, whose outcomes vary along a spectrum, rather than being rigorously “yes or not, right or wrong”».
The emphasis placed by IBM on Cognitive Computing is, at least in part, self-interested, it is one of the cogs of a powerful marketing machine that was not satisfied with the victory over Kasparov. AlphaGo, in fact, has a predecessor or, if you want, Deep Blue had a successor. A Supercomputer that has broken yet another barrier has robbed man of another hunting reserve, depriving him of a further certainty: in a word, Watson.
We will deal with Watson in the next post
