Artificial intelligence (AI) is not my area of expertise, but I suppose that being a generalist gives me license to comment on it. And anyway, the implications of AI could be far reaching for economics – and humanity in general, of course. In a liberating or in a very scary way, depending on which path we follow with it. But that’s for another time. Today I want to focus on the technology itself. Progress that we have made with chess and AI is truly mind blowing and I wanted to describe some of it for you.
I never used to believe it possible to construct a thinking machine. The human brain always seemed exceedingly complex – nothing that we could ever hope to replicate digitally. Information processing, sure. Computers have far surpassed us in speed and capacity, and they do not make mistakes. But thinking? I never thought it attainable, at least not in my lifetime. But a video that I recently watched put the lie to that!
A little context first. I happen to be a chess geek (though not a terribly good player) who has long been following the progress that computers have made against humans. When I was a child, even the best computer in the world would stand no chance against a master-level player. But they gradually improved and about 20 years ago a computer named Deep Blue was able to best the then-world champion, Gary Kasparov, though not without difficulty. Today, Magnus Carlsen, the world’s best player says that he struggles even to draw with the white pieces against the world’s best chess computer program, known as Stockfish (he almost invariably loses with black).
How have computers made such progress? Again, processing speed. Even decades ago, it was a relatively simple matter to store a massive number of quality games into a computer’s memory. But having it cycle through all possible combinations of moves during a game against a human would have taken days. Today, in contrast, a computer can go over literally millions of possibilities in minutes or even seconds – a decided advantage against any human. The human knows instinctively that there are usually at most a few serious moves to consider. The computer, in contrast, considers good moves, dubious moves, and even ridiculous moves before deciding on the best one. In other words, it is not thinking. But the technology has advanced to the point that what programs (or “engines,” as geeks refer to them) like Stockfish lack in intuition is more than made up for by speed.
Well, it now seems that they have also gained that intuition. The fascinating and scary thing about it is that, to my understanding, it does not happen gradually. Computers do not think – until, suddenly, they do. The above-referenced video is truly unbelievable. Anna Rudolf, an International Master in her own right, narrates a game between Stockfish and Alphazero, an AI program. I won’t bother with any details here, which would likely bore you if you are not a chess player. But I believe that you’d get something from watching it regardless – she is very good!
More to the point, this experimental AI technology, which was earlier used to destroy the Go world champion back in 2016, teaches itself. In other words, Alphazero was given no history, no prior games. Its only input was the rules of the game. After then playing millions of games against itself for four hours, it was able to beat Stockfish! The video I cite vividly describes its unorthodox, creative, almost human-like play. And really it is nothing more than a highlight. Stockfish and Alphazero played 1,000 games back in 2018 and drew 839 of them. But of the remaining 161 that were decided, Alphazero won 155 of them!
As an informatics non-expert, I cannot meaningfully comment on what this amazing cognitive advance signifies. It is true, however, that we still have lots of work to do. Alphazero knows how to think and learn, but it cannot feel, it has no passions, emotions, or preferences. As much as we like to think of ourselves as “rational” creatures, any psychologist worth her salt would tell you that we mostly let our judgments, intuition, and inclinations decide for us, not “data” or information. Still, after watching this video and mulling over what it implies, future possibilities are both exciting and frightening to contemplate.