Re: Other fads - AI is based on stats too
"It took Google about 100 computer scientists to feed the Alpha Go database with millions of past historic documented Go games."
I doubt it. In chess, databases of millions of games have long since been compiled. Without huge effort, you can also choose to compile a database of the extant million (or three) games played by the tested best players. If the quoted statement is true, it could just as well be argued that the 100 scientists (and it could more cheaply have been 100 lay people or 100 teenagers with direction) were first filling a surprising lacuna in the culture of the game of Go. I've read Go books, and the approach is a lot like chess books: here is a technique, and here is a game that illustrates the technique. Rinse. Repeat.
In chess, it is not difficult with a high degree of reliability to evaluate each move of the 6 million games. The, ah, Art has sufficiently advanced that the evaluations of expert human commentators have been checked by computer chess engines. I'm guessing, but I suspect that was not so easy in the game of Go, as the new program did not have so much in the way of shoulders to stand upon.
The failures of programmers to replicate the ways they thought the human brain worked, while winning chess games, led pretty much to redefinition of what AI was. The chess programs that play at human World Championship level (or higher) do not attempt to mimic how a theorist thinks the human brain works. The programs that did, were left in the dust decades ago. Although it might be fun to bring back one of those programs on modern hardware. Chess has ratings. If your rating is 200 points the higher, in a match of 4 games, you will win about 3. A rule of thumb is that doubling the processing speed of a competitive chess program will improve it by about 20 rating points. Between the World Champion and the level of those dusted programs on their contemporary hardware is in excess of 1,000 rating points. But I'd rather liken it to the exercise of looking back at Phrenology and rescuing the parts of it that sort-of were valid.