Saturday, September 13, 2008

The Fifth Generation

Back when I was an undergrad, there was a major effort going on in Japan known as the Fifth Generation. The idea was that by using massive parallelism, one could simulate thought, and build systems that were intelligent (for some definition of intelligent; there's certainly some debate on what that would mean). Neural networks, genetic algorithms, all that jazz. This wasn't a small effort; it dominated computer science research in Japan, and there were quite a few people in the US who also felt that this was the right direction.

I've got a book from that era, The Fifth Generation, by Edward Feigenbaum (a senior Stanford professor), and Pamela McCorduck. Reading through it is fairly surreal; there's a level of certainty that I find astonishing. I don't get a feeling of this is an interesting research direction, and we should follow it. Rather, it's more of this will certainly work, and we must direct our efforts toward it and away from other ideas.

The Japanese research community lost many valuable years chasing AI systems. Likely many promising researchers were shut down if they did not subscribe to the party line. Research dollars were wasted barking up the wrong trees. The psychological term for this is Groupthink. Scientific discourse is halted, dissenting views are quashed. Everyone falls into lock step, but not necessarily in the right direction.

So a question for the random person who found their way here... Suppose the US government offered funding for perpetual motion research -- they want a system with flywheels, and it has to generate electricity. Would you take the funding? Or would you ask hard questions of the funding director? And suppose they're offering lots of money? Would you take it then? And if you saw many of your colleagues line up, and say how they had always had misgivings about thermodynamics, and are enthusiastic about the prospects of perpetual motion. Previous failures, they assure you, were only due to a lack of effort and funding -- and now, we're serious about it, and it has to work or we're in trouble. If you felt that the research community was headed in the wrong direction, would you speak up, or go along? The only way to break out of Groupthink is for enough members of the group to try to shake things up.