It would be very interesting to see something similar to Douglas Lenat’s Eurisko applied here, as it seems like the winning “AI” had a lot more guidance than Lenat’s did when he originally implemented it.
Eurisko received a lot of guidance; in some accounts, Lenat gives himself 60% credit.
I would say that Eurisko and the Berkeley group worked on different problems: strategy and tactics, roughly. Lenat played the fleet Eurisko designed, never having played before. That suggests that he didn’t think tactics mattered in Traveller. It is not clear whether this belief lead him to the project or whether Eurisko discovered this in training. Eurisko had to have some opinion on tactics, since it simulated games.
It is not clear that Eurisko did well anything more than apply the heuristic “look for extremes.” It tried looking for interior local maxima, but we don’t know how well it, since it rejected them. The second win seems like evidence that Eurisko knew something general about the game, since it used 100x less computer time to assess the adjusted rules. But some accounts make it sound like it reused the same strategy and the rules just weren’t adjusted well to eliminate its strategy.
It would be very interesting to see something similar to Douglas Lenat’s Eurisko applied here, as it seems like the winning “AI” had a lot more guidance than Lenat’s did when he originally implemented it.
Eurisko received a lot of guidance; in some accounts, Lenat gives himself 60% credit.
I would say that Eurisko and the Berkeley group worked on different problems: strategy and tactics, roughly. Lenat played the fleet Eurisko designed, never having played before. That suggests that he didn’t think tactics mattered in Traveller. It is not clear whether this belief lead him to the project or whether Eurisko discovered this in training. Eurisko had to have some opinion on tactics, since it simulated games.
It is not clear that Eurisko did well anything more than apply the heuristic “look for extremes.” It tried looking for interior local maxima, but we don’t know how well it, since it rejected them. The second win seems like evidence that Eurisko knew something general about the game, since it used 100x less computer time to assess the adjusted rules. But some accounts make it sound like it reused the same strategy and the rules just weren’t adjusted well to eliminate its strategy.