Tuesday, October 09, 2007
Neurotic AI The Best? Don't Count On It.
There's a little bit of a stir caused from an article that appeared last Friday in the New Scientist claiming that Austrian Researches found that a "Neurotic" Artificial Intelligence player performed best in a series of test against the built-in AI in the game "Age of Mythology." Coming in second was an "Aggressive" AI personality.
According to the article, "The neurotic bot was more likely than the others to distort hard facts about resources - like the amount of timber around - and flip between extremes of behaviour. And it was better than the rest."
But the test was only made against the default AI in the game. I predict that against humans, it's not going to be nearly as interesting. This was the classic "Star Trek" attack against computers - be entirely illogical and make the computers short-circuit.The thing is, it works. For all one of you who has been reading the blog that long, you might remember how I held off the strongest AI setting in Rise of Nations for nearly four days using the "Do Nothing Because I Forgot To Pause The Game When I Went For Dinner And Then Forgot About It" defense. Undoubtedly, if I had never resumed that game, the AI would still be dealing with my impenetrable defense strategy today, nearly two years later. Dang, I'm good.
For some kinds of games - particularly those in which the AI must rely upon heuristics rather than brute-force problem-tree searches -the AI performs at its best only if the player is behaving rationally and employing similar, tried-and-true strategies. Most artificial intelligence "bots" in commercial games are completely incapable of learning or adapting. The same tricks work against the AI over and over again, so it is extremely poor at adapting if the opponent isn't playing along in the expected manner. Behaving irrationally is usually not a winning strategy; however, may trip up the AI.
So this finding isn't particularly surprising to me. It is amusing, and I'd be interested in hearing how it fares against humans. But at this point, I'd not read too much into it.
(Vaguely) related pontification:
* Jet Moto Memories
* Elements That Make Believable AI
* Game Moments #5: Rise of Nations
* Playing Lately: Go
* Computers Playing Go
.
Labels: Mainstream Games, programming
- posted by The Rampant Coyote @ 5:42 AM (Permalink)
Comments:
Links to this post:
<< Home
My first year computer science course was an advanced one, so rather then having rote lectures we would have professors come in and talk about their various specialties.
One of the guys was an AI specialist, and he described his work getting a computer to completely destroy an old EA soccer game (FIFA 97 or something). The computer eventually figured out a that from a corner kick it could force a penalty, and it could score on a penalty kick every time.
I learned from this that video game AI's are usually neurotic themselves. They are designed to be beaten. Why would anyone play against an unbeatable AI?
Chances are, the neurotic personality occasionally hit on the neurotic weakness of the built in AI, and exploited it unintentionally. Like you say, not especially surprising.
I'm curious about how their definition a neurotic AI vs. aggressive AI would apply to human players. There are people out there who carefully deconstruct games to find inherent weaknesses: I once a description of how to win Civ4 on the highest difficulty by throwing your country into a permanent state of anarchy. Whether that is rational behaviour or not depends on what level you look at it.
Post a Comment
One of the guys was an AI specialist, and he described his work getting a computer to completely destroy an old EA soccer game (FIFA 97 or something). The computer eventually figured out a that from a corner kick it could force a penalty, and it could score on a penalty kick every time.
I learned from this that video game AI's are usually neurotic themselves. They are designed to be beaten. Why would anyone play against an unbeatable AI?
Chances are, the neurotic personality occasionally hit on the neurotic weakness of the built in AI, and exploited it unintentionally. Like you say, not especially surprising.
I'm curious about how their definition a neurotic AI vs. aggressive AI would apply to human players. There are people out there who carefully deconstruct games to find inherent weaknesses: I once a description of how to win Civ4 on the highest difficulty by throwing your country into a permanent state of anarchy. Whether that is rational behaviour or not depends on what level you look at it.
Links to this post:
<< Home
