Another hat tip to Instapundit, there's a great link in New Scientist suggesting that neurotic AIs outperform others. Indeed it suggests erratic behavior may have some competitive advantages, maybe even for people.
The researchers took the AI from Age of Mythology and souped it up with four different emotional patterns based on human personality research. The aggressive and neurotic bots both won every match (out of 7) against the standard AI, but the neurotic bot, prone to extremes of emotion and to distorting facts, won faster.
Implications for gaming AI: enormous. Implications for human psychology: enormous.
ID:35506
 Oct 8 2007, 5:35 am (Edited on Oct 8 2007, 5:44 am)
|
|
Too bad Age of Mythology SUCKS
By the way, if you want a really good book on the issue of AI (esp. concerning emotions and how they may benefit AI), grab yourself a copy of Steve Grand's "Growing up with Lucy", if you haven't already. You should know who he is- the guy who created Creatures. | |
I disagree. The implications for broad based human psychology are minimal at best. For one thing, much of this is nothing new to game theory. Consider that the results are directly related to the experiments rule set. All this proves is that in the AoM rules, aggression and unpredictability are good for baseline strategy against computer opponents. This is consistent with many games which reward aggression to prevent boredom and where uncertainty rewards apparently erratic (but generally goal oriented) behavior.
Of course, human interactions are rarely so simple as even the most complex games (where at least the participants presumably know the rules, and are compelled to follow them in any case!), so contributions from game theory must be closely scrutinized before drawing any broader conclusions. Saying that one strategy is optimal in chess, for example, does not in any way mean that same strategy is even possible in normal behavior. Also, why were the sample sizes so small? 28 total games? Why weren't dozens or even hundreds run? With automated action on both sides, it would require little oversight. | |
Indeed bigger samples would be welcomed. What might be even better would be pitting the AIs against each other. The neurotic AI might outperform the aggressive one against a standard AI, but vs. each other that might not hold. And what about against a defensive AI?
| |
Excellent point! Those silly Austrians- they like to cook up grand schemes, but when it comes to the carry through.....
Ummm, my comment timestamp shows Monday, October 08, 2007 03:27PM- appx. 30 minutes before Lummox JRs, which I am responding to. How is this possible. Unless, I can see the future! | |
Something is messed up in the comments; the most recent one always appears in BST.
| |
I agree with Jmurph. This makes for a nice headline, but all the experiment really proves is that the AoM AI is not good at dealing with unpredictable opponents. Big frelling deal.
| |
That's true, but it does leave a lot of interesting questions open for further research. For example: Is that true of other AIs? Is it true of human behavior? And forget about the defensive end of it; does unpredictability and a tendency to make snap decisions offer any kind of competitive edge in the real world? This study may have been too small and proved too little, but it opened up a very interesting line of study.
| |
Strangely, the pics in the article are from Age of Empires 3.
Both of them are good games though. Age of Mythology is my favorite of the two though. Its the game that made me interested in the Norse, Egyptian, and Greek pantheons. I'd probably not even know ANY of the gods of any of those three civilizations without AoM. | |
Hopefully you will post the results of the tests against humans if they post it themselves. I would love to program this kind of thing into a game but at this time I'm not skilled enough nor am I creative enough to solve the problems associated with AI.