Flash of Steel header image 2

Building a Better Odin

October 8th, 2007 by Troy Goodfellow · 3 Comments · Design, Ensemble, RTS

AI researchers in Austria have come to the conclusion that a “neurotic AI” is more likely to win real time strategy games than other AI personalities. In a test of four different AI scripts, aggressive and neurotic personalities won all their matches against the default AI in Age of Mythology, but the neurotic one won its matches 25 per cent faster. A neurotic AI tended, for example, to make irrational evaluations of available resources and veer between extremes in its play style.

All the emotionally scripted AIs won handily against the default computer opponent, though. “Defensive” and “Normal” were 6-1. And the average “Defensive” win was only four minutes slower than the Neurotic one.

The research is trying to determine if “emotional” AIs would be more attractive to a human player, and, with only the power point presentation to guide me, I’m not quite sure how a strong win-loss record against the Hard AI of Age of Mythology gives any clue as to whether it would be more enjoyable as an opponent. Is the assumption that a stronger AI is also a more “fun” AI?

It is, undeniably, more fun to play a real time strategy game against a human, but human skill is only part of the appeal. Personally, I like the fact that, even though I suck, human control gives the potential of really “outsmarting” someone. Not to mention the trash talking.

A more human AI is one that, as one of the power point slides indicates, adjusts its calculations based observable factors. Not just stuff like “this opponent gets cheap cavalry” but things like distance to resource points, assessing choke points, mastery of the seas, etc. Take the new Honshu island map in Asian Dynasties. Like all maps in Age of Empires III, it has predictable features. There are treasures in the ocean as well as on land. The single trade route is not on the main map. And the “native tribes” are in the middle of the map.

Even if you haven’t played the demo, this description should give you enough information for a human to plan a strategy. A single monitor ship can destroy everything, so one option is go heavy naval. The one trade route means that even if you can’t hold it, you shouldn’t let your opponent grab it for no cost. On land, the map is won or lost with control of the trading posts; Europeans should just plunk some outposts there. And, like many Age 3 maps, there is never enough wood. Geography is as important as play style.

It’s worth noting, then, that the AI scripts were run on the Alfheim map in AoM, a hilly and heavily forested map that can either have lots of hunting (key for the Norse) or little. (I’m not sure how you could overestimate the wood supply on a map like Alfheim, in any case.)

If you have any interest or insight about RTS AI, please comment. Is this research helpful, or is it already obsolete given what is going on the industry at the moment?


3 Comments so far ↓

  • Alan Au

    Map knowledge is extrememly important for RTS games (and all map-based games). It’s the “local knowledge” factor and one of the reasons I’m kind of annoyed by competitive multiplayer, but I digress.

    This study, while interesting for the Alfheim map, has very limited transferability unless it can be shown that the same principles hold true for other map layouts.

  • Alan Au

    Also, here’s the obligatory plug for http://www.gameai.com. It’s infrequently updated (and a bit out of date), but it’s a pretty good resource and makes for a good primer on basic AI techniques. For example, it has a nice discussion about the AI used in AoE 1 & 2, both as intended, and as actually implemented.

  • Chris Nahr

    This research makes a lot of sense. If I understand correctly what the “neurotic” AI does, it’s acting very quickly and decisively along one — ANY one — axis of the game’s possibility space. For example, pretend there’s ton of lumber, so use it all up right now, build lots of stuff fast, and strike hard!

    That agrees with winning RTS strategies as well as conventional military wisdom. It’s better to fully commit to one path of action that might not be the wisest than to squander resources across all possibilities. Besides, AI players are terrible at reacting to strategic situations, so it’s better for them to take the initiative and force the other player(s) to react. Same thing in Civilization: the psychotic leaders are the most dangerous.