ATG recently had a conversation with A.I. Specialist / YouTube Host Tommy Thompson about the history of A.I. in Games Development, and where it is headed next.
Here are a few of our key takeaways
- Because AI will immediately find the most optimal solution to a problem, at decision level, it will typically make gameplay far too difficult. So the AI must be dumbed-down. In older games titles, such as Goldeneye, this was done by "trapping" the AI in a slow animation - example slowly dropping to one knee, raising a rifle and then firing at the player. Rare essentially played with animation delays / animation transistion times to dial in the difficulty.
- (in modern titles the above method is still used, but the granularity of available behaviours and animations is much larger and more time-consuming)
- AI must be adequately telegraphed to the player, through animation and audio, for the player to "feel" the AI making decisions. Example: Half Life - the Marines will communicate their intent to flank the player via radio-chatter. Example: Halo - upon spotting the player, the Elites will growl and assume something of a Māori-Haka-war-dance pose before attacking the player.
- Alternatively, you can have AI who's first concern is for its own safety, such as in F.E.A.R. "I just dont want to die" causes the AI to dive behind cover (upon seeing you) before it engages you in combat. Their decisions are based upon minimising risk to themselves.
Lastly, to our surprise and pleasure, it turns out that Tommy graduated from Strathclyde University in Glasgow, as did our Chairman. It’s a small world!
ATG got interested in Strathclyde a while back when we visited its Technology & Innovation Centre in 2018. Winning the Times Higher Education University of the Year award, it’s now home to over 23,000 students from 100 countries.
