Gaming AI So Human-Like it Fools Panel of Judges!

2K Games has sponsored a competition to find the most human-like gaming artificial intelligence, offering a prize of $7000 to the winner. One important goal was to try and find an AI that was considered more human than a real human – and they succeeded.

Popular game Unreal Tournament 2004, a twitchy, fast-paced shooter, had its built-in AI replaced with the competition entries for about half the onscreeen opponents and the human judges had to decide which players were human opponents or AI by the use of a “judging gun” which tagged the opponent as either human or bot. Interestingly, the humanness target for the competition was 50%, with the human players achieving a humanness rating of only 40%, while the two winning AI systems both achieved a humanness rating of 52%!

The two winning bot artificial intelligences were UT^2 and MirrorBot, which both tied for first place and share the $7000 top prize. Now we just need these bots made into official game plugins, which will make a great game even more awesome!

Intriguingly, the competition wasn’t run using the latest version of the game, Unreal Tournament 3, which was widely regarded as a bit of a disappointment, although it wasn’t terrible.

The victory comes 100 years after the birth of mathematician and computer scientist Alan Turing, whose “Turing test” stands as one of the foundational definitions of what constitutes true machine intelligence. Turing argued that we will never be able to see inside a machine’s hypothetical consciousness, so the best measure of machine sentience is whether it can fool us into believing it is human. “When this ‘Turing test for game bots’ competition was started, the goal was 50 percent humanness,” said Miikkulainen. “It took us five years to get there, but that level was finally reached last week, and it’s not a fluke.”


Comments are closed.