I have heard the claim from various sources that the AI of S.T.A.L.K.E.R: Shadow of Chernobyl, a Ukrainian FPS RPG-ish thing from 2007, had beta AI that was so good that an AI character could be given the goals of going to particular locations, fighting specific enemies, etc, equivalent to that a player would do to beat the game - and succeed. In other words, the AI was good enough to beat the game, and the developers were forced to dial it back in the final release. One example of a website making this claim, from the TV Tropes page for the game (brackets mine):

Word of God [TV Tropes speak for an interview or similar developer testimony] states the A.I. actually had to be toned down because in early builds, it would regularly and consistently outsmart the player, to the point that the game became outright frustrating - and eventually completely unplayable. The original pre-release A.I. apparently would have been more than capable of beating the game entirely without player intervention if it hadn't been dialed back.

While the AI in the final game is extremely impressive at its best, it's also known to be quite buggy, such that I struggle to imagine it actually being able to beat the entire game. (Of course, maybe this is because of its cut capabilities.) And anyway, even if the AI was that powerful, it would have been put up against many opponents with the same AI, so it isn't clear to me that or why the AI put in the shoes of the player would be able to outsmart/outgun the many equivalent AI put in the shoes of the enemy characters. I also know of no explicit citations of the supposed interview, nor have I seen someone unlock the AI's abilities in the code (supposedly the unrestricted AI still exists in the game, simply with many functions disabled) and prove it.

Does anyone know of the interview cited in the above excerpt, or other evidence that the AI was in fact this powerful, or is this merely a rumor or an exaggerated claim of the game's AI abilities?

  • 12
    Video game AIs are well known for cheating, and being able to do things that player characters can't do, so the idea that a video game could be completed by its own AI doesn't sound all that dubious to me.
    – F1Krazy
    Commented Sep 26, 2023 at 14:42
  • Even if this was possible, wouldn't the opponent ais just as competent as the playing ai?
    – pinegulf
    Commented Sep 27, 2023 at 8:02
  • @pinegulf This is the primary reason I don't believe it out of hand. It isn't at all clear to me how or why a single hyper-competent video game AI in the role of the player would be able to outsmart and outgun the many equally hyper-competent opponents - especially considering that at the beginning of the game the player would have inferior weapons and armor, giving an even further disadvantage. Even if it is cheating, it would need to cheat above and beyond its own exact counterpart to consistently defeat several of itself, or at least it would seem so to me.
    – Benyamin
    Commented Sep 28, 2023 at 1:32
  • 1
    @F1Krazy A competent AI is inherently going to be cheaty--it will be the equivalent of using a perfect aimbot. It also has more ability to see the detail than any human possibly could. It can also do things like perfect shooting on the move. While many game AIs get cheats like no fog of war you don't need that to have an AI outperform a human. Unless the opponents have weaknesses that a human can exploit I would expect an AI to outperform at a tactical level. They generally perform abysmally at a strategic level. Commented Oct 2, 2023 at 2:35
  • 5
    @F1Krazy In the case of a first person shooter, an AI can have perfect accuracy and near-instant reflexes. Programming a good shooter AI is easy- programming a bad one is the challenge. Commented Oct 2, 2023 at 9:38


You must log in to answer this question.

Browse other questions tagged .