Published Jun 25, 2026
Use case for AI in video games #32: Interrogation and Catching a Lie.
The detective gameplay mechanic lives on finding clues and solving puzzles, but also on improvising while interacting with suspects. The NPC should react to what you actually ask, and no two interrogations should play out the same.
In 2011, long before LLMs, a brilliant game by Rockstar and Team Bondi, "L.A. Noire," chased this. Much of its marketing pitch rode on one promise: you could catch a liar by reading their face. The studio scanned every actor with a 32-camera rig (MotionScan) so a suspect's eyes would flicker before a bad answer. Then the player watched the face and pressed one of three buttons. Truth. Doubt. Lie.
The face was real, but the conversation wasn't. A player never actually said anything, only picked a label and the scene branched. Most games are still designed like this: persuasion is a charisma stat or a dice roll. You don't win the argument, just pass the check.
A language model inside the game changes that. The suspect holds a story. You, the detective, notice that the alibi doesn't fit a timeline you set three scenes ago, and you say so, in your own words. The model weighs your claim against what it knows about the case and either holds the line or cracks. The argument carries the scene, not a number on your sheet.
Catching a lie stops being a menu pick. It becomes the contradiction spotted, the question you knew to ask, the NPC who finally slips because you found the one thread that did not hold.
What games come to mind that would be a good fit for this AI-backed mechanic? Share in the comments.
Picture credit: Screenshot L.A. Noire, Rockstar, Team Bondi.