I've long felt that the mystery game genre is cool in theory but doesn't work in practice on computers. Most games end up being menu clickers. Or you have to add puzzles. Which is fun, but a puzzle game isn't the same as a mystery game.
Generative AI changes things. It's possible to make a more open-ended game where you can act like a real investigator. Draw up lists of suspects, order forensic tests, call experts, interview witnesses, whatever you can think of.
You won't have to wait long, that space is red hot right now. People are rushing to commercialize LLM powered NPC dialoge. nVidia's ACE might cut everyone's lunch though we will see.
There is one tech demo which has you piecing together what happened at an accident by talking with the NPCs.
There's also a Skyrim mod that replaces all the NPCs with LLM actors.
I think the turning point for that technology will be locally run LLMs that can be very finely tuned and trained on your world data.
I think it will be a long long time before we see an LLM generating a genuinely novel, interesting and fun to play mystery game.
Good mysteries involve providing just enough hints so that the player can make the fun leaps of logic and experience "Aha!" moments. People don't want to play the job of a real detective (collect endless witness statements, construct extremely detailed timelines) they want to be the hero in a detective story, making clever deductions.
LLMs might be able to generate characters and dialogue but they're going to struggle to come up with completely new ideas and mystery mechanics like Her Story
> People don't want to play the job of a real detective (collect endless witness statements, construct extremely detailed timelines)
You underestimate how many people would be really into what you described. Lots of wildly popular games are work in a fun-looking package. Factorio is just programming. Football Manager, and every other sports manager game for that matter, is spreadsheet after spreadsheet with some cool graphics.
That is what the tech demo does, each actor has a prompt that includes what that person knows about the situation and how they should dicuss it but the story itself was designed by a human. The prompts for the actors were written to make the discussions interesting and useful for deducing what happened. So perhaps Actor A saw a mysterious man, Actor B heard the noises and has a better idea of the timelime, and a different detective has already examined the physical clues so you can discuss them with that actor.
It's really interesting but it's also of course really easy to break out of sensibility, and hard to properly gamify. Like you said people don't actually want the tedious part of detective work, they want the fun parts that make them feel like Sherlock.
The game Shadows of Doubt is a great example of how the information flow of detective work is really difficult to balance, I think they did a great job but if you run out of really strong clues, clues that link things together, the game becomes a brute force exercise in talking to everyone and asking the same questions.
Generative AI changes things. It's possible to make a more open-ended game where you can act like a real investigator. Draw up lists of suspects, order forensic tests, call experts, interview witnesses, whatever you can think of.
I wish a game like that existed.