How Watch Dogs: Legion’s ‘Play as Anyone’ Simulation Works | AI and Games

Tommy Thompson’s deep dives into AI in games frequently overlaps with procedural generation in games, and that’s certainly in the forefront of this latest video about the characters in Watch Dogs: Legion.

One thing that struck me is how the systems in the game are building on previous research. For example, the player-time character generation is partially inspired by research on alibi generation by Ben Sunshine-Hill and Norman Badler.

I’d like to know more details on how the generation of character details, because their ability to generate from partial data sounds like they’re using some form of constrain solving but I’m not sure.

It’s also worth pointing out that they don’t try to generate everything from scratch: recruitment missions, for example, have some basic narrative scaffolding that cab have the dynamic elements slotted in. This is a smart way to handle it: you don’t need to generate everything, and writing some solid content saves you time for enriching the generative system in ways that pay off more effectively.