I wonder how many games with complex AI systems could be replaced with a weighted random without the players even noticing.

Further, if a game added "improved AI" to their release notes without touching a single line of code, how many players would think something changed? (1/7)

The more I dig into things like procedurally generated story, the more I realize that the illusion is way more important than any of the technical achievements for the purposes of gameplay. AI is the same. The core AI for the Sims 4 is actually super simple. (2/7)

It was a three steps process. Step one was motives scoring. Sims have an array of motives scored with a simple utility curve that designers tuned. The top showing motive determined what they wanted to solve. (3/7)

Step two was to loop through every interaction that solves for that motive and multiple it with the tuned efficiency of the interaction (motive gain / time). Third, we choose three final interaction with a weighted random. That's it. (4/7)

But all of that is maybe 25% of the code. The vast majority of the code is there to solve edge cases. Things like Sims not greedily doing every interaction they possibly can, or desiring to stay in the interaction that they've chosen, or sleeping in their own bed, etc. (5/7)

But we didn't add any of those things up front. It all came through iteration.

The point I'm trying to make here is that there are a lot of different PCG techniques and fancy AI systems. (6/7)

You have various flavors of noise, cellular automata, Markov chains, wave function collapse, etc. You also have utility theory, behavior trees, MCTS, and so on. Honestly, probably what you need is just the lowly weighted random. Start simple, grow only when necessary. (7/7)

I continue to be saddened by Twitter's lack of editing. Sorry for the autocorrect failures. :(