Spotless

I set out to do what I did, what I found, and what I would do next.

A lot of AI safety related games are "bash you over the head" with ai safety ideas. I aimed for a more subtle approach with more emergent ah ha moments. In particular I wanted the player to notice that simple optimization goals (cleaning) in familiar environments (home) could be pushed towards maladaptive behaviors when competed or optimized against. I also liked the idea of playing with a reverse stealth game (the analogy being that your AI looks aligned as long as you keep looking at it in the right way).

I found that balancing showing these ideas in emergent ways is not easy! Especially in such a short timeframe. But I think I gave it a good shot and made something that kinda gets my idea across.

If I were to develop this more fully I would spend a lot more time on the mechanics and more deeply tying and balancing that tension between cleaning "cleanly" vs cleaning via "dirty" behavior.