[bb123]

I wonder if the model here is missing something important - it assumes there is no survival penalty for a false positive attack. In real life this isn’t true. Activating a fight or flight response in humans and animals has an energy (survival) penalty.

[goldfish]

It does include a survival penalty for a false attack. I laid out the assumed payoff matrix here: https://www.adamjuliangoldstein.com/blog/paranoia-parameter/.... The penalty for false positives is also baked into the simulations (see first few lines of code here: https://github.com/adamjuliangoldstein/anxiety_algorithms/bl...).

[zomglings]

You are right. It would work well if the models included a utility function and, instead of measuring likelihood of survival, the measurement were expected utility.