[z3c0]

It will take me some time to dig up with all the noise around AI, but this reminds me of a paper published around 2018 or so that explored the possibility of two such AI forming an accidental trust by optimizing around each other. For example, if the denying AI used frequency of denied claims as a heuristic for success, and the AI drafting claims used the claim amount for the same, then the two bots may unknowingly strike a deal where the AI drafting claims lets smaller claims get frequently denied to increase the odds of larger claims.

Note: not saying these metrics are what would be used, just giving examples of antithetical heuristics that could play together.

[0_____0]

I feel that this sort of autonomous agent co-optimization may happen more often over time as humans step farther away from the loop, and lead to some pretty weird outcomes with nobody around to go "wait what the f---- are we doing?"

[z3c0]

Agreed, and I'm even further worried about the plausible deniability these situations would create.

[asdf147]

Wow, that sounds very interesting, do you have some link to that paper?

[z3c0]

I'm on the hunt, but no luck. I've tried a myriad of search terms to dig it up, but none are able to surface the paper through all the vaporware and blog pieces on competitive AI.