Hacker News new | past | comments | ask | show | jobs | submit login

Interesting. It seems like what they tested here was not that each agent should get as much points as possible like in a usual prisoner dilemma, but just that they should have more than the opponent (and thus get selected for reproduction).

Not sure about the explanation in the article "This domination would last until enough errors accumulated in the code handed down between generations for dominant machines to stop recognizing each other.", shouldn't be the agents that stop recognizing the code forced out? I guess it depends on implementation details, if every agent can fight several opponents before being or not reproduced.

I wonder if it also has an efficiency metric: a shorter agreeable program is better than a longer agreeable program. Then, over time, you start failing to detect bad programs after long periods of peace.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact