Hacker News new | comments | show | ask | jobs | submit login
A.I. in Prisoner's Dilemma (medium.com)
10 points by StevenWeinberg 11 months ago | hide | past | web | favorite | 1 comment

> In the context of this A.I Prisoner’s Dilemma, If Luke snitches, Dave’s best response is to snitch, since 3 years is better than 5. If Luke doesn’t snitch, Dave’s best response is to snitch, since 1 year is better than 2.

Another way to look at this is that the AI has no "ethics" or "morals," and therefore the AI wouldn't care that it's betraying its buddy AI/human. In other words, the AI would "sell its own mother," if it came out ahead.

I think it's an important perspective to keep in mind as we start integrating AI more and more into our societies "because it's a more efficient/smarter way" of doing things. Perhaps higher efficiency won't always be a desired outcome.

Leaving an AI in charge of nation's healthcare funds comes to mind here. Would an "advanced AI" start denying people healthcare if they have a "low enough" chance of survival or because they've reached an age where continuing to pay for their treatment is not "cost-effective" for the society anymore? Is that a desired outcome?

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