Comment on AI CEO – Replace Your Boss Before They Replace You

• AlteredEgo@lemmy.ml ⁨3⁩ ⁨days⁩ ago

  They would generate strategies that maximize their objective function based on the training data. Obviously garbage in garbage out, but my point is they would not be prone to certain irrationalities like humans.

  It might be possible to regulate how AI CEOs are optimized and trained though. You can tell a human CEO a thousand times “we only have one earth, if you all externalize your cost we will all die and have zero profits” but an AI might actually get it. AI might also be connected to a kind of crowdsourced democratic economic global forum, where people can discuss, complain and make suggestions.

  AI also has a much higher bandwidth and might catch institutional problems much easier because it doesn’t have to rely on summaries of subordinates to understand how things are going.

  More broadly, it might be theoretically impossible for humans to act according to our shared values - no matter what rules, institutions, education or culture we create. Like “theoretically impossible, the system always degenerates” because individual humans will always follow their own greed and lust for power while pretending to comply, and then using that power to slowly pervert the system and it’s rules. I believe that is the root of our current malaise. But even non-sentient AI might be able to help us just enough to make it work. It’s much more likely that those in power will use it for the opposite, but that shouldn’t stop us from thinking about if it can be used for good.

  source