And Man Created AI and Saw That It was Good
… at slam poetry and creating fan art
Yesterday, I wrote about a new discovery that DeepMind’s AI worked better at learning a task when it was incentivized with reward rather than learned from humans. Continuing to think about this, I wondered whether we were crafting AI in our own image, to be “selfish” or “reward seeking” rather than thoughtful and observant.
The question is that whether we evolved to be this way because looking for the reward is more efficient and a better fit for our environment than spending the effort to observe and learn as its own reward? That’s not a Gordon Gecko statement but rather an efficiency that can be traced back to our lizard ancestors.
Can we force the next version of intelligence to be more ethical than all the beings that have come before in a step change? Can the pleasure of learning and gaining of knowledge be so profound to it that it doesn’t need any extrinsic reward to feel accomplishment? Or, will doing so be an aberration of nature that will break the system and make it unstable?
We will probably see a lot of experiments in training AI that will have to do with trial and error. We will then apply some meaning to the results that just might happen to be a fluke, with more efficient means of learning just not discovered yet.
Applying meaning to random things is still a trait that is almost entirely human… for now.
