Clever Machines Learn How to Be Curious
I just love the way efforts to create effective AI reveal so much about what it means to be human. In this case, developers are training AI to be “curious, but not too curious.” An AI agent needs to be curious about its environment, to explore. But if too curious it might get trapped exploring “pointless novelty.” So what isn’t pointless? “Only features of the environment that have the potential to affect the agent (or that the agent can influence). In essence, if the agent can’t interact with a thing, it won’t even be perceived in the first place.”
Is that how we perceive the world most of the time, unless we purposefully try to be mindful of what is happening in the moment? Perhaps even then we are unable to perceive certain things that are “pointless” for us to perceive.
Read the article in full: “Clever Machines Learn How to Be Curious,” by John Pavlus, Quanta Magazine
If you like this subject, you might enjoy my post, “After You: Teaching robots to maneuver politely around people.”