Novelty Search: A Heuristic for What Does Different Mean

3min Snip

00:00

Play full episode

Summary

Transcript

Episode notes

I think I'm a little confused about what Novelty Search really entails because for any problem, it seems like there's an infinite number of really dumb novel directions to go. So the idea is that if we had a true beautiful Novelty Search algorithm that was based in the rich sense of human novelty, you know, would be one way to go. But if you're trying to do this within an algorithm where you don't have access to that nuanced signal, then you might be driven to design a different kind of heuristic,. Not a heuristic about what is good, but insteadYou make a heuristic for what does different mean in a way that hopefully will capture something interesting about a

Read the full transcript here.

Where does innovation come from? How common is it for "lone wolf" scientists to make large leaps in innovation by themselves? How can we imbue AIs with creativity? Or, conversely, how can we apply advances in AI creativity to our own personal creative processes? How do creative strategies that work well for individuals differ from creative strategies that work well for groups? To what extent are models like DALL-E and ChatGPT "creative"? Can machines love? Or can they only ever pretend to love? We've worried a fair bit about AI misalignment; but what should we do about the fact that so many humans are misaligned with humanity's own interests? What might it mean to be "reverent" towards science?

Joel Lehman is a machine learning researcher interested in algorithmic creativity, AI safety, artificial life, and intersections of AI with psychology and philosophy. Most recently he was a research scientist at OpenAI co-leading the Open-Endedness team (studying algorithms that can innovate endlessly). Previously he was a founding member of Uber AI Labs, first employee of Geometric Intelligence (acquired by Uber), and a tenure track professor at the IT University of Copenhagen. He co-wrote with Kenneth Stanley a popular science book called Why Greatness Cannot Be Planned on what AI search algorithms imply for individual and societal accomplishment. Follow him on Twitter at @joelbot3000 or email him at lehman.154@gmail.com.

Further reading: