You should watch that. It’s the most enlightening talk I saw recently.
Yeah, definitely seems to jive with my perception of how things are discovered / invented. I don’t think all the details wrt to parallels between human accomplishment and the results of his studies have been fleshed out, but it’s definitely an attractive and believable explanation for quite a lot of phenomena.
I’d say the whole idea is not really new. There is some similarity with the idea of local minima. But the way he aggregates all the bits, and try to provide a kind of experimental proof through an AI experiment, is really novel to me.
Fabulous video of a point that’s often mentioned, but not followed. The diversity of Picbreeder and its focus on individual autonomy (rather than convergent consensus) reminded me of ARPA’s model in the past of hiring top talent, and paying them top dollar to research/develop whatever they wanted. The result was a diverse range of immensely valuable products, which have served as crucial ‘stepping stones’ to many innovations today.
Now for a bit of a side-track…
Another interesting ‘stepping stone’ from this video lead me to explore the idea of genetic algorithms for composing music. A really fascinating development in this field is GenJam, an interactive genetic algorithm that professor John Biles uses as an improvisation partner at gigs.
Fitness tests are often not useful for composing original music, since it’s very difficult to express the idea “this musical phrase sounds nice” in code. As a result, some genetic algorithms for composition respond to human feedback, like Picbreeder. The result seems to be a much more broad exploration of the musical search space, usually resulting in innovative musical phrases that arise from ‘stepping stones’.
A plausible next step to human feedback is the co-evolution of musical ideas and music critics, where musical ideas that are highest rated by music critics generate more offspring. A prevailing idea for the heuristic used by these music critics is exactly what you might expect after watching this video: Novelty Search.
Here’s a link to Biles’s paper on GenJam if anybody is interested. Also; a paper on co-evolution of musical phrases and music critics.