Thoughts as Thinkers: biology & computer science of inspiration within & without us
- Date published: 2026-10-31
- Speakers: Michael Levin (talk given in a joint session; he says he is presenting "together with Chris")
- Format: Talk
- Watch: https://www.youtube.com/watch?v=EBF0KUX_jpA
- Topics: Platonic space and ingressing minds, Xenobots and anthrobots, Cognition in cells, AI and biology
In one paragraph
A 35-minute talk on what Levin calls "free lunches": abilities a system shows that were not paid for by design, natural selection or training. He walks through examples at every scale, from people with very little brain tissue and tadpoles that see through an eye on their tail, to molecular networks that learn, xenobots and anthrobots, and simple sorting algorithms that do things nobody coded. He argues these abilities come from an ordered space of patterns (the Platonic space) that should be studied systematically, and that inspiration in human creatives is the same thing at a higher level. This is a different upload from the 2026-10-16 "Thoughts as Thinkers" talk on this site.
Main ideas (plain words)
- Design, selection and training are the three known ways to "pay for" a competency. He argues they don't explain or predict much of what we see.
- He defines a free lunch as what a system can do minus the effort put in, and says these lunches are heavily discounted rather than truly free.
- Examples: rare clinical cases with most brain tissue missing and normal or above-normal IQ; tadpoles that can see out of an eye placed on the tail; a turtle on a skateboard that keeps up with a cat.
- Modeled molecular networks (coupled differential equations) can learn in about six behavioral paradigms, and learning raises their integration, which doesn't drop back when they are forced to forget. He says random networks are already close to optimal for this.
- Xenobots (frog skin cells) and anthrobots (adult human airway cells) show abilities, like kinematic self-replication and helping heal a scratch in a layer of neurons, that no evolutionary history selected for.
- Simple sorting algorithms keep working on unreliable hardware and do unrequested "side quests", including something like delayed gratification. He says a bigger paper on this is coming.
- He rejects plain "emergence" as too thin an explanation and proposes a structured latent space of patterns, with low-agency ones that look like math and high-agency ones that look like minds.
- At the end he discloses that three or four companies have licensed some of the lab's inventions.
Key quotes (word for word from YouTube's auto-generated captions)
"the three ways we have of basically paying for competencies or good patterns that do something interesting and those are design, selection, and training."
https://www.youtube.com/watch?v=EBF0KUX_jpA&t=60s"I'm going to call these free lunches, although I don't think they're actually free. I think they're heavily discounted."
https://www.youtube.com/watch?v=EBF0KUX_jpA&t=270s"It turns out that random networks are incredibly close to being optimal for this kind of ratchet."
https://www.youtube.com/watch?v=EBF0KUX_jpA&t=750s"There's an ordered structured latent space of patterns."
https://www.youtube.com/watch?v=EBF0KUX_jpA&t=1320s"Inspiration all the way down."
https://www.youtube.com/watch?v=EBF0KUX_jpA&t=1980s"there are uh three or four companies that have now licensed some of our"
https://www.youtube.com/watch?v=EBF0KUX_jpA&t=2070sSource: YouTube auto-generated captions, speakers not labeled; wording is provisional. Timestamps point to the start of the 30-second stretch that holds the quote.