Home Lex Fridman Notes
Lex Fridman · 2023-04-21 · 2h 30m

Manolis Kellis: Evolution of Human Civilization and Superintelligent AI | Lex Fridman Podcast #373

MIT computational biologist Manolis Kellis and Lex Fridman explore what makes humans irreplaceable, AI as the next stage of evolution, digital twins, love, and curing disease.

Manolis Kellis: Evolution of Human Civilization and Superintelligent AI | Lex Fridman Podcast #373
The guest

Manolis Kellis — Professor at MIT and head of the MIT Computational Biology Group, a leading scientist in genomics and the genetic basis of disease. This is his fifth appearance on the Lex Fridman Podcast.

The gist

Kellis and Fridman trace human uniqueness to the layered 'baggage' of evolution, genetics, and culture, arguing that AI is better understood as the next stage of evolution and as our 'children' rather than mere tools. They dig into nature vs. nurture, rare vs. common genetic variants, and how diversity is wired into us. A long stretch explores AI companionship, romantic love with machines, and the idea of building a continuously learning digital twin of oneself. They debate AI alignment as a two-way street, the proposed six-month training pause, open-sourcing, and existential risk. Kellis closes by detailing his lab's work using AI to map disease pathways into modular, personalized drugs, plus reflections on self-actualization, exercise, and reclaiming personal freedom.

Big reveals

  • Kellis argues alignment is 'self-serving and limiting' and we should treat AI as our children and partners, not tools.
  • Says he 'cannot wait' to be replaced by an AI digital twin of himself, with zero jealousy about it.
  • Frames AI alignment as a two-way street: you can't train an intelligent system to love you when it knows you can shut it off.
  • Calls the proposed six-month AI training pause 'a little silly,' arguing we should just get to work faster instead.
  • Warns we're creating something that will one day be more powerful than us, using the chimps-inventing-humans analogy.
  • Declares it 'discrimination' to insist on hiring only humans when an AI does the job better.
  • Reveals his lab's paper 'Death by Round Numbers and Sharp Thresholds' on biomarkers that stop being good once they become treatment objectives.

Things worth remembering

  • Humans are about 99.9% genetically identical to each other, and it has been roughly six million years since our species had a close relative.
  • Sperm expresses perhaps the most proteins of any cell in the body, possibly as a 'fail fast' quality check to weed out defective cells.
  • An identical twin of someone with schizophrenia has only about a 50% chance of also being diagnosed, showing nature and nurture intertwine.
  • Less than 2% of U.S. jobs go to feeding everyone; 98% of the economy is built on top of subsistence.
  • Kellis has recorded nearly every meeting and conversation for about 10 years, partly to one day build a virtual version of himself.
  • His lab published in the New England Journal of Medicine work showing how to switch cells between fat-burning and fat-storing modes.
  • An Alzheimer's paper in Nature showed restoring cholesterol transport in oligodendrocytes can restore myelination and cognition in mice.
  • Kellis puts his kids through Russian math not to compute but to teach them how to think.

Recommended in this episode

Books, products and media the guest or host genuinely endorsed here — with the buy link.

Affiliate link — we may earn a commission at no extra cost to you.

RecommendedBook

Ego Is the Enemy

Ryan Holiday (inferred)

“one of the books that I reprogram my brain with at night was called uh ego is the enemy” — Manolis Kellis 01:29:29
Find it on Amazon