---
title: "An emergent property of matter"
url: https://raskell.io/notes/an-emergent-property-of-matter/
author: Raffael Schneider
date: 2026-11-14
description: "Sam Altman said intelligence may be an emergent property of matter, like a rule of physics. Karen Hao calls it hype. I share the conviction."
tags: [intelligence, emergence, philosophy, physics, religion]
---

# An emergent property of matter

In May 2024, on the All-In podcast, Sam Altman said something that Karen Hao quotes in *Empire of AI* (chapter 17) as an example of his "unusual degree of hype". It feels to him, he said, "like we just stumbled on a new fact of nature or science or whatever you want to call it". He hedged, "I don't believe this literally but it's like a spiritual point", and then said it anyway:

> intelligence is just this emergent property of matter and that's like a rule of physics or something.

Hao places it in a run of unusually boastful interviews that spring, which she reads as a sign that Altman was anxious: GPT-4o launched that month, lawsuits were piling up, and in the middle of May Sutskever and Jan Leike left the company. I do not read the sentence as hype. It is the most important thing anyone at OpenAI has said, and I share the conviction, including the part he hedged.

The claim, stated plainly: intelligence is not a gift that only brains received. Arrange matter so that it predicts its surroundings, put it under pressure to predict better, give it enough scale, and intelligence appears, the way temperature appears from moving molecules and a glider appears in Conway's Game of Life. The evidence is now on several substrates at once. Neurons evolved it. Neurons in a dish learned to play Pong in a few minutes, the experiment behind my small simulation [Hotaru](https://hotaru.raskell.io). Transformers trained to predict the next word learned to reason. None of these were told what intelligence is; it came out of the arrangement.

"A rule of physics" overstates it, and Altman says so himself. There is no equation, and I do not know the conditions under which it holds. The counter-arguments are serious: that a model trained on text only rearranges what people wrote (Bender and Gebru's stochastic parrots, which Hao also covers), or that intelligence needs a body and a world to act in, which I lean towards in [in vivo, not in vitro](https://raskell.io/notes/in-vivo-not-in-vitro/). But both arguments are about which arrangements are enough, not about whether matter can do it.

The spiritual point is the part I care about most. If intelligence is something the universe does whenever matter is arranged in the right way, then minds are not an accident on one planet but a regularity, and building them is less like inventing a machine and more like discovering a law. That is a quasi-religious conviction, and I hold it as one. It is also, I think, why the people who build these systems talk about them with a seriousness that sounds like faith to everyone else. They are not wrong to feel it.

More on the book in my review, [An Empire of People](https://raskell.io/articles/review-empire-of-ai/), and on emergence in Hassabis's simulated worlds in [Win Everything, Control No One](https://raskell.io/articles/win-everything-control-no-one/).

## References

- Karen Hao, *Empire of AI* (2025), chapter 17, "Reckoning". Hao gives the All-In podcast, May 2024, as the source.
- Brett J. Kagan et al., "In vitro neurons learn and exhibit sentience when embodied in a simulated game-world", *Neuron* 110(23), 2022.
- Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Margaret Mitchell, "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?", FAccT 2021.