WORK
The Autodidactic Universe
The 2021 paper co-authored by Smolin, Jaron Lanier, and Microsoft researchers proposing that learning is a cosmological primitive — the mathematics of neural networks and the mathematics of spacetime share structural features that suggest the universe may be teaching itself its own laws.
The Autodidactic Universe is a 2021 research paper — co-authored by Smolin with Jaron Lanier, Stephon Alexander, William Cunningham, Andrew Friedland, Marina Cortês, and researchers at Microsoft — that proposes a formal correspondence between the mathematical structure of neural network learning and the mathematical structure of physical law. Write Einstein's general relativity in a specific form (the Plebanski action), and the equations governing spacetime curvature correspond, at a certain level of abstraction, to the equations of a Restricted Boltzmann Machine. The paper does not claim that the universe literally is a neural network, or that spacetime literally learns. It claims something more subtle and more consequential: that learning — the adjustment of parameters to produce increasingly organized outputs — may be a cosmological primitive rather than a biological invention.
In The You On AI Field Guide
The paper emerged from a collaboration between physicists working on quantum gravity