# Toward an AI Physicist for Unsupervised Learning

arXiv: https://arxiv.org/abs/1810.10525 · HTML: https://mlsanity.com/paper/1810.10525/

## Key facts

- Cited by 12 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 24 Oct 2018.

## Built on by (most cited first, 12 of 12)

- [Machine-Learning Non-Conservative Dynamics for New-Physics Detection](https://mlsanity.com/paper/2106.00026/) (2021)
- [Discovering conservation laws from trajectories via machine learning](https://mlsanity.com/paper/2102.04008/) (2021)
- [Oracle-Preserving Latent Flows](https://mlsanity.com/paper/2302.00806/) (2023)
- [Similarity-Based Equational Inference in Physics](https://mlsanity.com/paper/2103.13496/) (2021)
- [Learning Hamiltonian dynamics by reservoir computer](https://mlsanity.com/paper/2104.14474/) (2021)
- [AI Research Associate for Early-Stage Scientific Discovery](https://mlsanity.com/paper/2202.03199/) (2022)
- [Automated Gadget Discovery in Science](https://mlsanity.com/paper/2212.12743/) (2022)
- [Advances in Electron Microscopy with Deep Learning](https://mlsanity.com/paper/2101.01178/) (2021)
- [Toward Building Science Discovery Machines](https://mlsanity.com/paper/2103.15551/) (2021)
- [Neuronal diversity can improve machine learning for physics and beyond](https://mlsanity.com/paper/2204.04348/) (2022)
- [Microscopy is All You Need](https://mlsanity.com/paper/2210.06526/) (2022)
- [How AI Agents Discover Scientific Equations: From Hydrotope Rediscovery to New Water-Wave Amplitudes](https://mlsanity.com/paper/2610.00435/) (2026)

## Abstract (arXiv)

We investigate opportunities and challenges for improving unsupervised machine learning using four common strategies with a long history in physics: divide-and-conquer, Occam's razor, unification and lifelong learning. Instead of using one model to learn everything, we propose a novel paradigm centered around the learning and manipulation of *theories*, which parsimoniously predict both aspects of the future (from past observations) and the domain in which these predictions are accurate. Specifically, we propose a novel generalized-mean-loss to encourage each theory to specialize in its comparatively advantageous domain, and a differentiable description length objective to downweight bad data and "snap" learned theories into simple symbolic formulas. Theories are stored in a "theory hub", which continuously unifies learned theories and can propose theories when encountering new environments. We test our implementation, the toy "AI Physicist" learning agent, on a suite of increasingly complex physics environments. From unsupervised observation of trajectories through worlds involving random combinations of gravity, electromagnetism, harmonic motion and elastic bounces, our agent typically learns faster and produces mean-squared prediction errors about a billion times smaller than a standard feedforward neural net of comparable complexity, typically recovering integer and rational theory parameters exactly. Our agent successfully identifies domains with different laws of motion also for a nonlinear chaotic double pendulum in a piecewise constant force field.