What matters in AI.

Subscribe

Ai2 launches Theorizer, an AI agent that proposes science theories

Theorizer links each theory to published studies, but the user must check each claim of novelty.

Ai2 launched Theorizer in Asta, its platform of AI agents for science, and opened the code for other researchers to examine. Theorizer reads scientific papers and proposes explanations for observations, for example a tumor that grows for no known cause. On the web, Ai2 pays the token costs, the text units the model reads, to a maximum.

How it works

Theorizer examines papers for the question of the user. It makes a template of the items to read from each paper, and compares the results to make theories. Each claim links to the papers behind it, and the tool flags claims that it thinks are new.

The proof

A cancer doctor used it for lung tumors with no known cause of growth. It gave 3 ideas, and the other tool, AutoDiscovery, tested them on open tumor data. In brain research, it gave 50 candidate theories, where the scientist expected 1 or 2. Ai2 gave no test of precision.

What is not known

The cancer results must be tested in the laboratory. A check of papers can miss work from before, and the model can misjudge which claims are new. In the brain work, the central idea was in the literature before.

Insight: The tool is strongest where it agrees with experts

The cancer doctor said that the first theories held ideas that the doctor formed in 15 years of work. The brain scientist found that the central idea was known before, and valued the joined proof. This reads as a tool that gathers known proof fast, which shows that its value is speed of work, not new theories. For leaders, the right test is the time that it saves.

Sources

  1. Theorizer comes to Astaallenai.org
AI MATTER · NEWS · AI MATTER · NEWS ·19 NOV2026

Posted

Learn the terms in this story

More in Models

All Models news