Ai2 ships Theorizer in Asta for generating testable theories
- Source
- x.com
- Date
Theorizer is now available in Asta. It draws on published research to propose theories scientists can test. Watch @AllenInstitute neuroscientist Jérôme Lecoq explain how he used an early version to explore a link between neurons and vision. 🧵
Context
Ai2 used this post to announce that Theorizer is now available in Asta, its platform of AI agents for scientific research. Theorizer draws on published studies to propose theories a scientist can then test, and Ai2's announcement adds that it links each proposed explanation to evidence from those studies. The post itself is three short sentences. What it adds is a short video of Jérôme Lecoq, an Allen Institute neuroscientist who develops methods for recording brain activity in mice, talking about using an early version to explore a link between neurons and vision. The video's speech was captured by an automatic transcript that mishears some words, and its visuals were not reviewed, so what follows paraphrases only the passages that are clear.
Lecoq says he expected one or two theories and got 50. Ai2's blog places that run on VIP interneurons, brain cells that suppress the activity of other neurons. He says that within a couple of minutes the tool can produce a theory consistent with current understanding of a topic without him reading all of the papers. Most of his remarks, though, are about what he wants to keep for himself. He describes himself as driven by understanding and sees these tools as augmenting his ability to understand and helping him connect the dots. He warns that if researchers stop understanding what such a system generates and leave it running, it will produce things nobody cares about, and he says people have to stay in the loop, placed at the right point. His case for the tool is about volume. Papers now appear faster than anyone can keep up with, and neuroscience is split across many labs each pursuing its own question, so he sees Theorizer helping a researcher survey a field broadly and find the promising areas before going deeper into one.
The video is a researcher describing a way of working, not evidence that a theory held up: nothing in its transcript describes a tested result. The details sit in the blog post, covered in our earlier Intel on Theorizer's launch. That post describes one of the theories, which matched Lecoq's own view of these cells, says its central idea had already been proposed in the literature, describes tests by scientists in three other projects, on lung cancer, climate data and potato diversity, and states that Theorizer can miss relevant studies and misjudge whether a theory is new.
Checking sign-in…
Loading comments…







