IV Co-Scientist: Multi-Agent LLM Framework for Causal Instrumental Variable Discovery
Ivaxi Sheth 1 Zhijing Jin 2 Bryan Wilder 3 Dominik Janzing 4∗† Mario Fritz 1 1 CISPA Helmholtz Center for Information Security 2 MPI for Intelligent Systems, Tuebingen; Vector Institute; University of Toronto 3 Carnegie Mellon University 4 Amazon · 2026
Read IV Co-Scientist: Multi-Agent LLM Framework for Causal Instrumental Variable Discovery by Ivaxi Sheth 1 Zhijing Jin 2 Bryan Wilder 3 Dominik Janzing 4∗† Mario Fritz 1 1 CISPA Helmholtz Center for Information Security 2 MPI for Intelligent Systems, Tuebingen; Vector Institute; University of Toronto 3 Carnegie Mellon University 4 Amazon (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07943.txt.
Description
Causal inference from observational data often needs an instrumental variable, a factor that shifts the treatment without directly affecting the outcome, but finding valid instruments takes creativity and domain knowledge. Ivaxi Sheth and four co-authors test whether large language models can help.
In a two-stage evaluation, they first check whether models recover established instruments from the literature, then whether they avoid instruments already discredited by theory or evidence. Their system, IV Co-Scientist, is a team of agents that drafts, criticizes, and improves candidate instruments for a treatment-outcome pair, with a statistical test for consistency without ground truth. Economists and epidemiologists gain a promising AI collaborator for a traditionally human craft.
Catalog record
| CREATOR | Ivaxi Sheth 1 Zhijing Jin 2 Bryan Wilder 3 Dominik Janzing 4∗† Mario Fritz 1 1 CISPA Helmholtz Center for Information Security 2 MPI for Intelligent Systems, Tuebingen; Vector Institute; University of Toronto 3 Carnegie Mellon University 4 Amazon |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07943