Spelda, Petr and Stritecky, Vit (2026) How Artificial Intelligence Steers the Course of Science. [Preprint]
![[img]](https://philsci-archive.pitt.edu/style/images/fileicons/text.png) |
Text
spelda-stritecky-ai-steering-science.pdf - Submitted Version Available under License Creative Commons Attribution. Download (452kB) |
Abstract
General-purpose AI models used as co-scientists are developed with the help of public evaluations that represent performative feedback from scientists pushing for greater levels of AI research assistance and acceleration with every model improvement. Formal methods that can ensure the risk of mismatches between model capabilities and scientific requests approaches the optimum over time are insufficient to guarantee science will be accelerated evenly, remain diverse and keep widening its scope. Should validity be the sole peer review criterion determining problem acceptance to impactful AI benchmarks when co-scientist models steer science by disproportionately improving in areas covered with this performative feedback? We show that while validity is necessary, it does not always support strategic selection. Peer reviewing problems for AI benchmarks as performative feedback requires a new competence that weighs validity and steering of co-scientist models jointly and with foresight. A failure to meet the challenge could ruin even acceleration by valid feedback that does not exploit performativity for steering AI co-scientists away from a monoculture. The key insight is to maintain benchmark diversity by keeping model scores on a benchmark hard to predict from existing evaluations.
| Export/Citation: | EndNote | BibTeX | Dublin Core | ASCII/Text Citation (Chicago) | HTML Citation | OpenURL |
| Social Networking: |
| Item Type: | Preprint | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Creators: |
|
|||||||||
| Subjects: | Specific Sciences > Artificial Intelligence General Issues > Science and Policy |
|||||||||
| Depositing User: | Dr. Petr Spelda | |||||||||
| Date Deposited: | 15 Apr 2026 12:31 | |||||||||
| Last Modified: | 15 Apr 2026 12:31 | |||||||||
| Item ID: | 29128 | |||||||||
| Subjects: | Specific Sciences > Artificial Intelligence General Issues > Science and Policy |
|||||||||
| Date: | 4 April 2026 | |||||||||
| URI: | https://philsci-archive.pitt.edu/id/eprint/29128 |
Monthly Views for the past 3 years
Monthly Downloads for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](http://philsci-archive.pitt.edu/style/images/feed-icon-32x32.png){kind=icon}