Peter Clark is interim CEO and a senior research director at the Allen Institute for AI. He studies machine reasoning and automated scientific discovery, co-leading Asta after earlier work on science-question answering and systems that expose their reasoning.[1]
Start with CodeScientist →
Career
Clark earned degrees in physics at Oxford, information technology at Edinburgh and computer science at Strathclyde. His career has included university research, Boeing and Vulcan before Ai2. He was interim CEO in 2022–2023 and returned to that role in 2026. His research-director work spans Aristo, structured reasoning and the scientific-discovery program Asta.[2][3]
Contributions
Evaluate the experiments behind generated discoveries
CodeScientist combines research articles with reusable code blocks to search for experimental ideas and implementations. Clark co-authored the work, which evaluates outputs through more than paper review: it also examines code and attempts replication. Of 19 discoveries returned in the reported study, six were judged at least minimally sound and incrementally novel. The result illustrates both the promise of automated experimentation and the need to check the artifacts it produces.[4]
Measure scientific agents under comparable conditions
AstaBench evaluates agents across the scientific research process, including literature use and data analysis. Its environment supplies reproducible tools and accounts for differences such as cost and tool access. Clark is a co-author. The paper’s evaluations show progress on individual tasks while finding that broader research assistance remains difficult.[5]
Tools, writing & talks
Writing, research & talks
Start here
CodeScientistarxiv.orgA research system that proposes and runs experiments, with code review and replication checks.
Start with the system overview and the evaluation of generated discoveries. Some machine-learning experience helps; inspect the replication results before treating a generated finding as established.
Co-author[4]