DOE should judge Genesis Mission awards by science delivered, not compute pledges

The projects have varied aims, from developing a virtual model of a fusion device to improving particle-physics calculations and designing enzymes. A University of California, San Diego project aims to expand an RNA structure database fivefold. That is a target, not an achieved result.

Those aims need project-specific tests. A database expansion might be assessed differently from a better microchip design or a faster research calculation. But each team can report what researchers did before, what AI-assisted workflow they tested and whether it improved a defined scientific task.

DOE’s Phase I guidance points in that direction: teams are to demonstrate a research workflow that incorporates AI and quantitatively assess whether it offers an AI advantage. Phase II supports larger teams pursuing promising directions. The agency should make evidence of progress, not a project’s place in a large portfolio, central to judging results.

A DOE-hosted summit report records participants’ concerns that publication counts alone are poor measures of impact and that research effects can be hard to quantify. That is a reason to choose relevant measures, not to abandon measurement. Participants suggested indicators such as validated models, new capabilities and successful use in practice.

Compute credits matter only if they reach researchers

The strongest case for company credits is that they are intended to make computing resources available to research teams. Politico reported that as a goal, but the pledge does not establish that teams have received usable capacity.

The reported $2.4 billion commitment from 11 companies includes compute credits and AI tools; it is not DOE grant money for the 12 Phase II projects. Nextgov/FCW and FedScoop report the pledge, but their coverage does not establish how much has been delivered, which teams will receive it or how the resources will be valued.

Those details matter. A promised credit is not the same as capacity in a researcher’s hands, and access alone would not show that a research workflow improved. The credits may prove valuable; the public record should show what teams received and what that access enabled.

The same distinction applies to DOE’s awards. The announcement describes intended work, not verified scientific outcomes. The coverage to date does not establish an independent assessment of the new projects’ performance. DOE’s account of a project’s potential is a rationale for funding, not proof that the promised capability has arrived.

The strongest counterargument is that researchers need room to experiment and that scientific impact can take time to emerge. Both points are fair. A rigid deadline or single score could favor easy-to-count outputs over important work that takes longer. But flexibility is no reason to leave taxpayers unable to tell what was tested, what changed and what remains uncertain.

Make progress visible, project by project

DOE should require a public progress record for each award: the starting workflow, the scientific task and comparison method, the results, and whether another party has validated any claimed advance. Teams should also report limitations and unsuccessful tests. A workflow that fails to improve a task can still produce useful evidence if the test is clear and the result is reported honestly.

The measures should fit the science. A fivefold expansion of a database, for example, would not by itself prove that researchers can answer scientific questions better. Nor should every team be judged by publication totals or whether its work has already moved into practical use. The point is to connect each project’s stated aim to evidence that it was met, and to explain when that evidence is not yet available.

Independent review would make results more credible, especially when claims of acceleration come from a funded team or a company providing tools. DOE need not promise that every project will succeed. It should make the evidence, limitations and terms of any success claim visible.

The Genesis Mission should be judged by what research teams can demonstrate, not by $159 million in awards or billions in pledged resources. DOE’s Phase I guidance calls for quantitative assessment of whether an AI-enabled workflow offers an advantage. The department should make clear, project-level evidence of results central to public reporting across the mission.