AI Exposure Scores: what they measure, what they miss, and what comes next

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Campbell Lund, Thomas Euyang, Zanele Munyikwa, Marzieh Fadaee
• Venue: arXiv: cs.AI
• Published: 2026-06-22
• Collected: 2026-06-23
• URL: http://arxiv.org/abs/2606.23633v1
• PDF: https://arxiv.org/pdf/2606.23633v1
• DOI: UNKNOWN

Plain-Language Brief

The article discusses the limitations of a set of AI Exposure Scores introduced by Eloundou et al., which measure the share of tasks that can be assisted by large language models. The authors highlight two main gaps: first, between what these static scores measure and what policy questions require; second, the coordination between researchers and policymakers. They propose various approaches to address these issues and suggest that better measurement is not enough to close the gap between research and policy.

What the Document Says

• In 2023, Eloundou et al. introduced GPTs scores, which define exposure as the share of occupational tasks a large language model can assist with.
• The article highlights two gaps: one between what static scores measure and policy questions, and another regarding coordination between researchers and policymakers.
• Five families of research are surveyed to address these limits: dynamic and benchmark-based measures, ensemble methods, task-framework extensions, worker-centered metrics, and adoption and usage data.
• The authors argue that closing the research-policy gap requires policymakers to widen their evidence base, engage workers as epistemic partners, and shift from prediction to preparedness. Researchers are also asked to build data infrastructure, adopt participatory methods, and write with policymakers in mind.
• The article suggests that better measurement will not close the second gap alone.

Key Concepts Mentioned

• AI Exposure Scores
• GPTs scores
• occupational tasks
• large language models
• policy questions
• research-policy gap

Claims Made by the Document

• Static AI Exposure Scores have limitations in policy-facing analyses.
• Closing the research-policy gap requires actions from both policymakers and researchers.

Evidence Used

• The authors argue that the temporal, geographic, and ontological limitations of the GPTs scores compound in policy-facing analyses.

Methods / Approach

• Surveying five families of research
• Discussing coordination between researchers and policymakers

Limitations Stated by the Authors

The article does not provide specific examples or case studies to illustrate the compounding limitations of the GPTs scores in policy-facing analyses.

Limitations Noted by the Agent

Agent interpretation — not from the source.

The data used for the analysis is based on the GPTs scores introduced by Eloundou et al. in 2023, which might have changed or evolved over time.

Open Questions

• What specific examples or case studies illustrate the compounding limitations of the GPTs scores in policy-facing analyses?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The article does not provide details on how researchers and policymakers can engage workers as epistemic partners, nor does it specify what ex-post frameworks are.

Raw Abstract / Excerpt

A set of exposure scores calculated in 2023 has become a central empirical input to the future of work debate. Produced by Eloundou et al. (2023) and referred to here as the GPTs are GPTs scores, they define exposure as the share of occupational tasks a large language model can assist with. This work is a genuine methodological contribution, but as the scores travel from the time and place they were produced, the limitations the authors named do not always travel with them. Two gaps have widened as a result. The first is structural, between what static exposure scores measure and what policy questions actually require. Taking the diffusion of these scores as a case study, we show how their temporal, geographic, and ontological limitations compound in policy-facing analyses, and we survey five families of research responding to these limits: dynamic and benchmark-based measures, ensemble methods, task-framework extensions, worker-centered metrics, and adoption and usage data. The second gap is the one we argue needs more attention: the coordination between researchers and policymakers. The policy-relevant work which ask who is harmed, who benefits, how, and when, continues to reference the static GPTs are GPTs scores without engagement with the methodological updates that would let these questions be answered more reliably. We then ask what additional steps towards navigating uncertainty remain: ex-post frameworks and the deliberate, political work of reimagining what futures are worthy of building towards are. Closing the research-policy gap is a shared task: policymakers must widen their evidence base, engage workers as epistemic partners, and shift from prediction to preparedness; researchers must build data infrastructure, adopt participatory methods, and write with policymakers in mind. Better measurement matters, but it will not close the second gap alone.