As we highlighted in our September 2026 report, AI for Discovery: New Frontiers, Challenges, and Opportunities for Scientific Research, artificial intelligence is reshaping the practice of discovery. Researchers are using AI to accelerate their work across all parts of the scientific process—including reviewing literature, generating hypotheses, designing experiments, and analyzing data—thereby expanding what scientists can study and how quickly they can study it. These new technologies also raise consequential questions about who has access to the resources required for AI-enabled research, how scientific data should be governed as it moves through increasingly complex computational systems, and how the integrity and trustworthiness of research can be maintained as AI becomes more deeply embedded in the scientific enterprise.
These considerations are closely connected. Access to computing resources shapes who can participate in AI-enabled science. Rules governing data determine what research can be conducted, by whom, and under what conditions. Prevalence of AI throughout the research process affects reproducibility, research integrity, scientific communication, and public confidence.
While the tools available to science will continue to change, the more immediate task is to build the infrastructure, governance systems, and institutional practices that allow technology to advance scientific discovery while protecting the conditions on which a healthy research ecosystem depends.
To chart the path toward such an ecosystem, the Aspen Institute Science & Society Program convened experts from universities, government, technology companies, scientific publishing and journalism, civil society, healthcare, and research and policy organizations. Throughout the summer of 2026, our conversations with these leaders focused on research infrastructure, privacy and data governance, and scientific trustworthiness.
The roundtables examined both the opportunities created by AI and the practical constraints emerging as researchers adopt these technologies. Universities must decide how to allocate limited AI resources across researchers and fields. Funders must determine how to evaluate proposals that depend on new models, datasets, and computational methods. Research institutions, journals, and scientific communities must decide what evidence and practices are sufficient to establish the credibility of AI-enabled research. Understanding these institutional decisions—and the tradeoffs, incentives, and standards that shape them—is an important next step in understanding how AI will change not only the practice of science, but the systems that determine what research is pursued and how scientific knowledge is evaluated.
Our participants debated specific questions including:
- How can public research institutions secure sustained access to computing resources and technical expertise?
- How can governance systems protect sensitive data while enabling legitimate scientific use and reuse?
- How can researchers preserve provenance, reproducibility, and accountability when working with rapidly changing AI systems?
- How can institutions ensure that AI supports scientific expertise without weakening the human capabilities on which future discovery depends?
The Science & Society Program synthesized the discussions into policy recommendations reflecting perspectives raised across the conversations. We then brought the draft recommendations to panels of reviewers (who had not participated in any of the original roundtables) with experience translating policy ideas into practice. These review sessions included civil-society leaders, current and former government officials, legislative staff, researchers and university leaders, and think-tank experts working across state, federal, and international policy. Reviewers were asked to assess whether the recommendations addressed decisions that policymakers and institutions can realistically make, whether responsibility was assigned to the appropriate actors, and where proposals needed greater specificity or attention to implementation. Their feedback informed the final recommendations presented in this playbook.
Across the roundtable discussions and review sessions, our 47 participants brought perspectives from the United States, Canada, Chile, Colombia, the Philippines, South Africa, and the United Kingdom. Models shared during the roundtables and review process, such as the Latin American Artificial Intelligence Index, revealed that each recommendation must be adapted by sector, geography, and audience. In particular, participants highlighted that implementation and downstream implications depend on both cultural and infrastructural realities.
The policy choices made now will help determine whether AI strengthens the scientific research ecosystem or deepens existing disparities and vulnerabilities. This playbook is intended as a practical contribution to decisions already confronting governments, research institutions, funders, technology companies, and scientific organizations.
This work is supported by Google.org.
