The National Academies of Science, Engineering, and Medicine (NASEM) convened for a two-day workshop titled “Enhancing Scientific Integrity: Progress and Opportunities in the Social and Behavioral Sciences.” Hosted by the Board on Behavioral, Cognitive, and Sensory Science (BBCSS) on April 23 and 24, the event brought together researchers, funders, publishers, and others to discuss how social and behavioral sciences can strengthen research integrity, transparency, and collaboration, especially as the challenges of artificial intelligence (AI) evolve.
Day 1
Daniel J. Weiss, PhD, BBCSS Board Director, opened the workshop by laying out three guiding questions: (1) how stakeholders can adopt effective frameworks for transparency, (2) how journals can refine their review policies to sift out unverified data, and (3) what barriers impede journals and researchers from upholding data integrity goals?
Pamela Davis Kean, PhD, University of Michigan, moderated the two days, starting with factors that undermine the integrity of research practices. Klass Sijtsma, PhD, professor emeritus at Tilburg University, reviewed common warning signs which might indicate a breach in scientific integrity within research environments. These include isolating students, opposing collaboration from peers and colleagues, and producing overly linear data. While extreme cases of fabrication, falsification, and plagiarism (FFP) are not common, more pervasive errors originate from questionable research practices (QRP) and structural weaknesses. He emphasized that integrity challenges often stem from systemic issues rather than isolated misconduct, and suggested several approaches groups can adopt to mitigate them:
- Mandate researchers to preregister their research to prevent data dredging and post hoc hypothesis manipulation.
- Promote data transparency.
- Encourage collaboration across specialties.
- Promote institutional oversight by establishing internal scientific review committees.
John Ioannidis, PhD, Stanford University, expanded on this point, detailing how the increasing use of AI in data analysis is introducing new uncertainties into the research process, especially in competitive environments where research quantity outweighs quality. To remain ahead of these issues, scientists and institutions should strengthen education in research methodology and design, encourage research preregistration, and reinforce the ethical use of AI, especially as a new generation of researchers enters the field.
Day 2
The second day of the workshop turned to proactive mitigation strategies, with speakers focusing on fraud detection, publishing practices, and the systemic incentives that shape scientific behavior. To improve the detection of fraudulent research activity, the first speaker of the day, Uri Simonsohn, PhD, Ramon Llull University, introduced his online platform AsCollected, created at the Wharton Credibility Lab. The tool documents “results provenance”—tracking how, when, and by whom data is collected, processed, and analyzed. These efforts aim to reduce fraud by establishing a system of transparency and accountability while supporting reproducibility.
The first panel, featuring Ivan Oranksy, MD, Retraction Watch; Elisabeth Bik, PhD, scientific integrity consultant; and Leif Nelson, PhD, University of California, Berkeley, explored how paper mills and AI-generated content amplify the potential for scientific misconduct. Speakers warned that as misconduct grows, public trust will continue to erode. The panelists proposed solutions centered on prioritizing reproducibility, strengthening oversight, and enforcing consequences for research misconduct.
In her presentation, Simine Vazire, PhD, Editor-in-Chief at Psychological Science, emphasized that while transparency is essential for scientific credibility, it is difficult to implement effectively and does not always ensure reproducibility. Looking at research from Nicholas Hardwicke, PhD, University of Sydney, she noted a rise in data transparency and preregistration since the start of psychology’s “reproducibility crisis,” noting that significant challenges remain. Vazire attributes part of the problem to journals’ limited capacity in detecting errors during peer review and the prestige-based ranking system in publishing. She concluded that while transparency is not an automatic fix, if paired with organized skepticism and stronger evaluation systems before and during the publication process, reproducibility can be improved.
Reproducibility is hard to achieve, especially when statistical methods are misused and cross-collaboration is underencouraged. In an interview led by Simonsohn, Andrew Gellman, PhD, University of Columbia, discussed improving statistical practices and data transparency. He noted that researchers often fall into the trap of using older methods that can misrepresent their data, while modern approaches can produce more accurate data analysis.
The next panel, featuring editors Stavroula Kousta, PhD, Nature Human Behavior, and Magdalena Skipper, PhD, Springer Nature, focused on research integrity in publishing practices. They encouraged using pre-prints and strengthening the review process. While journals have adopted preemptive strategies, the system continues to fall short due to expensive, inequitable, and slow operations.
Alongside publishers, universities play a key role in strengthening transparency and research integrity. Arthur “Skip” Lupia, PhD, University of Michigan, argued that academia needs to strengthen incentives to disclose methodology. He recommended that universities expand credit to all research contributors, not just primary researchers.
In his fireside chat, Jay Bhattacharya, PhD, Director of the National Institutes of Health (NIH), also discussed incentives and their influence on scientific behavior. He argued that the common metrics of publication volume and citation influence are too narrow to support collaboration and replication. In an effort to support replication research, Bhattacharya stressed the need for funding structures and evaluation systems that reward openness and innovation. NIH is working to establish stronger metrics of scientific productivity, release new funding opportunities for replication research, launch a replication journal, and enhance the PubMed search engine to include a knowledge graph of related research. Bhattacharya said, “All of that, in the works. We will have that this year.”
A concluding panel offered valuable insights on the ethical issues and possible reforms in publication, funding, and tenure review systems. Steve Smith, STEM Knowledge Partners, moderated a conversation featuring Michael Dougherty, PhD, University of Maryland (COGDOP representative to the FABBS Council); Andrew Bacher-Hicks, Arnold Ventures; and Michelle Meyer, PhD, Geisinger College of Health Sciences. The discussion focused on how incentives at institutions shape research practices, with data sharing identified as a practical starting point for reform. Dougherty covered the successful policy changes in the Psychology Department at the University of Maryland, where improved data sharing practices and established positive incentives support open science. Bacher-Hicks noted that funders can also support transparency by funding replication studies and requiring replicability goals in grant proposals. Meyer discussed how institutional review boards (IRBs) can take a more proactive ethical approach by using consent processes to support responsible data sharing and research integrity.
Throughout the two days, discussions consistently circled back to a central theme: while new technologies such as AI present major challenges, they are not the only cause of structural weaknesses within research review systems. Instead, technology amplifies vulnerabilities that already exist. Confronting these challenges is more urgent than ever. Making headway will require sustained investment in education, strengthened policies, and a collective effort to realign incentives to foster a research culture grounded in trust.
