Societal AI Research Has Become Less Interdisciplinary

Computer science teams are now the primary drivers of societal AI research: While interdisciplinary teams remain more likely to produce societally-oriented work, CS-only teams now account for a growing share of the field's overall societal output, challenging the assumption that interdisciplinary collaboration is the main pathway for integrating ethical considerations into AI development.

Technical researchers are increasingly addressing diverse societal concerns: CS-only teams are tackling a wide range of domains including fairness, safety, healthcare, and misinformation, suggesting that technical communities are becoming more responsive to calls for responsible AI without necessarily partnering with social scientists or humanists.

The shift raises critical questions about AI governance and safety: The paper highlights an important tension: if most societally-oriented AI research is now conducted by exclusively technical teams, this may affect how emerging understandings of AI safety and governance develop, potentially limiting perspectives from social sciences and humanities disciplines.

Interdisciplinary collaboration may be declining in practice despite being championed: The study of over 100,000 ArXiv papers from 2014-2024 reveals a "striking shift" away from interdisciplinary approaches, even as experts continue to advocate for cross-disciplinary work as essential for responsible AI development.

The field faces a potential expertise gap: The findings suggest that as technical teams take on more societal concerns, there may be a growing disconnect between what social scientists and humanists can contribute and where the actual research momentum is occurring in AI development.