Interdisciplinary Symposium on AI and Society — a report
In June 2025, researchers gathered at the University of Warwick for the Interdisciplinary Symposium on AI and SocietyLink opens in a new window. The 2-day event was designed to foster interdisciplinary exchanges between scientists and scholars in computer science, statistics, sociology and other social sciences and humanities disciplines.
Supported by two University of Warwick Research Spotlights — Society & CultureLink opens in a new window and Digital, Data Science & AILink opens in a new window — the symposium brought together leading international researchers in this area with academics from 8 different departments at Warwick, and was structured around different themes and provocations, such as the future of theory in an age of machine learning, the question of scale in social science & the power of models. All sessions, which included plenaries, debates and round tables, were designed to enable exchanges across disparate fields, career stages and backgrounds.
The opening plenary session Modelling the Social started the conversation with big questions: How much of social behaviour can be captured by mathematical laws? Is modelling mainly a methodological challenge, or is social life too contingent to be reduced to models? And what roles can machine learning models play both as theory-informed frameworks and as sources of new theoretical insight? Speakers offered distinct perspectives from computational social science (Dirk Helbing), political science (Giuditta Fontana), and philosophy/sociology (Steve Fuller). What seems to have united all speakers is a shared recognition of the need to engage with the methodological limits of modelling precisely at the time that new forms of machine learning are expanding its capacities. As Dirk Helbing alluded to in the open discussion, technical innovation can only offer solutions for societal problems when combined with social innovation. Giuditta Fontana’s discussion of using machine learning methods in civil war research offered a robust reminder that while machine learning can be useful for generating new insights about social phenomena, many aspects of social life are too complex to be modelled.
Steve Fuller’s talk took us back to the University of Pittsburgh, where both Herbert Simon and Geoff Hinton worked at the time, when he was a PhD student. He commented that models inscribe ontological assumptions about society, such as methodological individualism and rational action, into social science and its applications and thereby offered a useful prelude to the Social Physics or an interpretative turn? debate, which saw an interdisciplinary group of speakers from sociology and computer science — Carrie Friese (London School of Economics), Paolo Turrini (University of Warwick), Justus Uitermark (University of Amsterdam, and Theo Damoulas (University of Warwick) — respond to four debating motions formulated by the symposium organisers regarding the future of social science in an age of AI.
A couple of points of agreement and disagreement stand out. First, while all speakers agreed that we need to re-examine the perceived distinction between qualitative and quantitative methods to understand the relevance of AI to social science and society, there were divergent takes on how to think about researcher bias: some computational researchers argued for the importance of de-biasing models used to study social life, while interpretative social researchers embraced situatedness of knowledge and suggested that one promising aspect of using AI is its potential to surprise a researcher by providing new ways of seeing and opening up alternative perspectives. Relatedly, the discussion about whether machine learning, especially so-called Large Language Models (LLMs), will transform social research raised questions about the level and unit of analysis: Carrie Friese, for example, expressed an interest in AI as a method to scale up the analysis of textual data while others – and this also came up in the Scientific, Causal and Agentic AI stream – defined the potential of LLMs in terms of their ability to add interactional detail to (multi-agent) models by assigning, for example, a “persona” to an agent.
We returned to the question of scale in the Scale/Multiscale/Multiscalar/Abstractions plenary on day 2. This session explored how the notions of abstraction and scale inform our understanding of socio-technical systems. The speakers shared distinct perspectives on the topic: Sander Beckers focused on causal abstraction in computational social science, arguing that social constructions, such as race, are still amenable to rigorous causal analysis; Joel Dyer explored the potential of multi-scale modelling in social science and policy analysis; Matt Spencer argued that it would be analytically fruitful to attempt to bridge social theory and interventionist theory of causation through field study of multi-scalar entities, such as computer viruses. In the ensuing debate, the notion of intervention gained traction: how to justify the selection of the context of intervention? Sander Beckers commented that it depends on what a researcher is interested in, but it also depends on what scale of analysis is favoured by our disciplinary conceptual and methodological toolkits.
The late morning was dedicated to stream plenaries, including a panel about Health & AI, which hosted presentations by Busola Oronti (Statistics) about predictive diagnostics, Emily Rowe (WBS) about AI and the efficiency drive in the NHS, and Davide Piaggio (Engineering) about frugal innovation. Each in different ways, these presentations demonstrated that in what some define as “downstream” settings, some of the most significant challenges and implications of AI and its inscribed objectives of “optimization” and the “maximization of the utility function” for society become apparent. This includes the key question of practical credibility, or lack thereof, of any possible fit between AI and the requirements placed on innovation in low-resource settings.
To conclude the symposium, three speakers reflected on How to envision society with AI? Justus Uitermark distinguished between two different logics of complexity underpinning the digital (e.g. social media) and -- at least for a long time -- non-digital (e.g. cities) spaces; Nick Gane drew our attention to political economy of AI by centering the ideal of the sovereign individual and the question of who funds AI; Michael Castelle put forward the notion of “language ideology” inscribed in LLMs and the associated challenge to long-standing assumptions in computational linguistics, such as that linguistic meaning is largely syntaxic During the discussion, the question of values – which was also raised in the opening plenary session and streams on politics and agent research – gained urgency: What values should guide the development of (responsible) AI methods in social science? How should we do research with – not necessarily for – the public in understanding the relation between (mostly proprietary) AI and society? And, perhaps most importantly in the context of this symposium, how do we facilitate productive interdisciplinary exchange that is based not on extra-disciplinary criticism but grounded in shared problematics?
In their introduction to the symposium, the organisers suggested that it is important to suspend some disciplinary assumptions in order to develop the “infra-language” (Galison, 2010) that is necessary for inquiry across disciplines. But perhaps it is as important to develop "careful research" (Law and Lin, 2022) practices in which disciplinary differences are not sidelined – or silenced – but approached as generative tensions that should be worked with – not against – to engage with questions at the intersection of AI and society.
The Interdisciplinary AI & Society Symposium also served as an inaugural event of the Society, Economics, and TechnologyLink opens in a new window research cluster initiated by the sociology department in collaboration with CIM.
REFERENCES:
Galison, Peter. 2010. “Trading with the Enemy.” In Trading Zones and Interactional Expertise: Creating New Kinds of Collaboration, edited by Michael E. Gorman. MIT Press. https://direct.mit.edu/books/edited-volume/2155/chapter/57431/Trading-with-the-Enemy.
Law, John, and Wen-yuan Lin. 2022. “Care-Ful Research: Sensibilities from Science and Technology Studies (STS).” In The SAGE Handbook of Qualitative Research Design, edited by Uwe Flick, 127–41. SAGE Publications Ltd. https://doi.org/10.4135/9781529770278.
The Interdisciplinary AI and Society symposium was co-organised by Greta Timaite, Noortje Marres, Yorgos Felekis, Federico Perlino, Theo Damoulas, and Michael Castelle
Photos: Luke Robert MasonLink opens in a new window
