Computational methods and simulations have become a fruitful methodology for philosophers, particularly for understanding how social relations, norms, and communicative structures could and should shape inquiry. Work in computational social philosophy has illuminated phenomena ranging from the epistemic value of diversity and the dynamics of epistemic belonging to the emergence and erosion of norms and processes of polarization.
Generative and agentic AI tools stand to profoundly reshape computational social philosophy in at least two ways:
As objects of study for computational social philosophy. Generative and agentic AI are increasingly embedded in epistemic life: as sources of information, interlocutors, surrogates for social participants, gatekeepers within epistemic communities, and even partially autonomous epistemic actors. This should motivate a broadening of purview in social epistemology, which has tended to focus primarily on personal, interpersonal, and institutional factors that impact inquiry, but comparatively less on technological factors.
As tools of study for computational social philosophy. These systems can lower the technical barriers to computational work. More interestingly, they may expand the range of social phenomena philosophers can model, including richer representations of agents’ beliefs and behavior, more complex interactions and environments, and even digital-twin-style models of particular communities or institutions.
This cluster brings together philosophers working in, or seriously engaging with, computational social philosophy. We are eager to gather a cohort who can bring rigor, clarity, and intellectual generosity to advancing understanding in relation to the above themes.
We’re particularly interested in philosophical contributions that engage one or more of the following cross-cutting threads:
Modeling of AI systems as socio-epistemic phenomena: How might explicitly modeling the effects of AI systems within social epistemic processes challenge, deepen, or expand existing understanding in social epistemology and philosophy of science?
Modeling of socio-epistemic phenomena with AI-based tools: What new forms of social epistemological phenomena can AI-based tools help us represent and investigate beyond traditional methods, such as typical agent-based models?
Understanding aims and tradeoffs: How should we think about the tradeoffs between promises of AI-based tools (e.g., flexibility, expressive power, scale) and other desiderata (e.g., robustness, justification, tractability, understanding) that have long been central to computational social philosophy? What can we learn about these issues from the use of AI-based tools in other disciplines such as computational social science?
Methodological standards: What standards should guide reliable, reproducible, and more generally epistemically and ethically responsible philosophical inquiry using AI-based methods in computational social philosophy?
Please note that, while we welcome expressions of interest from all disciplines, this is not a general workshop about AI or AI-based simulations. It is a workshop about how deeper engagement with AI systems as objects of modeling studies and as a modeling tool can ultimately enrich and expand philosophical inquiry.
Date: September 21, 2026
Speaker: Sahar Heydari-Fard
Title: AI and Scientific Progress
Abstract: How will AI affect science’s capacity for paradigm shifts? Using a variation of Zollman and Heydari Fard’s model of scientific communication, I show that AI can reshape scientific networks in ways that accelerate convergence and make alternative approaches harder to sustain. This can improve efficiency while undermining the conditions for transformative scientific change. I then examine several ways to mitigate this risk, including interventions in network structure, AI design, and scientific institutions. The broader lesson is that AI’s effect on scientific progress depends not only on what AI can do, but on how it reorganizes the social structure of inquiry.
Date: October 5, 2026
Speaker: Leonard Kahn
Title: Rule Consequentialism: Toward a Game-Theoretic Interpretation with Monte Carlo Simulations
Abstract: According to Rule Consequentialism (RC) the moral status of actions is determined by whether they are required, permitted, or prohibited by the rules contained in the code whose acceptance, under the theory's specified conditions, would have the best consequences. In this paper, I explore the game-theoretic and computational groundwork that can be used to improve this theory. I begin with a disputed question among discussants of RC concerning the appropriate acceptance rate for a given code when evaluating it. Advocates of fixed-rate approaches evaluate codes at a specified rate but often disagree about what that rate should be, with 100% and 90% getting the most attention. In contrast, advocates of variable-rate approaches evaluate codes across multiple rates, usually averaging the results at each of the percentage rates from 1% to 100%. This disputed question helps to expose a challenge for RC concerning how candidate rules are to be specified precisely enough for their consequences to be compared. RC theorists typically formulate candidate rules in ordinary, folk-theoretic terms, (for example, in terms of lying or harming). However, rules formulated in this way are often too vague and too disconnected from outcomes to allow us to compare the consequences of codes especially at different levels of acceptance.
My response is two-fold. I suggest that we begin by rethinking codes of rules in terms of game-theoretic strategies because game-theoretic strategies can represent candidate rules with greater precision, while game-theoretic models make the consequences of their interaction more explicit. I follow up on this suggestion by constructing a simple Monte Carlo simulation in which interactions among these strategies can be evaluated at various rates of acceptance. (For reasons of tractability, I study competing rules governing one recurrent interaction, while acknowledging that whole-code evaluation and interactions among rules remain downstream problems to solve.). My test case in this paper is the Stag Hunt. My simulation compares five strategies across ten acceptance rates. It includes two simple strategies (Always Stag and Always Hare), one randomized strategy (Two Face), and two adaptive strategies (Cooperative Adapter and Empirical Best Responder). My results suggest (1) that a comparatively simple cooperative rule can perform remarkably well against the alternatives tested across a broad range of acceptance rates and (2) that, under the baseline payoffs and the present method of aggregating results, high fixed-rate and variable-rate approaches can yield surprisingly similar evaluations of candidate rules. My results illustrate how game theory and computational modeling might begin to clarify questions about rule specification and acceptance rates in RC as well as provide answers to these questions, though a long road remains ahead of us.
Date: October 19, 2026
Speaker: Elias Brockman [Joint work with Ted Rubin]
Title: Generative AI, Interpretive Provenance, and the Modeling of Philosophical Systems
Abstract: Can AI-assisted formal modeling produce genuinely philosophical discoveries? Accounts divide between reliabilism, which permits reliance on opaque but reliable processes, and skepticism, which denies that systems without truth-oriented agency can participate in discovery. We argue that these positions conflate hypothesis generation, interpretive justification, and epistemic agency. Using fragmentary evidence from early Greek philosophy, we show how generative AI can expand the space of candidate reconstructions without determining which are historically or philosophically adequate. Researchers evaluate proposals through textual evidence, source criticism, competing interpretations, and judgments about contested standards. Formal modeling then tests consequences of reconstructions, revealing implications, equivalences, unsupported assumptions, and missing distinctions obscured in prose. These results redirect inquiry toward the evidence, creating a recursive process of interpretation, implementation, diagnosis, and revision. We call the record of this process "interpretive provenance": documentation connecting evidential assessments, AI proposals, human judgments, formal rules, consequences, rejected alternatives, and conditions for reconsideration. Interpretive provenance does not itself justify a reconstruction; it makes its generation and justification available for criticism, comparison, reproduction, and revision. AI-assisted philosophical discovery is therefore neither machine knowledge nor reliably generated output, but accountable inquiry in which computational systems help researchers discover novel, defensible possibilities they might otherwise have overlooked.
Date: November 2, 2026
Speaker: Chaoqun Ni
Title: Broader Science, Narrower Roles: Scientific Expertise in the Age of AI
Abstract: Modern science depends on specialization and teamwork, with researchers combining complementary expertise through collaboration. Large language models (LLMs) offer a new way to access knowledge and methods across fields, raising questions about how scientific expertise is now assembled. Drawing on publication histories of over 775,000 scientists and contributor statements from more than 137,000 papers, we examine changes in what scientists work on, whom they work with, and how teams divide their work during the period of widespread LLM diffusion. We find that research portfolios broadened, collaboration networks became more interdisciplinary, and individual responsibilities within teams became more specialized. These patterns were more pronounced among authors with stronger AI-associated writing signals. Our findings point to a shifting relationship between the breadth of scientists' research and the work they perform.
The series will consist of bi-weekly virtual meetings, in the style of CSPS, starting in October 2026. The meetings are on Mondays at 11:30am ET.
You can submit your application to join the seminars by using this application form: https://forms.gle/SC2iqbZUBU6iB52v6
The form includes:
Basic details: Your name, email address, affiliation, career stage
Extended abstracts of up to 750 words and prepared for anonymous review (no identifying information in the abstract itself).
The submission deadline for applications is July 27th, 2026. p
Sina Fazelpour (Northeastern)
Luca Garzino Demo (UPenn)
Please direct inquiries to lgarzino@sas.upenn.edu