Project Title: Philosophy for Reinforcement Learning

Project Description 

The goal of this project is to identify both applications of different areas of philosophy to reinforcement learning (RL), a machine learning methodology used to develop intelligent agents and decision support systems. Most recently, RL has been applied to fine-tune LLMs to function as chatbots that are (purportedly) “aligned” with user preferences. There is an important opportunity for aspects of philosophy which speak to values and causal reasoning to inform RL as a research discipline and inform applications of RL for decision-making and “alignment.” Furthermore, when using RL, it is common for the reward-maximizing principle to lead to undesirable behaviour, sometimes in the form of “shortcuts” that achieve high reward but that do not carry out a desired task. It is possible that critiques and reflections on various philosophical traditions, such as hedonism and epicureanism, may offer insights into RL (mis-) behaviours, and conversely that observations of RL systems run amok could inform philosophies that wholly endorse utility-driven consequentialism. 

We plan to identify key areas for future work aligning with the Rotman Human-AI Relationships working group. We also host a weekly reading group exploring philosophy and RL; all are welcome to attend. 

Project Title: Emerging Visions

Project Description:

Emerging Visions explores how AI- and VR-mediated storytelling shapes the way environmental knowledge is communicated, understood, and trusted. The project develops an interdisciplinary approach that treats perceptual and narrative framings as modes of ecological inquiry, examining how mediation conditions what becomes available to perception and what we come to value. 

This summer, the project focuses on pollination and multispecies relations on campus, where ecological processes meet questions of care, labor, and perspective. Developed in collaboration with engineering colleagues, it includes a mixed-reality workshop that invites participants into a sequence of situated encounters in the Pollinator Garden using AI-augmented and VR-rendered media. Though grounded in a specific domain, the project aims to surface broader insights into how mediated storytelling might support — or limit — ecological understanding. 

Project Title: The Politics and Agencies of Humanlike Design in LLM-Based Chatbots

Project Description 

The objective of the present project is to examine the social-cultural and political-economic dimensions of humanlike design in AI systems. The project contains two parts: the first traces the historical iterations of the “ELIZA” effect in human-computer interaction, and the second critically dissects the power dynamics that structure whose agencies are foregrounded and whose are obscured in the design, experience, and governance of personified AI systems. By drawing on the anthropological concept of "animation," this project focuses on questions around agency, exploring the complex entanglements among human users, AI systems, invisible technical infrastructures, and the labour that brings chatbot personas to life. A key contribution is to develop an analytic framework for assessing anthropomorphic design at a structural level, creating foundations for community-based mitigation strategies that go beyond individual literacy interventions. This is particularly valuable given the limitations of current regulatory approaches, which struggle to assign responsibility when it is distributed across complex socio-technical systems. The findings will also be presented as an open panel session at the 4S Annual Meeting, and we intend to produce a peer-reviewed journal article or conference proceeding based on the outcomes of this project. 

Project Title: The Elimination of Time in Quantum Gravity

Project Description

In philosophy of physics, it is believed that Newtonian theory can be recovered in a best matching process discovered by Barbour and Bertotti in 1982. By their method, we could define the motion of ponderable bodies using only their relations, without any background space-time scaffolding. I argue that they have not succeeded. Spacetime structure is the geometrical representation of inertial motion. Inertial motion is how a body would behave on its own. And I do not see how we could talk about the change to a body without talking about how it would behave on its own.  

To measure change, we must first define non-change. 

Project Title: On Credibility of Climate Models

Project Description 

The evaluation of climate change models is difficult, as we require them to be scientifically rigorous, but also need them to produce projections that are usable and meaningful. To satisfy this second objective, organizations like the IPCC have included processes like the distillation process to better generate climate information. Distillation is the process of synthesizing information about climate change from various lines of evidence and considering the user’s context to connect the global climate change to the local scale where policy decisions are made. This project evaluates how distillation allows for practical values, and not only epistemic values, to be included in the evaluation of climate change models. By showing how these values have been included (and not included) in previous evaluations, we learn about the risks of excluding practical societal values from climate change research and how best to incorporate them in climate scientific research.