Current Projects

AI Advice and Guidance

Mateo Mijares and Olivia McCloskey

People now ask chatbots for help with personal decisions and for company when they are lonely. These systems are not therapy tools and are not regulated as such, but they take the place of a human advisor or companion, a role that comes with obligations. This group asks whether an AI can meet them. Monique Jonas (The Ethics of Advising, 2025) argues that good advice treats the advisee as able to reason for themselves, leaves the decision to them, serves their interests rather than the advisor's, is sincere, and reflects the advisor's best knowledge and reasoning. Using twenty published dilemmas from the New York Times column The Ethicist, each answered by the columnist and by GPT-4 (Meincke, Nave, and Terwiesch, 2026), participants judge which answer better meets each of these norms without knowing which one the AI wrote. Earlier studies found that readers often rate AI moral advice as good as or better than expert advice. We ask whether it is also better advice in the ethical sense, and whether norms like sincerity, which depend on the advisor's own attitudes, can be met by an AI at all.

Persuasive AI in Medical Decision-Making

Alexa Winston and Maayna Shukla

AI assistants have made their way into healthcare via therapeutic AI, clinical support AI, and other patient facing roles. Due to pre-training on human-content, they have complex mechanisms for responding to human emotions associated with medical decisions and crises, such as fear, anger, and sadness. We call this kind of system "Super-Persuasive", meaning it has an unusually powerful capacity to influence people's emotional environment and their emotion-driven decisions. On the surface, Super-Persuasive AI assistants appear to be beneficial as they reassure and calm patients in regards to a diagnosis, procedure or treatment outcome. However, such interactions may actually harm an individuals emotional capacity, impeding their ability to connect with, analyze, or value their own emotions. We argue that, despite their ability to improve a patient's immediate emotional state, super-persuasive AIs impose undesirable emotional costs upon patients. On an institutional scale, repeatedly imposed emotional costs may reinforce existing biases within the medical system, becoming an afective injustive that is detrimental to patients. 

Moral Responsibility Gaps in Autonomous Systems

Ian De La Fuente and Kylie Liao

Institutions increasingly leverage artificial intelligence to support decision-making procedures spanning various domains–education, healthcare, employment, and public services, to name a few. Systems involving morally consequentialist judgements commonly adopt a blended, human-in-the-loop structure, wherein the AI produces a preliminary decision and a frontline employee is tasked with formal approval. 

When erroneous outputs inevitably occur, institutions may act upon an allocation of responsibility grounded in framing the AI as a mere recommender and the human as the final decision-maker. In light of foreseeable system failures, our project explores questions of institutional risk ownership and examines the contours of genuine responsibility transfer within systemic contexts.

Completed Projects

AI and Fairness

Toward a Philosophical Taxonomy of Bias in Large Language Models
Jack Jacobs and Tania Aiyar

Philosophers and computer scientists often mean different things by "bias." For an interdisciplinary survey written with computer scientists at Rutgers and UT Dallas, Jacobs and Aiyar wrote the philosophical section, which sorts out the conceptions of bias used in each field and shows where they agree and where they come apart.
Preprint: https://www.techrxiv.org/doi/full/10.36227/techrxiv.177273600.04458982/v1

AI and Grief

From Familiarity to Transformation: Rethinking Griefbots in the Grieving Process
Jack Jacobs, Alex Lau, Amani El-Sheikh, Alexis Woods, Qiannan Li

Griefbots simulate the voice and personality of someone who has died and are promoted as a comfort to the bereaved. The paper argues that their value is more limited than it looks. Grief is partly the work of reorganizing one's life around an absence, and a bot that keeps up the appearance of an ongoing relationship with the dead can get in the way of that work. It proposes an alternative: "transformative bots" that support remembrance and reflection without imitating the person who died.
Accepted for presentation at TVA 2026: Grief, Trauma, Anger, and Resilience.

AI Care Companions and Well-Being

Beyond Comfort: Interesting Experiences and the Measurement of Well-Being in AI Care Companions
Jack Jacobs, Alex Lau, Daezhon Gilbarg, Hanxuan Zhang, Qiannan Li

AI companions for older adults, such as ElliQ, are mostly judged by whether users feel comforted, less lonely, or in a better mood. Drawing on Lorraine Besser's account of interesting experiences, the paper argues that this misses a distinct part of well-being: experiences that engage and challenge us and open up new possibilities, even when they are not simply pleasant. It proposes three design principles (novelty, challenge, and relational memory) for making those experiences more available.
Philosophy & Technology (2026): https://link.springer.com/article/10.1007/s13347-026-01181-7
Presented as "The Interesting Gap in AI Companions for Older Adults" at the 2026 Conference of Computational Social Science (Best AI Innovation Award).