2026 workshop program
Including AI Agents on Scientific Teams: Some Open Ethical Questions
The work
Abstract
Artificial agential systems are increasingly being integrated into scientific research, anywhere from ideation and experimental design all the way to data analysis and full manuscript writing. I will sketch some possible risks to science that might come with integrating AI agents into scientific workflows. Arguably, if their integration is done carefully and reflectively, many such risks might be effectively mitigated. Unfortunately, it’s very difficult to know what careful and reflective integration entails. Afterall, these tools are very new and rapidly evolving, established scholars weren’t trained on how to do good science with them, and even the model builders themselves don’t fully understand their capacities and limitations. Furthermore, the incentive systems in science arguably generate a “tragedy of the commons”, where indiscriminate use of AI agents provides individual scientists with distinct advantages, but where widespread indiscriminate use would likely damage whole scientific communities. During the discussion, we’ll brainstorm possible solutions.
The speaker
Biography
Matt is the Program Director for Embedded EthiCS at Harvard, an educational program that embeds short modules on the ethical and social implications of computing technologies into courses across Harvard’s CS curriculum. Originally trained as a generalist philosopher of science, his current research focuses mostly on issues related to computing technologies, like online content moderation, AI companions, and big tech regulation. He’ll be teaching a course at Harvard College next fall entitled “Artificial Agency and Society”.