Humanity is at the dawn of a new era in artificial intelligence, one in which machines can speak and understand natural language. This will transform the nature of machine-assisted work in many domains, cultural heritage included. Indeed, because language is a cultural artifact, cultural heritage research is particularly vulnerable to transformation by this technology. And change is happening rapidly.
Led by Damon Crockett, AI@IPCH is endeavoring to meet the urgency of the present moment with a dedicated line of research into large language models and their multimodal variants. Of particular interest is shaping the behavior of these models to better align with the epistemological and ethical commitments of the field. We aim to make progress on two fundamental questions:
- How do we evaluate the performance of generally intelligent systems in complex domains like cultural heritage research?
- Which interventions on model behavior—things like training curriculum, context engineering, architecture design, and loss/reward modeling—give us the control we need to align the models with our values?
Recent work on the nature of these values will be published in a forthcoming interactive essay, “Discourse Machines”, and an in-progress essay on the ethics of using AI for cultural heritage research.
