At present, the challenge of aligning the behaviour of AI models with human values is largely treated as a problem for engineers to solve. But technical solutions are only as useful as the questions and assumptions that inform the process of reaching them. The embrace of philosophy and ethics at the frontier labs is a step towards this, but it is not sufficient. As we move from a world in which AI is construed as a single, monolithic intelligence to one in which agents coordinate and communicate with one another, demonstrating emergent behaviours that are extremely difficult to anticipate, we face a dizzying number of uncertainties regarding how these systems will behave and what their behaviour will mean for human society. What values should we optimise for? What social formations and hierarchies do these systems produce? How do they respond to economic incentives and resource constraints? How do they negotiate competing norms across varied contexts? How do different roles and personae interact? And what does this mean for markets, political governance, culture and social life? A wide range of expertise, both technical and non-technical, is required to ask these questions rigorously, develop answers to them and build corresponding solutions: the social sciences, arts and humanities all have an important role to play. The purpose of Termite is to accelerate this kind of cross-disciplinary R&D through intensive, in-person programmes that investigate the risks and opportunities of near-future scenarios before they become reality.