Grounded Models: The Future of Sensemaking in a World of Generative AI

Abstract

The promise of generative AI technologies is seductive to product leaders: frictionless research in which synthetic data can be both generated and analysed via a simple end‐to‐end UI, enabling teams to speed up research timelines and reduce costs. However, our evidence suggests we should be sceptical of these maximalist claims. Over the last 18 months our combined team of NLP data scientists and ethnographers has conducted a series of experiments to explore, assess and define the value of LLM‐driven research techniques. First, we explore this value pragmatically, as new tools for sensemaking; and second, epistemologically, as we unpack their broader implications for ethnography. We demonstrate how ethnography can usefully “ground” LLMs in two “complex” worlds: that of the user and that of the organisation. We argue the future of research is not automation, but more collaboration between ethnographers and data scientists, as they better integrate their tools and ways of knowing.