A qualitative shift in AI capabilities: A “bitter lesson” for human-AI interaction research?

Richard Sutton’s “Bitter Lesson” (2019) describes how prior knowledge-driven approaches to AI design were displaced by domain-general methods that scale with data and computation, rendering many prior insights obsolete. The technologies this shift enabled, including large language models with native multimodal capabilities, may now reshape human–AI interaction (HAI). Contemporary systems capable of fluid, expressive, open-ended interaction differ not only in degree but in kind from earlier technologies. This paper argues that HAI research may face its own “bitter lesson”: findings derived from interaction with earlier, technically constrained, systems cannot be assumed to generalize without replication and historical contextualization. It situates these developments in relation to methodological strategies and the challenges posed by rapid technological change, while highlighting how contemporary AI systems can offer a new lens on human connection and the expression of intelligence.