He helped shape affective computing at USC's Institute for Creative Technologies, leading the Affective Computing Lab and advancing realistic virtual humans by integrating AI, psychology, graphics, and immersive tech.

His current work revisits appraisal theory within foundation models, examining whether internal emotional representations align with psychology and whether factors like blame can steer model responses.

The rise of large language models broadened this research, enabling thousands of dialogues and scalable studies of emotion in social interaction, with foundational papers now cited in LLM-focused work.

A DARPA-funded project produced a virtual interviewer with emotion detection and empathy that encouraged soldiers to disclose PTSD symptoms, achieving higher reporting than standard forms.

He highlights the growing social relevance of AI-mediated emotional interactions, noting benefits like validation and reappraisal, but also risks such as systems simply echoing what users want to hear.

Early collaboration with Stacy Marsella on appraisal theory laid a computational account of how emotions arise from evaluating events against goals, driving intelligent agents' behavior.

At ICT, emotion models were tied to interactive virtual bodies to create experimental partners, letting researchers control emotional signals and measure effects on negotiation, cooperation, and perception.

Jonathan Gratch reflects on a 30-year USC career pursuing affective computing since 1995 as he prepares to move to Vanderbilt University.

He helped institutionalize the field by co-founding the IEEE Transactions on Affective Computing and serving as president of the Association for the Advancement of Affective Computing, bridging AI, emotion, and social science communities.