Hume AI
Hume AI is a research-oriented AI company founded by Alan Cowen, Keltner's former student and collaborator on the computational emotion research that demonstrated the human voice conveys at least twenty-four emotions without words. Keltner serves as Chief Scientific Advisor, bringing two decades of emotion science into direct engagement with AI development. The company's mission — ensuring AI is built to improve human emotional well-being — represents a specific theoretical position: that emotion is not noise to be filtered out of human-AI interaction but signal to be integrated, that AI systems should be designed around what humans actually are rather than around simplified models of rational utility.
In the AI Story
Hume AI emerged from the recognition that contemporary AI systems were being built without adequate grounding in emotion science. The dominant paradigm treated emotion as secondary to cognition, something to be handled through rule-based safety systems or reinforcement learning from human feedback. Cowen and Keltner's position was different: emotion is primary in human experience, and AI systems that fail to engage emotion accurately will fail to engage human beings accurately.
The company's research has contributed foundational findings to empathic AI — the discovery that facial expressions map onto at least twenty-eight distinct emotional states, that vocal expressions map onto at least twenty-four, that these mappings hold across cultures and can be modeled by machine learning. This research is the scientific foundation for systems that can recognize and respond to human emotional states — a capability with transformative potential and profound risks.
Keltner's involvement represents a specific stance in the AI discourse: engagement from within rather than critique from outside. His position is that the conditions for human flourishing in the age of AI will be shaped by those who build the systems, and that scientists who understand emotion should be at the table when those systems are designed. This engagement is costly — it implicates him in the commercial dynamics of the industry — but he has argued that the alternative (academic observation from outside) cedes the ground to actors without his framework.
The tensions in this position are real. The same computational emotion research that enables systems to support emotional well-being could enable systems to manipulate emotional states at scale. The ecology of wonder framework demands that institutions ensure the emotional intelligence being built into AI serves human flourishing rather than corporate extraction. Hume AI's mission represents one attempt to build such an institution.
Origin
Hume AI was founded in 2021 by Alan Cowen, who had completed his PhD at Berkeley working with Keltner. Cowen had previously worked at Google AI, where he applied computational methods to emotion research. The company is based in New York and has grown from a research startup to a significant presence in the empathic AI space, with Keltner's ongoing scientific advisory role providing continuity with the Berkeley emotion science tradition.
Key Ideas
Emotion as primary. The founding insight that emotion is signal, not noise, in human-AI interaction.
Twenty-four vocal emotions. The research finding that anchors the company's approach to voice AI.
Engagement from within. Keltner's position that scientists should build, not merely critique.
Structural tensions. The same technology could support flourishing or enable manipulation — the difference is institutional.
Ecology at the infrastructure level. An attempt to build ecology-of-wonder principles into AI development itself.
Appears in the Orange Pill Cycle
Further reading
- Cowen, A. S. & Keltner, D. (2017). Self-report captures 27 distinct categories of emotion. PNAS.
- Cowen, A. S. et al. (2019). The primacy of categories in the recognition of 12 emotions in speech prosody.
- Hume AI research publications, humeai.com.