Hume AI
AI must account for human emotion to ensure that it is built to serve our goals. However, based on traditional theories of emotion, biased measures, and data rife with stereotypes, machine learning has long overlooked the bulk of what we express and perceive in real-world emotional behavior. Hume introduces a new, data-driven scientific approach to emotion that will enable the development of truly empathic AI. With large-scale studies, Hume is collecting richer, more balanced, more diverse, unbiased emotion data by providing a degree of experimental control over the emotions experienced and expressed by participants around the world. Our team has used this data to train facial and vocal expression models that are much more accurate, nuanced, and generalizable than any that have previously been developed. These discoveries set the stage for fundamental advances in affective science and empathic technology and pave the way for AI that verifiably serves human emotion.