A computational model of emotion — so technology can understand and respond to people, not just data.
Purpose
Emotion is not noise on top of thought — it shapes attention, memory, and every decision we make. Software that ignores it will always misunderstand the people it serves.
The Human Emotion Engine is 42.AI's effort to model human emotion computationally: to recognize how someone feels, interpret it in context, and respond appropriately — while taking seriously how hard, and how easy to get wrong, that problem really is.
Try it: the shape of emotion
Psychologists often map emotion on two axes — how pleasant it feels, and how much energy it carries. Drag the dot (or use arrow keys) to explore. An illustrative model, not a diagnosis — it implements the circumplex model of affect (Russell, 1980).
Excited
pleasant · high energy
valence +0.42 · arousal +0.42
Research & writing
Foundations
The idea that computers should understand emotion
In 1997, Rosalind Picard of the MIT Media Lab named and defined the field of "affective computing" — computing that relates to, arises from, or deliberately influences emotion. Her argument broke with the assumption that feeling was the opposite of intelligence.
Drawing on neuroscience showing that people who lose emotional processing also lose the ability to make sound everyday decisions, Picard argued that emotion is functional, not decorative — and that machines meant to work fluently with humans would need to perceive and reason about it. That claim is the founding premise of the Human Emotion Engine.
It is tempting to assume a scowl means anger and a smile means happiness. In a 2019 review of the evidence, Lisa Feldman Barrett and four co-authors examined a large body of research and concluded that facial movements are far more variable and context-dependent than that: the same emotion can produce many different expressions, and the same expression can mean many different things.
The lesson is a caution the Human Emotion Engine is built around. Reliable emotion understanding cannot come from a single signal read out of context; it has to combine multiple cues, weigh situation and culture, and stay honest about its own uncertainty. Systems that skip that step don't just underperform — they mislead.
Reference
Barrett, L. F., Adolphs, R., Marsella, S., Martinez, A. M., & Pollak, S. D. (2019). Emotional Expressions Reconsidered: Challenges to Inferring Emotion From Human Facial Movements. Psychological Science in the Public Interest, 20(1), 1–68. https://doi.org/10.1177/1529100619832930