Computational Affective Intelligence
What is CAI
Computational Affective Intelligence
A research framework that treats emotion computationally all the way through an AI: recognition, reasoning, generation, internal representation, and social interaction.
01 — Definition
Definition
Computational Affective Intelligence is the study of how affect is represented, inferred, generated, learned, and utilized by artificial intelligence systems.
Affective Computing has traditionally been built around human-centred emotion technology: detecting emotion, responding to it, expressing it. CAI takes in more than human emotion. It also asks how an AI expresses, interprets, reasons over and generates emotion, and how it holds the uncertainty and the value judgements that come with it.
02 — Lineage
Not another name for emotion recognition
What is usually called emotion AI has mostly meant the two inner boxes below. CAI contains them, and widens the question outward.
Computational Affective Intelligence
Goes on to ask how an AI understands emotion, how it structures it inside, and how it uses it in reasoning and decisions.
Affective Computing
Builds computational systems that recognise, generate and respond to emotion.
Emotion Recognition
Guesses the emotion.
03 — Why now
Why the LLM era needs it
The central problem used to be estimating an emotion label from a face, a voice or a sentence. Now the LLM itself reasons about emotion, reaches different emotional judgements depending on persona, values and culture, and produces emotional responses of its own.
Recognition alone no longer covers that. Emotional intelligence itself has to be studied as an object of computation.
And once emotion surrounds society like an atmosphere — what we call the Affectosphere — the AI that reads it has to be designed as an intelligence, with its uncertainty and its limits in view, not only as a sensor.
04 — Questions
What CAI asks
- 1 How is human emotion represented inside an AI?
- 2 How should an AI handle the ambiguity of emotion labels, and the differences between people?
- 3 On what grounds does an LLM judge anger, sadness or empathy?
- 4 What structure, and what bias, sits in the emotional expression an AI produces?
- 5 Is there any sense in giving an AI emotional states or emotional reasoning?
- 6 How should emotional interaction between people and AI be designed and evaluated?
- 7 What does the technology do to society, ethics and culture?
05 — Structure
Four tiers
Understanding human emotion, holding it computationally inside an AI, letting the AI reason and act with it, and placing the result among people. Under each tier: the lab's domains that work there, and how many of our publications sit on it.
-
I
Human Affect
Understanding human emotion itself.
Stages
Human Affect
Our domains
1 publications
-
II
Computational Representation
Giving it a form a machine can hold and compute over.
Stages
Perception · Representation
Our domains
12 publications
-
III
Affective Intelligence
Where the system reasons, generates and decides with it.
Stages
Reasoning · Generation
Our domains
16 publications
-
IV
Human–AI Society
What all of it does between people, and in society.
Stages
Interaction · Society
Our domains
- Ethics and Philosophy of CAI
- CAI and Human Interaction
- CAI and Business
- CAI and Art
- Development Based on CAI
12 publications
Each publication is counted once, at the stage it contributes to most.