Affective computing is the work of making machines that read and respond to emotion. Rosalind Picard named the field in her 1997 book of the same title. A modern system usually does three things: it estimates how a person seems from their words, voice or face, it adjusts its pace, warmth and length to fit, and it remembers context so the next conversation starts where the last one ended.
The capability is useful and easy to overrate. The system reconstructs a likely feeling from evidence and produces a fitting reply. It feels nothing, and its reading can be wrong in ways the person never sees. Warmth also earns trust faster than accuracy does, so people confide in these systems on health, money and relationships long before anyone has tested how well they handle those subjects.
A sound rule for deployment: let emotion readings shape tone and pace only. Once a reading feeds a decision about a person’s job, credit or care, the system has moved from adjusting a conversation to judging a human, and the law in several jurisdictions treats that very differently.