Human–AI Interaction (HAI)
What is Human–AI Interaction (HAI)?
When a person uses an AI, what happens beyond the screen?
Human–AI Interaction (HAI) is the subfield of HCI that studies what passes between a person and an AI system. It goes past whether a button is easy to press, and asks what a person expects of an AI, how they explain it to themselves when that expectation is broken, and how they change their own behaviour in response.
Why classical HCI is not enough on its own
A classical interface can be specified. The designer can write down in advance what a button does, and most usability problems live in the gap between that specification and what the person expected.
An AI is not. The same input can produce a different output, and a model change can break today what worked yesterday. So a person handles the system by inference rather than by specification — does this AI remember me, is it thinking or is it broken, can I trust this answer. HAI takes that inference as its object of study.
What it looks at
Expectation and how it forms. Trust calibration — how a person drifts toward over- or under-reliance. Attribution of intent — people assign intent to probabilistic output. Interpretation of failure — a missing answer is sometimes more uncomfortable than a wrong one. And time, because a relationship is built by repetition rather than by a single exchange.
How I use it
What comes out of the research does not stop at a paper; it becomes an app. You have to know what a person expects of an AI to design something that does not betray that expectation, and you have to run a product with real users to find out whether the research holds outside the lab. They are two halves of one job.