Trust, Agency & Human–Robot Interaction
What happens to trust, agency, and a sense of control when people work beside machines that are usually — but not always — right.
A decision support system that is right ninety percent of the time is not ninety percent of a good system. It is a different kind of system entirely, because the failures arrive without warning and land on a person who has already learned to defer.
Much of my work has followed that problem through production settings. When a support system gives wrong advice, people comply anyway more often than they should, and how badly this goes depends on the interface rather than on the error rate alone. Usability turns out to matter most precisely when a system is wrong, which is the opposite of how usability is usually justified. People also attribute the outcome to themselves or to the machine depending on cues that have nothing to do with who was actually responsible.
The same question reappears with collaborative robots, where the stakes include not just performance but the operator’s sense of agency. Trust after a cobot failure is repairable, and what repairs it is communication rather than reliability statistics. Context and design shape whether people extend trust to a robot at all, and people will delegate morally weighted tasks to a machine under conditions that can be characterised rather than guessed at.
Underneath all of it is a question I take seriously and think the field often waves at: whether the human beside the automation still has meaningful autonomy, or only the appearance of it. That is the thread running through the systematic review of agency in human-robot interaction and the interdisciplinary work on Industry 5.0.
