What is a mental model?
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
A mental model is a person’s understanding of how a system works. It includes what the system’s objects and controls mean, what will happen after an action and what the person believes the system remembers or can access. It is an interpretation in the user’s mind, not an AI model inside the software.
Suppose someone sees a button labeled “Share.” They may expect it to prepare a link, send an email or make the document publicly accessible. Those expectations have different consequences. A technically functioning button can still create a serious misunderstanding if the person’s expectation does not match its actual behavior.
People develop mental models from familiar products, labels, previous experience and the feedback they receive. Consistent behavior helps them predict what will happen next. Inconsistent labels or unexplained changes force them to guess. The goal is not to teach every internal technical detail, but to support a sufficiently accurate understanding for the task.
AI products create particular opportunities for confusion. A conversational assistant may sound as though it has read every file in a project when only selected material was supplied. It may appear to remember a preference even though the application saved and reintroduced it. Its language can suggest capabilities that the actual software does not provide.
A useful design test is to ask people what they think the system is doing and what they expect an action to cause. Then compare those explanations with the implementation. If users believe a draft has already been sent, the problem is not solved by pointing out that a technical status field was correct somewhere else.
Mental models are also relevant to correction. People need to understand whether editing a sentence changes only the current document, updates a reusable instruction or affects a stored preference. Different kinds of change should not be hidden behind the same vague language of “learning.”
A clinic coordinator seeing “referral complete” may assume the patient has an appointment, while the software may mean only that required documents were received. Both interpretations are plausible; the mismatch is a design problem. Ask staff what the status means to them, then compare their answers with the underlying records and behavior. More precise labels such as “documents complete; booking pending” can make the next responsibility visible. A useful mental model lets people predict what remains to be done.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy