What is a chatbot?
The 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 chatbot is a program that interacts with people through a conversational interface. Most use written messages, although some include speech. The term describes how a person communicates with the program, not one particular kind of AI underneath it.
A chatbot may follow scripted rules, search a collection of answers or use a language model to generate replies. Some combine all three. Earlier systems were not all simple menus, and modern chatbots are not all trained with the same feedback methods. The implementation matters more than the label.
A language-model chatbot can respond flexibly to unfamiliar wording, explain material and maintain a conversation using the context supplied by the application. It can also produce plausible but unsupported statements. The fact that an answer is immediate and confidently written does not tell the reader whether it came from an approved source, a live record or the model’s learned patterns.
A chatbot is not necessarily an agent. An agent can choose actions while pursuing a goal. A chatbot may only exchange messages, or it may provide the interface to an agent that searches, books or updates information. “It told me the appointment was changed” and “the appointment was changed in the booking system” are different claims.
Good design makes those distinctions visible. Users should know when they are interacting with software, what information it can access, what it can do and how to correct a mistake. A system should not imply that a person is monitoring the conversation when none is.
In health care, a chatbot answering administrative questions has a different role from one offering individualized clinical advice. Even an administrative assistant needs a response to unexpected symptoms or distress. Its boundaries, escalation route and performance across languages should be tested. Availability outside office hours can be useful, but an always-available interface is not the same as an always-available clinical service. The organization remains responsible for what the interface leads people to believe and do.
Sources: Ji and colleagues — Survey of Hallucination in Natural Language Generation (2023); Anthropic — Building Effective Agents (2024)
To learn more, take a look at 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