What is responsible AI?
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.
Responsible AI is the practice of making defensible choices about how AI is designed, used and monitored. It asks who benefits, who may be harmed, what evidence supports the use and who is answerable for the consequences. It is not a particular model architecture or a feature that can be switched on.
Several related terms help clarify the work. Ethics examines values and obligations. Governance assigns decision-making authority and oversight. Safety addresses harm. Fairness examines how benefits and burdens are distributed. Responsible AI brings these concerns into the actual choices made around a system.
The work begins before model selection. A proposed use may solve the wrong problem, exclude the people most affected or optimize an unsuitable measure. A technically accurate model trained to predict spending, for example, is not necessarily a good way to identify unmet need.
It also spans several levels. The model needs appropriate evaluation. The application needs reliable information handling, permissions and recovery from errors. The organization needs people, processes and capacity to respond. A weakness at one level cannot always be corrected by improving another.
Responsibility includes deciding not to use AI where a simpler method is sufficient or where the evidence does not support the proposed use. It also includes comparing AI with the real alternative. Existing manual processes can be slow, inconsistent or unfair, so “keep the current system” should not be treated as a risk-free option.
For a clinic considering automated follow-up, the question is not only whether the messages are accurate. Can patients understand and respond to them? Does someone act on a reply? Are people without digital access served adequately? Can a mistake be corrected before it delays care?
A responsible implementation makes those questions part of design and evaluation, then continues checking after launch. Documentation helps preserve decisions, but documents alone do not create responsibility. The practical standard is whether the organization can explain its choices, observe their consequences and change course when the evidence requires it.
Sources: WHO — Ethics and Governance of Artificial Intelligence for Health (2021); NIST — Artificial Intelligence Risk Management Framework (AI RMF 1.0, 2023)
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
Lead with AI in Health Care
The AI Pragmatist is a monthly newsletter from ALA faculty, with new explainers & essays from the Advanced Learning Academy