What is iteration?
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.
Iteration is a cycle of making or changing something, examining what happens and using that evidence to shape the next attempt. In product development, it is a method for learning through successive versions. It does not mean making endless changes without deciding what those changes are intended to improve.
A useful cycle is Understand, Specify, Build, Test and Decide. Understanding concerns the people and problem. Specifying makes the intended behavior and limits explicit. Building produces something to examine. Testing generates observations. Deciding determines what should happen next, including whether to continue at all.
Consider a tool that summarizes meeting notes. The first version may produce accurate paragraphs but leave users unable to find the decisions. The next iteration might change the structure rather than the AI model. A model is the trained component generating responses; changing it is only one possible intervention.
An iteration can also reveal that the original problem was wrongly framed. Perhaps people do not need shorter notes; they need a reliable record of who agreed to do what. Continuing to polish the summary without revisiting that discovery would create activity without useful learning.
Not every observed difference is evidence of improvement. A new version may perform better on one familiar example and worse on another. Keep representative test cases, including situations that previously failed. A regression is the reappearance of an error or a loss of behavior that previously worked. Rechecking earlier cases helps reveal it.
The size of the change matters too. Changing the model, instructions, data and interface simultaneously may produce a better result while leaving the team unsure why. Where practical, make changes that can be interpreted and reversed. A comparison is more useful when the team knows what differed.
Suppose staff trying a dental follow-up dashboard find that the main problem is not seeing overdue cases but discovering who owns each next action. The next iteration might add ownership and a handoff process rather than a more elaborate AI summary. If the real problem is stale records, the next change belongs in the data flow. This is what it means for observation to change the brief. Another version is useful when it addresses what was learned, not merely when it contains more features.
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