Patterns / Self-Improvement & Feedback
AI products and systems should have a clear, low-effort way to tell the system it was wrong . This feedback should be used to improve the system.
When the human provides feedback to the system it should show something the feedback was taken onboard by showing visible improvements. This has two positive side effects: one it encourages people to provide feedback and secondly it improves the system or product.
In some contexts, the human is the expert in which case systems and products should provide clear and accessible interfaces for the human to label, correct or annotate to improve the AI’s future outputs.
Feedback should be embedded directly into the AI product or system and thought of as a core feature rather than surfaced as a one-off survey.
When an AI product or system retains information about a person in order to adapt to them, that information should be viewable, editable and deletable item by item rather than as one opaque store. People should be able to tell which retained detail produced a given behaviour.