Problem

The AI system is wrong and the user needs to edit, correct, refine, or recover the system’s behavior.

Solution

Enable the user to alter the AI system’s behavior by editing, correcting, or refining its output and making clear that their correction will be used as feedback for its learning over time (see Guideline 15, Encourage granular feedback).

Use when

• The user would benefit from correcting the AI system on a granular level during regular interaction.

How

Enable the user to correct the system through a mechanism that uses feedback patterns (see patterns for Guideline 15).

User benefits

• It offers the user an efficient mechanism for correcting the AI system.
• It can help the user feel in control.
• Corrections indicate the system may not be perfect and give the user insight to how well the system can do what it can do (Guideline 2,  Make clear how well the system can do what it can do).
• When corrections are used to help the system learn (Guideline 13, Learn from user behavior), a better user experience can be provided over time.

Common pitfalls

• It’s unclear to the user that they can correct the AI system.
• It’s unclear to the user how to correct the AI system.
• The user’s correction to the respective behavior is delayed.
• Completing the correction interaction is costly for the user.
• In some contexts, such as driving a car,  it might be inefficient, dangerous, or disruptive for the user to correct the AI system.

Keep in mind that repeated correction of the AI system can be costlier to users than performing the task themselves. If the AI needs frequent correction, consider disabling the feature or making it possible for users to disable the feature (see Guideline 17, Provide global controls). Also, the AI system should learn from this behavior so it can improve over time (see Guideline 13, Learn from user behavior).