Understanding and influencing decisions

Counterfactual explanation

A box saying "chance of rain 87%", below this box is more information saying "if below 64%, your quote would have been cheaper"

Description

Counterfactual explanations provide a simple way for people to grasp how an AI system generates a result. These explanations rely on examples and occur after the fact. They present statements related to the variables used in reaching the outcome.

For example, an insurance quote could be cheaper if there was a lower chance of rain.

IF thinks that using a counterfactual explanation alongside UI elements that let someone play with the variables, helps build a useful mental model of how an AI system works.

Advantages

  • Helps teams and organisations give people their digital right to “explanation” in a way that does not overwhelm them with technical details.
  • More intuitive for people to comprehend, and provide actionable feedback.

Limitations

  • Could be gamed by people or service providers and lead to an increase in the number of false inputs.

Examples

  • Skyscanner →

    It list the flights that are possible to take across the day. The prices for those flights are also listed, so a user can see that travelling at different times of day impacts the price they pay

Related patterns