Summary
In this chapter, we approached XAI with a top-to-bottom method. We learned that the counterfactual explanations method analyzes the output of a model unconditionally. The explanation goes beyond explaining why a prediction is true or not. The model does not come into account either. The counterfactual explanations method is based on four key pillars: belief, trust, justification, and sensitivity.
A user must first believe a prediction. Belief will build trust in an AI system. However, even if a user believes a prediction, it must be true. A model can produce a high accuracy rate on a well-designed dataset, which shows that it should be true.
The truth alone will not suffice. A court of law might request a well-explained justification of a prediction. A defendant might not agree with the reasons provided by a bank to refuse a loan based on a decision made by an AI system.
Counterfactual explanations will provide a unique dimension: sensitivity. The method will...
