Google has been using TCAV (testing with concept activation vectors) technology to help explain how an AI comes to a decision using higher-level human-friendly terms.
The use of TCAV aims to avoid perpetuating biases in society and data, as well as to inform experts when evaluating models.
Read here.
