Summary
In this chapter, we went right to the core of XAI with WIT, a people-centric system. Explainable and interpretable AI brings humans and machines together.
We first analyzed a training dataset from an ethical perspective. A biased dataset will only generate biased predictions, classifications, or any form of output. We thus took the necessary time to examine the features of COMPAS before importing the data. We modified the column feature names that would only distort the decisions our model would make.
We carefully preprocessed our now-ethical data, splitting the dataset into training and testing datasets. At that point, running a DNN made sense. We had done our best to clean the dataset up.
The SHAP explainer determined the marginal contribution of each feature. Before running WIT, we already proved that the COMPAS dataset approach was biased and knew what to look for.
Finally, we created an instance of the WIT to explore the outputs from a fairness perspective...
