Microsoft AI Dashboard Demo

A demonstration site shows how the Responsible AI Dashboard operates. In the provided example, the data scientist has created a model for home prices based on training data. However, when comparing actual sale prices against those predicted, errors are apparent. In the screenshot at the top, the data scientist can see that many errors stem from older homes (built prior to 1978) with interior square footage more than 1,078 square feet. In this case, the data scientist can create a new data cohort, which can be analyzed separately.

 

The bottom chart shows how the various features of the house affect the prediction. Note that “YearBuilt” (far right) is weighted more heavily for new houses than old, potentially affecting results.

 

Other features allow the data scientist to change the values (called “counterfactuals”) for a given element of the data set and see how that affects the prediction. The tool also supports aggregate causal inferencing, which can show how each parameter, or combinations of parameters (for example, two-car garage and fireplace), directly affect the result.