Model transparency

Transparent artificial intelligence is becoming an important requirement for many businesses. In risk management, decisions need to be explainable, and opaque predictive models are not allowed. In marketing, the policy for the transparency of models might be less strict and allow for the use of opaque models.

Each model type that comes with Pega Platform™ is assigned a transparency score ranging from 1 to 5, where 1 means that the model is opaque, and 5 means that the model is transparent. Highly transparent models are easy to explain, whereas opaque models might be more powerful but difficult or not possible to explain. For example, a decision tree has a high transparency score, whereas a neural network model has a low transparency score.

Depending on the company policy, models are marked as compliant or non-compliant for a specific business issue. By default, the transparency threshold is set to 1 and all model types are allowed in all business issues. Lead data scientists can modify transparency thresholds for different business issues. For example, they can increase the threshold for Budget to indicate that opaque models are non-compliant in that area. To set the thresholds in Prediction Studio, go to the Settings section and choose Model transparency policies.