Details in model explanations?

How much disclosure and accuracy should be given when explaining a model?

The Dutch DPA has indicated that one of its focus areas for the coming years is explanation of models. The explanation of algorithms is a concretization of the core principle of transparency. Some initial interpretations the Dutch DPA gives about the explanation of automated decisions are that a data subject should expect an explanation of:

• The process (where the data comes from and how it is processed)
• The key elements of the algorithm (what is the type of algorithm and how does training it work)
• How the process and algorithm result in the outcome. This should include the possible outcomes and the effects of the outcomes on the data subject

The data controller will have to train a person, who will have to be able to provide a reasoned explanation of the process of automated decisions. They also have to give their own assessment on the decision, because the data subject has the right to human intervention. This allows data subjects to enforce their right to transparency of processing.