scrutability

Terms from Artificial Intelligence: humans at the heart of algorithms

The glossary is being gradually proof checked, but may have typos and misspellings.

Scruitability is the potential for an algorithm or model to be understood and interpretted by a human. Some representations, such as decision trees, are more naturally scrutible, at least by experts. Other, such as neural networks, are opaque and hence there is a need for techniques in explainable AI to expose aspects of these black-box models. Scrutability is important to ensure legal compliance and expose potential bias in the model.

Used in Chap. 18: page 283

Used in glossary entries: bias, decision tree, explainable AI, neural network