Robust and Reliable

Summary

This chapter begins with some of the primary concepts in the area of robust and reliable artificial intelligence (AI). Reliable AI performs as expected even given inputs that were not included in training data, what are called out‐of‐distribution (OOD) inputs. These are data points that are different from the training set, and reliable AI must be able to detect whether data is OOD. The subset of machine learning known as deep learning has led to powerful innovations and AI applications transforming industries and business models. AI tools might extrapolate correlation between data points, but they struggle to determine causation. As with all AI use cases, the organization deploying the tool should calculate for itself the importance of AI reliability relative to AI value. AI tools are deployed into a real‐world environment full of threats to safety and security.