uncertainty

Terms from Artificial Intelligence: humans at the heart of algorithms

The glossary is being gradually proof checked, but currently has many typos and misspellings.

Uncertainty in AI may arise from the data (e.g. inaccuracy of sensors) or due to the processing of the data (e.g. stochastic nature of some neural-network training methods). Uncertainty may affect both the training sets used to create machine learning or other forms of AI; and the data needed for decision maing when the AI model is deployed. For example, if an image recognistion algorithm is trained using past radiographic images, some may have been misclassified by previous human processes (uncertainty in training data), but also when deployed some scans may be poor quality due to patients moving during scanning (uncertainty during deployment).
There are various methods used in AI for uncertain reasoning. In addition, when designing the human processes and user interactions that work with AI, it is important that uncertainty is adequately communicated and overall processes created that are robust.

Used in Chap. 3: pages 27, 30, 34; Chap. 7: page 100; Chap. 11: pages 147, 158; Chap. 15: page 228; Chap. 18: pages 276, 283, 288, 291; Chap. 19: pages 293, 294, 298, 304, 308, 309