XAI—Explainable artificial intelligence

See allHide authors and affiliations

Science Robotics  18 Dec 2019:
Vol. 4, Issue 37, eaay7120
DOI: 10.1126/scirobotics.aay7120


  • Fig. 1 Performance versus explainability tradeoff for ML techniques.

    (A) Learning techniques and explainability. Concept adapted from (9). (B) Interpretable models: ML techniques that learn more structured, interpretable, or causal models. Early examples included Bayesian rule lists, Bayesian program learning, learning models of causal relationships, and using stochastic grammars to learn more interpretable structure. Deep learning: Several design choices might produce more explainable representations (e.g., training data selection, architectural layers, loss functions, regularization, optimization techniques, and training sequences). Model agnostic: Techniques that experiment with any given ML model, as a black box, to infer an approximate explainable model.

    Credit: A. Kitterman/Science Robotics

Stay Connected to Science Robotics

Navigate This Article