12.1  AI and statistics working together

12.1.1  AI as an alternative to statistics

12.1.2  Statistics under the bonnet of AI

12.2  Explainable AI

12.2.1  Why we need explainability

12.2.2  Explainable statistics

12.2.3  Explainability in AI

12.3  Evaluating AI

12.3.1  Holdout

12.3.2  Limits to validity

12.3.3  Comparing AI algorithms

12.4  Case study: regret

12.5  User testing of AI systems

12.6  AI as statistics guru

accuracy measure, ANOVA (Analysis of Variance), average, Bayes rule, bayesian network, bias, big data, Binomial distribution, black box models, chatgpt, cloud infrastructure, cognitive model, confidence in estimates, confidence measure, decision tree, deep neural network, discrete classification, diversity, empirical testing, estimation, explainable ai, explainable statistics, factor analysis, generalisation, global explanation, gpu, hold out, hyperparameters, hyperplane, interaction effect, jackknifing, knowledge base, large language model, legally compliant, Likert scale, linear discriminant, linear regression, local explanation, machine learning, main effect, neural network, neural network weights, overfitting, personal data sovereignty, perturbations, probability distribution, recommender systems, regret, reinforcement learning, replications, reproducibility, residual sum of squares, safety, selection bias, sensitivity analysis, SHAP, standard deviation (s.d., σ), stochastic, summative evaluation, symbolic ai, test data, three cs, training data, transparency, user experience design, user studies, Wizard of Oz prototyping, xai