Mitigating belief projection in explainable artificial intelligence via Bayesian teaching.

State-of-the-art deep-learning systems use decision rules that are challenging for humans to model. Explainable AI (XAI) attempts to improve human understanding but rarely accounts for how people typically reason about unfamiliar agents. We propose explicitly modelling the human explainee via Bayesian teaching, which evaluates explanations by how much they shift explainees' inferences toward a desired goal. We assess Bayesian teaching in a binary image classification task across a variety of contexts. Absent intervention, participants predict that the AI's classifications will match their own, but explanations generated by Bayesian teaching improve their ability to predict the AI's judgements by moving them away from this prior belief. Bayesian teaching further allows each case to be broken down into sub-examples (here saliency maps). These sub-examples complement whole examples by improving error detection for familiar categories, whereas whole examples help predict correct AI judgements of unfamiliar cases.

Age affects reinforcement learning through dopamine-based learning imbalance and high decision noise-not through Parkinsonian mechanisms.

Probabilistic reinforcement learning declines in healthy cognitive aging. While some findings suggest impairments are especially conspicuous in learning from rewards, resembling deficits in Parkinson's disease, others also show impairments in learning from punishments. To reconcile these findings, we tested 252 adults from 3 age groups on a probabilistic reinforcement learning task, analyzed trial-by-trial performance with a Q-reinforcement learning model, and correlated both fitted model parameters and behavior to polymorphisms in dopamine-related genes. Analyses revealed that learning from both positive and negative feedback declines with age but through different mechanisms: when learning from negative feedback, older adults were slower due to noisy decision-making; when learning from positive feedback, they tended to settle for a nonoptimal solution due to an imbalance in learning from positive and negative prediction errors. The imbalance was associated with polymorphisms in the DARPP-32 gene and appeared to arise from mechanisms different from those previously attributed to Parkinson's disease. Moreover, this imbalance predicted previous findings on aging using the Probabilistic Selection Task, which were misattributed to Parkinsonian mechanisms.