Explainable artificially intelligent (XAI) systems form part of sociotechnical systems, e.g., human+AI teams tasked with making decisions. Yet, current XAI systems are rarely evaluated by measuring the performance of human+AI teams on actual decision-making tasks. We conducted two online experiments and one in-person think-aloud study to evaluate two currently common techniques for evaluating XAI systems: (1) using proxy, artificial tasks such as how well humans predict the AI's decision from the given explanations, and (2) using subjective measures of trust and preference as predictors of actual performance. The results of our experiments demonstrate that evaluations with proxy tasks did not predict the results of the evaluations with the actual decision-making tasks. Further, the subjective measures on evaluations with actual decision-making tasks did not predict the objective performance on those same tasks. Our results suggest that by employing misleading evaluation methods, our field may be inadvertently slowing its progress toward developing human+AI teams that can reliably perform better than humans or AIs alone.
Understanding how machines learn is critical for children to develop useful mental models for exploring artificial intelligence (AI) and smart devices that they now frequently interact with. Although children are very familiar with having conversations with conversational agents like Siri and Alexa,children often have limited knowledge about AI and machine learning. We leverage their existing familiarity and present Zhorai, a conversational platform and curriculum designed to help young children understand how machines learn. Children ages eight to eleven train an agent through conversation and understand how the knowledge is represented using visualizations. This paper describes how we designed the curriculum and evaluated its effectiveness with 14 children in small groups. We found that the conversational aspect of theplatform increased engagement during learning and the novel visualizations helped make machine knowledge understand-able. As a result, we make recommendations for future iterations of Zhorai and approaches for teaching AI to children.