Saliency maps can explain how deep neural networks classify images. But are they actually useful for humans? The present systematic review of 68 user studies found that while saliency maps can enhance human performance, null effects or even costs are quite common. To investigate what modulates these effects, the empirical outcomes were organised along several factors related to the human tasks, AI performance, XAI methods, images to be classified, human participants and comparison conditions. In image-focused tasks, benefits were less common than in AI-focused tasks, but the effects depended on the specific cognitive requirements. AI accuracy strongly modulated the outcomes, while XAI-related factors had surprisingly little impact. The evidence was limited for image- and human-related factors and the effects were highly dependent on the comparisons. These findings may support the design of future user studies by focusing on the conditions under which saliency maps can potentially be useful.

The evaluation of explainable artificial intelligence is challenging, because automated and human-centred metrics of explanation quality may diverge. To clarify their relationship, we investigated whether human and artificial image classification will benefit from the same visual explanations. In three experiments, we analysed human reaction times, errors, and subjective ratings while participants classified image segments. These segments either reflected human attention (eye movements, manual selections) or the outputs of two attribution methods explaining a ResNet (Grad-CAM, XRAI). We also had this model classify the same segments. Humans and the model largely agreed on the interpretability of attribution methods: Grad-CAM was easily interpretable for indoor scenes and landscapes, but not for objects, while the reverse pattern was observed for XRAI. Conversely, human and model performance diverged for human-generated segments. Our results caution against general statements about interpretability, as it varies with the explanation method, the explained images, and the agent interpreting them.

Conversational case-based reasoning (CCBR) systems retrieve past cases that are similar to a current problem by eliciting situation descriptions in interactive dialogues with their users. To find out how such human-machine cooperation is put into practice, the present article reviews the CCBR literature and extracts a list of dialogue principles – interaction techniques by means of which CCBR systems communicate with their users. Seven dialogue principles are identified and explained: mixed initiative, question selection and ordering, dealing with abstraction and expertise, explanations, visualisation and highlighting, dialogue termination, and evaluation support. The results reveal that current CCBR systems already make great efforts to put user needs into the centre of the interaction. At the same time, the current implementation of dialogue principles that adjust CCBR systems to user needs raise questions about who should be in control of these adjustments, what levels of human-computer interaction should be adjusted, and what goals should guide adjustment decisions. Moreover, the present review highlights a number of limitations concerning the methodology and contents of CCBR research, and points out questions for future research on human-computer interaction in CCBR systems.

In complex production systems, the diagnosis and correction of faults requires operators to possess a deep understanding of the specific processes and machines as well as general knowledge about the interactions between different system components. However, in actual work environments operator qualification and experience is quite diverse, which leads to an immense variability in the time required for fault diagnosis and in the quality of corrective actions. Both time and quality of fault diagnosis and correction are vital parameters in the functioning of a plant, because downtimes have severe economic consequences and thus should be kept to a minimum. With the introduction of highly complex and flexible cyber-physical production systems, these problems are aggravated as fault sources vary and diagnosis becomes even more challenging. The paper presents a concept for a self-learning assistance system that supports operators in finding and evaluating solution strategies for complex faults. This concept applies a question-answer approach which allows for an incremental, dialogue-based establishment of common ground between operators and the assistance system.