Saliency map visualizations explain AI predictions on images by highlighting specific regions, but these methods often lack intuitiveness and semantic clarity, leading to interpretability gaps.
We argue that AI explanations should be intuitive—coherent with user knowledge yet simple and selective to accelerate understanding. Inspired by artistic drawings, we propose SketchXplain to generate sketch-based visual explanations for intuitive image-based explainable AI (XAI).
SketchXplain integrates techniques from saliency maps, concept-bottleneck models, and sketch optimization, selecting coherent observation artifacts, forming concepts for knowledge coherence, using cues for representation, and achieving simplicity through abstraction.
Evaluations on face expression recognition showed that SketchXplain supported quicker interpretation with more aligned visualizations than saliency maps or simple drawings. Further evaluation on skin lesion diagnosis revealed that SketchXplain visualized disease symptoms more coherently, better supporting lay diagnosis.
Thus, this work illustrates the value of sketches for intuitive, simple, coherent, and quick image-based XAI visualizations.
Blogger's Review: The introduction of SketchXplain is a significant enhancement to traditional saliency map visualization methods. By incorporating artistic sketches, it improves the intuitiveness and comprehensibility of human-computer interaction, providing a more user-friendly explanation tool for non-experts, greatly enhancing the interpretability and usability of image classification.