Explainability via Interactivity? Supporting Nonexperts' Sensemaking of Pretrained CNN by Interacting with Their Daily Surroundings

Interactive learning helps non-experts understand AI limitations: The study found that design students using the mobile app to photograph everyday objects gained "vivid understandings about the capabilities and limitations of pretrained CNNs in real-world environments," suggesting that hands-on experimentation with AI systems is more effective than traditional explanations for non-technical audiences.

Playful exploration supports deeper AI literacy: Rather than passive instruction, the "playful learning tool" enabled students to actively test the AI model themselves, resulting in "different depths of thought" in their sensemaking processes—indicating that interactive engagement can lead to more nuanced understanding of how AI actually works in practice.

Visual explanations make AI decisions more intuitive: The app used Class Activation Maps to show users which parts of images the AI focused on when making decisions, allowing students to see the model's reasoning directly rather than relying on technical jargon, making the "black box" nature of neural networks more transparent.

Current XAI research overlooks non-expert needs: The paper highlights that "current research on Explainable AI heavily targets expert users (data scientists or AI developers)," pointing to a significant gap in making AI understandable to the broader public who increasingly interact with these systems in their daily lives.