Core Insights - The research published in "npj Artificial Intelligence" highlights the fundamental differences between human and algorithmic visual attention in the context of autonomous driving [1] - It introduces a three-stage quantitative framework for human driving attention and identifies a critical flaw in algorithmic visual understanding: the lack of "semantic salience extraction capability" [1] - The study suggests that incorporating human semantic attention can effectively bridge the "semantic gap" and "grounding gap" in professional algorithms without relying on large-scale pre-training [1] Research Methodology - The research team conducted experiments with both expert and novice drivers to complete tasks related to hazard detection, usability recognition, and anomaly detection [1] - Eye-tracking data was used to categorize attention stages, which were then integrated into various professional algorithms and visual language models (VLM) [1] Key Findings - The core difference in attention between humans and algorithms is not in "spatial localization" but in "semantic understanding" [1] - Humans can assign semantic priority to scene features through top-down cognition, a capability that algorithms struggle to learn independently [1] - This discovery offers a non-scaling new pathway for enhancing the performance of autonomous driving algorithms, which is significant for resource-constrained real-time systems [1]