Part 7/9:
Despite its achievements, deep learning is not without challenges. Many models exhibit a lack of interpretability, leading to potential consequences in high-stakes scenarios. A prominent example is the Tesla autopilot incident, where a neural network misclassified a tractor-trailer as a billboard, resulting in a tragic accident. Such instances highlight the need for deeper understanding and accountability in deploying these technologies.
Additionally, deep learning models are vulnerable to adversarial attacks — subtle perturbations in input data that can drastically change output classifications. This weakness raises concerns about the reliability of these systems in critical situations, revealing a contrast with human cognitive robustness.