The widespread adoption of deep learning (DL) models in neuroscience research has introduced a fundamental epistemological paradox: while these models demonstrate remarkable performance in pattern recognition and prediction tasks, their inherent opacity contradicts neuroscience’s foundational goal of understanding biological mechanisms. This review article examines the growing trend of using DL models to interpret neural dynamics and extract insights about brain function, arguing that the black box nature of these models fundamentally undermines their utility for mechanistic understanding. We explore the distinction between computational performance and scientific explanation, analyze the limitations of current interpretability techniques, and discuss the implications for neuroscience research methodology. We propose that the field must critically evaluate whether DL models can genuinely contribute to our understanding of neural processes or whether they merely provide sophisticated curve-fitting tools that obscure rather than illuminate the underlying biology.

Upper-limb impairments affect a substantial portion of the global population, often limiting the ability to perform daily activities. Robotic rehabilitation systems offer a promising solution by enabling high-dose, task-oriented therapy with consistent and objective feedback. However, user acceptance and perceived comfort are critical for their successful adoption. This work presents a feasibility, performance, and comfort evaluation of a 2-degree-of-freedom upper-limb rehabilitation exoskeleton capable of performing elbow flexion/extension and forearm pronation/supination. A total of 47 healthy participants were enrolled and tested across three rehabilitation modalities: passive assist, active assist, and active resist. Passive assist enabled full range-of-motion execution, active assist supported movement, and active resist provided variable resistance via a sliding bar (0–100%). Objective performance metrics, including position, current, and temperature, were recorded and analyzed, revealing trajectory-tracking errors during passive assistance of 4.82° ± 0.02° for forearm movement and 1.20° ± 0.04° for elbow movement, with actuator temperatures remaining below their rated limits throughout the study. The active assist mode did not achieve a true assist-as-needed performance, indicating a need for further refinement. Subjective evaluation included the System Usability Scale, yielding a score of 87.1 ± 9.6, indicating excellent usability, and a safety and comfort assessment averaging 4.4 ± 0.4 out of 5. Perceived effort was assessed using the Borg CR-10 scale and generally scaled appropriately across modalities, although some variability suggests the need for further investigation. Qualitative feedback identified areas for improvement, particularly in ergonomics and control behavior. Overall, the results support the feasibility, usability, and safe operation of the proposed exoskeleton and provide insights for future device refinement and evaluation with target user populations.