A decade of comparing brains to DNNs: Progress and perspectives

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Comparisons between deep neural networks (DNNs) and the human brain are increasingly central to cognitive neuroscience.
• Early research focused on vision, with convolutional neural networks (CNNs) driving progress in object recognition.
• More recently, large-scale language models have spurred comparisons in language processing.

Purpose of the Study:

• To review the progress and ongoing debates in comparing DNNs to the human brain.
• To highlight the impact of these comparisons on understanding information processing, connectivity, and computational goals.
• To identify future directions for advancing the mechanistic understanding of cognition.

Main Methods:

• Review of existing literature and research comparing DNNs to human brain function.
• Analysis of insights gained from vision and language domains.
• Discussion of challenges in interpretability and validity of DNN-brain mappings.

Main Results:

• DNN-brain comparisons have validated existing hypotheses and generated new ones regarding cognitive processes.
• These comparisons have challenged traditional views of information processing and computational goals in the brain.
• Persistent debates exist regarding the interpretability and validity of these neuro-computational mappings.

Conclusions:

• Despite ongoing debates, DNNs provide valuable frameworks for understanding cognition.
• Integrating cross-modal insights (vision and language) is crucial for future progress.
• Refined models and experimental frameworks are needed to advance mechanistic understanding of the brain.