How Can Deep Neural Networks Inform Theory in Psychological Science?

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Computational Neuroscience

Background:

• Deep neural networks (DNNs) achieve human-like performance in complex tasks.
• The inscrutability of DNNs challenges their relevance to human cognition theories.
• Existing debates polarize views on DNNs' utility for psychological science.

Purpose of the Study:

• To review research on interpreting DNN internal mechanisms.
• To challenge the assumption that DNNs are irrelevant to cognitive theories.
• To demonstrate how DNNs can inform theories of cognition and development.

Main Methods:

• Literature review of recent studies on DNN interpretability.
• Analysis of functional similarities between DNN mechanisms and psychological constructs.
• Argumentation against extreme viewpoints on DNNs in cognitive science.

Main Results:

• Evidence suggests DNN internal mechanisms can be interpreted functionally.
• This interpretability counters the notion that DNNs are psychologically irrelevant.
• DNNs offer a new avenue for understanding human cognition and its development.

Conclusions:

• The interpretability of DNNs reconciles AI advancements with cognitive science.
• DNNs can inform and potentially reshape theories of human cognition.
• A nuanced perspective on DNNs is crucial for advancing interdisciplinary research.