What Can Deep Neural Networks Teach Us About Embodied Bounded Rationality

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Decision Theory

Background:

• Simon's bounded rationality defines decision-making as limited algorithmic reasoning.
• Traditional computation (Turing-Church) is algorithmic but lacks interactivity.
• Embodied cognition posits that minds are interactive machines.

Purpose of the Study:

• To introduce and explore the concept of embodied bounded rationality.
• To differentiate embodied bounded rationality from traditional computation and bounded rationality.
• To provide empirical evidence for embodied bounded rationality using deep neural networks.

Main Methods:

• Conceptual analysis of rationality, bounded rationality, and embodied cognition.
• Comparison of Turing-Church computations with interactive machines.
• Examination of deep neural networks as examples of embodied bounded rationality.

Main Results:

• Embodied bounded rationality is both more limited and more powerful than traditional computation.
• Interactive machines, unlike Turing-Church computations, can achieve unique capabilities.
• Deep neural networks demonstrate empirical support for embodied bounded rationality.

Conclusions:

• Embodied bounded rationality offers a more powerful model for understanding cognitive decision-making.
• Interaction is a key component that enhances decision-making beyond algorithmic limits.
• Deep neural networks validate the principles of embodied bounded rationality in AI.