Untapped Potential in Self-Optimization of Hopfield Networks: The Creativity of Unsupervised Learning

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Computational Neuroscience

Background:

• The self-optimization (SO) model is the third operational mode of the classical Hopfield network.
• It leverages associative memory for enhanced optimization.
• The SO model exhibits characteristics of minimal agency, relevant to Artificial Life studies.

Purpose of the Study:

• To explore the creative capacity of the self-optimization (SO) model.
• To determine if the SO model meets the criteria for a creative process.
• To investigate the role of learning and its parameters in SO model creativity.

Main Methods:

• Analysis of the SO model through the lens of creativity studies.
• Demonstration of the necessity of learning for creative outcomes.
• Examination of four distinct regimes arising from modifications in learning parameters.

Main Results:

• The SO model satisfies the conditions for a creative process.
• Learning is essential for SO models to produce creative outcomes beyond chance.
• Modifying learning parameters results in four regimes, explaining creative products and inconclusive outcomes.

Conclusions:

• The SO model possesses inherent creative capabilities.
• Learning is a key factor in enabling creativity within the SO model.
• The SO model provides a framework for understanding artificial creativity and learning.