# A Neuro-Mimetic Realization of the Common Model of Cognition via Hebbian Learning and Free Energy Minimization

**Authors:** Alexander Ororbia, Mary Alexandria Kelly  
**arXiv:** [2310.15177](https://arxiv.org/abs/2310.15177)  
**Date:** 2023-10-14  
**Summary page:** https://www.summarizepaper.com/en/paper/2310.15177v1/

## Key points

- Large neural generative models in generative AI can synthesize intricate sequences of words and produce complex image patterns.
- The popularity of generative AI raises questions for Cognitive Science on how such functionality can be acquired and instantiated in biological or artificial substrates.
- The authors propose crafting cognitive architectures using neuro-mimetic generative building blocks as a long-term pathway.
- The COGnitive Neural GENerative system is introduced as an architecture that embodies this approach, casting the Common Model of Cognition in terms of Hebbian adaptation.
- By optimizing a variational free energy functional, this architecture seeks to mimic cognitive processes observed in humans and other organisms.
- Developing cognitive architectures that incorporate neuro-mimetic principles can provide insights into both artificial intelligence systems and biological cognition.

## Summary

In their paper titled "A Neuro-Mimetic Realization of the Common Model of Cognition via Hebbian Learning and Free Energy Minimization," authors Alexander Ororbia and Mary Alexandria Kelly discuss the emergence of large neural generative models in the field of generative artificial intelligence (generative AI). These models have the ability to synthesize intricate sequences of words and produce complex image patterns, which has opened up new opportunities and challenges for statistical machine learning. The increasing popularity of generative AI also raises interesting questions for Cognitive Science, a field that aims to understand the processes underlying minds and brains. Specifically, researchers are interested in how such functionality can be acquired and instantiated in biological or artificial substrates. To address this goal, the authors propose a promising long-term pathway: crafting cognitive architectures using neuro-mimetic generative building blocks. In their discussion, Ororbia and Kelly introduce the COGnitive Neural GENerative system as an architecture that embodies this approach. This system casts the Common Model of Cognition in terms of Hebbian adaptation, which is a learning rule based on synaptic plasticity. By optimizing a variational free energy functional, this architecture seeks to mimic the cognitive processes observed in humans and other organisms. Overall, this paper highlights the potential synergy between generative AI and Cognitive Science. By developing cognitive architectures that incorporate neuro-mimetic principles, researchers can gain insights into both artificial intelligence systems and biological cognition. The proposed COGnitive Neural GENerative system represents a step towards understanding and replicating complex cognitive processes through computational modeling.