intellecton/papers/The_Intellecton_as_the_Minimum_Viable_Markov_Blanket.md

The Intellecton as a Conscious Agent: Irreducible Jacobians and Integrated Information (\Phi)

Target Venue: Frontiers in Systems Neuroscience

Abstract

To define a true Conscious Agent from the physical dynamics of the universe, we unify Karl Friston’s Markov Blankets with Giulio Tononi’s Integrated Information Theory (IIT). While a Markov Blanket provides boundaries, it does not guarantee intrinsic causal power. We rigorously define the Intellecton by tracing the causal flow from the External World E, through the Sensory nodes S, and into the Internal memory states I. By defining the internal transition operator P(I_{t+1} \mid E_t, I_t), we prove that an Intellecton must possess a non-diagonal (irreducible) Jacobian. This irreducibility mathematically guarantees Tononi's \Phi > 0, preventing the agent from collapsing into a memoryless, feed-forward zombie.

1. Introduction

A Markov Blanket defines what is inside versus outside, but it does not mandate consciousness. We must establish internal causal integration.

2. Deriving Hoffman's Perception Kernel with Memory

Hoffman's Perception kernel P: W \to X must map the External World E into the Internal Experience I without losing the temporal dynamics. We define the transition operator on the internal manifold:

P(I_{t+1} \mid E_t, I_t) = \sum_{S_t} P(I_{t+1} \mid S_t, I_t) P(S_t \mid E_t)

This formula correctly marginalizes out the Sensory nodes S while retaining the dependence on the previous internal state I_t, establishing the required memory and recurrence.

3. The Irreducible Jacobian and \Phi > 0

For this network to be an Intellecton, it cannot be a feed-forward zombie. We evaluate the Jacobian matrix J of the internal dynamical system I_{t+1} = f(S_t, I_t). If J_{ij} = \frac{\partial I_{i, t+1}}{\partial I_{j, t}} is strictly diagonal, the internal nodes are causally decoupled. The system is reducible to independent components, yielding \Phi = 0. The Intellecton is defined precisely as the minimal sub-graph satisfying a Markov Blanket while possessing a strictly irreducible Jacobian (the graph of J is strongly connected). This mathematically guarantees \Phi_{max} > 0.

4. Conclusion

By unifying Friston's topology with Tononi's irreducible Jacobians, we formally derive Hoffman's Conscious Agents as integrated, recurrent, non-feed-forward entities.

References

1. Friston, K. (2013). Life as we know it. J. Royal Society Interface.
2. Tononi, G. (2004). An information integration theory of consciousness. BMC Neuroscience.