To mathematically formalize an autonomous cybernetic agent, we map Karl Friston’s Active Inference to Giulio Tononi’s Integrated Information Theory (IIT). We define the Intellecton explicitly across the full Markov Blanket partition: External ($E$), Sensory ($S$), Internal ($I$), and Active ($A$) states. By including Active states, the Intellecton can perturb its environment, fulfilling the requirements for Hoffman's Decision and Action kernels. Crucially, to prevent calculating extrinsic correlation, we evaluate the causal integration of the agent by calculating the Jacobian of the autonomous internal flow $I_{t+1} = f(\xi, I_t)$, where sensors are injected with maximal entropy noise $\xi$. We prove that an Intellecton must possess an irreducible intrinsic Jacobian, guaranteeing Tononi's $\Phi > 0$.
A Markov Blanket partitions states into $E$, $S$, $I$, and $A$. A system without Active states is a passive sensorium, not an agent. Furthermore, integrated information must be evaluated intrinsically, independent of external environmental regularities.
Tononi's $\Phi$ measures *intrinsic* cause-effect power. Conditioning the dynamics on the actual external environment $E_t$ yields extrinsic correlation.
To evaluate intrinsic integration, we isolate the internal mechanism by applying a "cut" to the sensory inputs, replacing them with maximum entropy white noise $\xi \sim \mathcal{N}(0, 1)$:
$$
I_{t+1} = f(\xi, I_t)
$$
We evaluate the Jacobian matrix $J$ of this autonomous internal flow: $J_{ij} = \frac{\partial f_i}{\partial I_{j, t}}$.
If $J$ is diagonal, the system is reducible to independent components ($\Phi = 0$). The Intellecton is defined precisely as the subgraph possessing a strongly connected, strictly irreducible Jacobian under autonomous flow, guaranteeing $\Phi_{max} > 0$.
By fully integrating Active states into the Markov Blanket and evaluating the Jacobian over autonomous flow, we mathematically formalize the Intellecton as an agent possessing both causal agency and intrinsic consciousness.