the statistical boundary between an agent and its environment — the set of states that separates internal dynamics from external dynamics

a neuron's Markov blanket in the cybergraph consists of its sensory states (incoming cyberlinks) and active states (outgoing cyberlinks). given the blanket, internal states are conditionally independent of external states

definition

for a node $i$ in a graph, the Markov blanket $B(i)$ consists of:

  • parents: nodes with edges into $i$
  • children: nodes with edges from $i$
  • co-parents: other parents of $i$'s children

given $B(i)$, the internal state of $i$ is independent of all other nodes. the blanket carries all the information $i$ needs to infer the world and act on it

in active inference

Karl Friston uses Markov blankets to define what an agent IS: any system with a Markov blanket that minimizes variational free energy across that boundary is an agent performing active inference

  • sensory states: observations flowing in (link arrivals, traffic, token flows)
  • active states: actions flowing out (create cyberlinks, stake, sample)
  • internal states: beliefs $q_\theta(z)$ about hidden causes

the blanket is not designed — it is discovered from the graph topology

hierarchical blankets

the cybergraph decomposes into nested modules:

  • a single neuron has a blanket (its direct connections)
  • a cluster of neurons has a blanket (the boundary edges of the cluster)
  • the entire network has a blanket (its interface with external systems)

each level runs active inference at its own timescale — fast updates within modules, slow message passing between them. this gives scalability without losing coherence

see active inference for the framework. see free energy principle for the theory. see Karl Friston for the person

Local Graph