CONCEPT
The Markov Blanket
Karl Friston’s mathematical definition of individuation—the statistical boundary that separates a system’s internal states from its environment by making them conditionally independent of the environment, given the blanket states—and the concept that asks whether AI systems, as currently built, have selves at all.
The Markov blanket is the statistical edge of a self. Named for the Russian mathematician Andrey Markov, whose work on conditional independence underlies it, and given its current significance by
Karl Friston’s
free energy principle, the Markov blanket of a system is the set of states that, once known, render the system’s internal states statistically independent of the external environment. A cell membrane is a physical instantiation of a Markov blanket; so is the skin of an organism; so, Friston argues, is any statistical structure that separates an entity from its environment in the relevant sense. The blanket does not need to be a physical surface—it is a formal, information-theoretic object—and this is precisely what makes the concept philosophically powerful and practically consequential. It allows the question “where does the self end?” to be asked with mathematical precision. Applied to AI systems, the question becomes: do large language models have Markov