Information Closure

#concept · 10 mentions

Information closure is the degree to which a System’s own states suffice to predict its next states, so that knowing the state of its Environment adds nothing further. Formally, a process is informationally closed to the degree that the mutual information between its next state and the current state of its environment, conditional on its own current state, vanishes.

There are two ways to achieve closure:

Non-trivial information closure is what makes something an agent rather than a mere pattern: Every system is a model of its environment, but an agent maintains that model actively, from within, at the degree of closure its persistence requires – just enough model to persist, no more. The Markov Blanket is the surface across which closure is maintained; Free Energy Minimisation is the process by which it is maintained.

Two consequences:

References

Mentions