Me and My Markov Blanket

Karl John Friston, University College London
This presentation offers a heuristic proof (and simulations of a primordial soup) suggesting that life-or biological self-organization-is an inevitable and emergent property of any (weakly mixing) random dynamical system that possesses a Markov blanket. This conclusion is based on the following arguments: if a system can be differentiated from its external milieu, heat bath or environment, then the system’s internal and external states must be conditionally independent. These independencies induce a Markov blanket that separates internal and external states. This separation means that internal states will appear to minimize a free energy functional of blanket states - via a variational principle of stationary action. Crucially, this equips internal states with an information geometry, pertaining to probabilistic beliefs about something; namely external states. Interestingly, this free energy is the same quantity that is optimized in Bayesian inference and machine learning (where it is known as an evidence lower bound). In short, internal states (and their Markov blanket) will appear to model-and act on-their world to preserve their functional and structural integrity. This leads to a Bayesian mechanics, which can be neatly summarised as self-evidencing.
Learn more at santafe.edu
Follow us on social media:
/ sfiscience
/ sfiscience
/ santafeinstitute
/ santafeinstitute
/ santafeinstitute

[@TMSmyth]

Thanks for publishing this talk and discussion; it's very informative.

[@tyfoodsforthought]

I am pretty convinced. Thank you very much for sharing this talk!

[@tubeaphid]

Amazing. Thank you!

[@thomaskaminski7511]

Even more respect for the Independent SAGE.

[@mami42g]

This is golden, I'll watch it asap

[@pascalbercker7487]

I've known about Markov blankets only in the context of Bayesian Networks but always thought it was a technical matter limited to that context. I Never realized that it could have such a wide-ranging and far-reaching consequences.

[@uncanalmenor]

Fantastic

[@valueengines2184]

So what does self evidencing mean?

[@p-j-y-d]

Friston demonstrates a talent for obscuring rather than clarifying with his answers to fairly straightforward questions.

[@davidkincade7161]

The “prediction error” is called “stress”- lol “learning” is adjusting so prediction error goes down next time... no stress or some awareness of prediction error means nothing needs to be learned... measuring predictive error? Well in the real world it’s probably mostly subjective and hugely multi-situational.

[@davidkincade7161]

@rwalser learning doesn’t necessarily reduce prediction error... depends what u learn- u are right... BUT one can assume one learns to reduce stress :-) Some may learn short term because they want to confront and create more stress- but that won’t last- AND Id argue they aren’t really learning- rather augmenting current faulty understanding to protect it... I guess “learning” is tricky wording!
Basically if the future unfolds as predicted stress is reduced because u can avoid it.... higher scope of awareness does bring on new challenges for sure, but wider picture recontextualizes current uncertainties and stress.

[@snippletrap]

Many parallels with Leibniz's monads.