Free energy principle Puns

I've been trying to get a grip on the Free Energy principle; it seems like a pretty interesting idea. Unfortunately I have some trouble understanding Friston's writing a lot of the time. The 2017 tutorial by Bogacz really helped with the basics. Unfortunately now I am stuck:

The part that I would like to understand most is the predictions the Free Energy Principle makes for action. His 2010 "Action and behavior: a free-energy formulation" is meant to cover these topics and he gets into how retinal stabilisation can be explained. In essence, instead of minimizing "Surprise" by adjusting neural weights, a second way is by using action (e.g. eye movement) to change the input.

I find the paper quite hard to follow, since high level, almost philosophical discussion, precise mathematical statements, and jargon seem to just be mixed throughout. I would like to understand how exactly active inference can be implemented in a free energy context. Could anyone help me, perhaps with some other references or good background material? Has anyone tried to implement this (especially for vision)? Am I alone in finding the relevant papers hard to read?

Those who are accommodated to Karl Friston's free energy principle will know that the original paper is really difficult to understand if you don't have a physics/mathematical background.

I finished a degree in Psychology and continuing my education with a master's in Computational Neuroscience this year. It is only last autumn and this whole year - up until now - when I actually did mathematics for the first time in my life (I didn't have the chance to study it in high-school, or any other science whatsoever.) However, I was able to catch up quite quickly in maths, physics, biology and computer science during undergraduate years (still quite rusty in physics for now though).

My question is: Would it be worth saving money for further education in mathematics and physics later in life, only to be able to understand Karl Friston's paper (and implicitly many other things), given that my purpose is to investigate the relationship between the nature of perception and the nature of reality?

Neurobiologist Karl Friston's free energy principle is the theory that any existing thing (in jargon, any thing with a boundary (a Markov blanket, resisting entropy) separating what's in the thing and what's out of the thing) MUST minimise variational free energy/surprise/entropy/incoherence between prediction and senses, i.e. maximise Bayesian model evidence or approximate Bayesian inference. (The mathematical proof is technical, but the basic idea is that simply in virtue of existing and resisting entropic forces of the environment, the thing MUST in some sense (however primitive) encode information about its environment, minimising free energy/entropy/surprise (these are information-theoretic notions).)

It is a supremely general principle which has been shown to explain cognition, life, homeostasis, action and perception. Moreover, some papers argue that POPULATIONS, via say, natural selection, also approximate Bayesian inference (obeys FEP). In a nutshell, action and perception strive for the same thing: minimise prediction error: Perception bends mind/beliefs to fit senses/world, action bends world/senses to fit mind/beliefs. (Moreover, the optimal scientific methodology is Bayesian, as Nola and Sankey's Theories of Scientific Method argues. So our best route at nontrivial truth is Bayesian inference)

Some accuse it of unfalsifiability but Hohwy defends it as a mathematical/philosophical conceptual analysis, akin to Hamilton's principle in physics.

The extreme scope, explanatory power and ambition (and purported unfalsifiability) reminds me of Hegel's dialectics. But the key similarity is that both principles of Friston and Hegel are about synthesis/unification/"aufheben", or decreasing incoherence/contradiction. Note moreover the beautifully simple premise of the FEP: that the system exists. So one can view the mathematical proof of it as almost like a Kantian-esque transcendental deduction: a necessary condition of the possibility of existence. (Friston's slogan: I am therefore I think)

One immensely interesting feature of Friston's free energy principle is that it might be a naturalistic way of grounding normativity, by approximating Bayesian inference, as this paper by Hohwy shows. (Bayesian inference is optimal

So lemme start off by stating that I'm still an undergrad studying neuro, so by no means do I have in depth knowledge of the principle.

As I understand it, the principle states that brains seek to reduce surprise by reducing the difference between the brain's model of the world and the information coming in through its sensations and perceptions. But does the principle explain how this "model of the world" exists in the brain. How does the brain form it in the first place? Where is the model residing in the brain that this overarching principle could even access it?

I know that these questions are largely unanswered with our current understanding. I guess what I'm trying to glean is just what does the free energy principle do to help us understand consciousness?

Some of my friends have asked me about the free energy principle because they know I'm studying neuro. And I tell them the definition I stated earlier. But this seems unsatisfactory. Like ok, sure that's pretty intuitive and makes sense but it sort of leaves you blue balled for lack of a better word. (Sorry for the langauge y'all)

And why is it that the principle is called unfalsifiable? Because all brains seem to do this but it fails to explain further how they are doing it?

Since Friston showed up as an east egg in 304 and his (real life) brain mapping tools apparently would've been the benchmark for Delos' immortality data project, I wonder if his Free energy principle can somehow explain the relationship between Rehoboam and outliers, if it's in any way connected to mental health, and if anything regarding free will can be inferred from it - something similar to the Bicameral mind theory of S1.

I wanted to look into it myself but, since it's a fairly complex subject, I thought I'd ask here in case someone already thought about it.

PS: there is already a post that presents the general connections, I wonder if anyone can take the idea a bit further.

How would you explain the free energy principle simply? Any analogies/examples to better understand it?

I was wondering if anyone had any insight on Karl Friston's Free Energy Principle. I have been trying to read and understand it online but am having trouble grasping what he means. What is "Bayesian Modeling" and how does it related to neuroscience and consciousness?

I have seen many people talk about how important it is, even that article claming that Friston and his principle might hold the key to AGI. Why is it so important?

Recent article / more information:

1. https://www.wired.com/story/karl-friston-free-energy-principle-artificial-intelligence/

2. https://en.wikipedia.org/wiki/Free_energy_principle#Active_inference

There was a recent computer science presentation whose site that got death-hugged [cached copy].

Does anyone know if Sam or others have looked into this?

I’m wondering if the analogy I’ve been using for the energy-time uncertainty principle is an accurate analogy. If you measure the average energy output of a wave over time you get a better and better measurement the longer you take the measurement since If you measure any system long enough small fluctuate contribute less and less to the average. So the longer you measure the average energy output the lower your uncertainty is, but the flip side is that you don’t know when the wave has any specific energy output.

Is this analogy accurate? Or does it just coincidentally work and it’s not a good representation of reality?

https://www.wired.com/story/karl-friston-free-energy-principle-artificial-intelligence/