The Free Energy Principle: Living is Alignment

The mental tool for this session is the “Free Energy Principle (FEP),” first proposed by British neuroscientist Karl Friston around 2005. We’ve mentioned it briefly before, and you might have heard the term elsewhere, but you may not realize just how powerful it is.

Friston’s 2010 review paper on free energy theory [1] is now one of the most cited works in neuroscience. The Free Energy Principle is hailed as the “Grand Unified Theory of Neuroscience,” considered a masterpiece integrating physics, biology, neuroscience, and cybernetics…

Because it answers both “what life is” and “what life should do.”

✵

Think for a moment about what life is.

A star, so magnificent in its prime, with terrifying mass and staggering energy output, will eventually collapse and scatter; it won’t give birth to a new star. Yet a seed, blown away by the wind, can grow structures layer by layer, maintain them, replicate them, and eventually become a great tree that reproduces. What is the difference?

From a physics perspective, moving toward disorder is the most natural thing—that’s the Law of Entropy! No matter how strong you are, you must eventually merge completely with nature until you are indistinguishable.

Yet a living organism is something that remains “unblended” for a long time. It maintains its shape, its boundaries, and its self over time. This doesn’t violate the Law of Entropy, of course, because if you account for both the organism and its surroundings, entropy still increases—but the individual organism, as an open system, manages to keep its structure from falling apart and even reproduces.

How does life do it? If we knew this secret, could we proactively do it better and gain more vigorous vitality?

Friston’s Free Energy Principle says that life’s strategy is to treat survival as an optimization problem: to keep living without being scattered, you must minimize “Free Energy.”

Let me explain. This might be the most accessible explanation of the Free Energy Principle ever given.

✵

Imagine you are a fish in water. How can you live happily? Many wellness philosophies talk about “merging with nature”—but you cannot completely merge with the environment, or you would become a pile of water instead of a fish.

Being water in water is easy; you have no distinction between inside and outside. As the environment changes, so do you. You don’t need perception, memory, or intelligence; you go with the flow and don’t even have a shape to speak of. But being a fish is not easy.

Because you insist on having a shape and retaining independence, you are fragile.

Having a shape means having a boundary, and having a boundary means there is “me” and “not-me.” Having a “me” requires ensuring that this “me” isn’t flattened by the environment.

The Law of Entropy strongly wants to flatten you. To avoid being flattened, you must ensure you can coexist harmoniously with the environment. To maintain harmony, you must do two things—

One is to transform yourself, making your structure more adaptable to the environment; the other is to transform the environment (or change environments) to make the order of the environment more friendly to you.

Simply put, living is about a two-way “alignment” between yourself and the environment.

This is the fundamental insight of the Free Energy Principle. Let me compare it to illustrate. Previously, people thought life lived by “stimulus-response”: see food and open your mouth, see danger and run. The Free Energy Principle tells you: no. That’s too slow and you’ll likely die because the environment is complex and ever-changing! True life must have intelligence; it must be a “prediction machine.” You cannot wait for the environment to act upon you; you must stay ahead of time. Alignment means you constantly predict what will happen in the environment, ensuring the gap between your expectations and your sensory inputs isn’t too wide, and that environmental changes align with your survival needs.

“Free Energy” was originally a physics term that Friston borrowed specifically to describe the degree of failure in this alignment. There is complex mathematics involved, strictly called “variational free energy”… but you don’t need to worry about that. You just need to know [2]:

Free Energy ≈ Surprise.

This surprise isn’t a “pleasant surprise,” but rather statistical unexpectedness—the degree of disharmony between you and the environment.

For example, if you are a tropical fish whose genes are set for 25°C water, and your senses tell you the water is 25°C, you aren’t surprised at all. Your free energy is low, and you are comfortable. But if your senses tell you the water is 5°C, you are very surprised. If you don’t resolve this high free energy, your system will collapse.

The core idea of the Free Energy Principle is that any system that exists over the long term is constantly working to minimize free energy—that is, to minimize surprise.

✵

To understand specifically how life does this, we first need to understand a concept called the “Markov Blanket.” It’s essentially the “skin” that separates the living system from the external environment—the interface between “me” and the world.

A Markov Blanket consists of two types of states: “sensory states,” which are the only way the outside affects the inside; and “active states,” which are the only way the inside affects the outside.

You can only infer what is happening in the outside world through the sensory states on your Markov Blanket. You don’t see the world itself, but the projection the world casts on your blanket. Similarly, you intervene in the outside world through your active states.

The Markov Blanket is your information input-output interface connecting inside and out. To stay alive, life must manage this blanket well: make the inputs reliable enough, the outputs effective enough, and form a sustainable coupling between the inside and the outside.

If your predictions match your sensory states, you are fine. But our predictions often don’t match our sensory states; there is always surprise. You have two basic strategies for this, corresponding to the “transforming self” and “transforming environment” mentioned earlier—

The first trick is to change your thoughts to adapt to the world, called “Perceptual Inference.” For example, if you see a dark shadow ahead while walking at night and think it’s a ghost, but realize it’s a tree stump as you get closer, you update your belief: “Ah, it’s a tree stump.” This lowers free energy by changing your internal model.

The second trick is to change the world to fit your thoughts, called “Active Inference.” For example, if your biological makeup requires a warm environment but the room is cold, you can’t just change your biology. You can put on a coat or turn on the heater. This changes the surrounding micro-environment through action, also lowering free energy.

However, active inference isn’t just about transforming the environment; it also includes actively probing the environment to update your beliefs. The precision of perceptual inference is also critical; you must be able to judge which information is a meaningful signal and which is just noise, what to believe and what to ignore.

Consider the depth of this idea. The “core self” we discussed earlier can be seen as an aggregate surrounded by an information boundary. Everything is information.

From this perspective, not just animals and plants, but also a cell, a company, an organization, or a culture—as long as they exist stably as a system for a period—must follow the strategy of minimizing free energy. To maintain a boundary in an uncertain environment, you must form a “prediction-error-update/action” closed loop.

Take a company, for example. Its environment is the market; its internal structure consists of strategy, culture, processes, and team capabilities. The sensory states on its Markov Blanket are customer feedback, financial data, and market signals; the active states are product iterations, sales strategies, and organizational adjustments. A company is a surprise-eliminating machine. Managing a company means quickly gathering surprises (customer complaints, market anomalies) and then eliminating errors through high-frequency perceptual inference (reviews, strategy adjustments) and active inference (releasing new products, changing market rules).

To prolong your independence and prevent collapse, your system must live in harmony with the environment—that is, minimize free energy.

✵

This sounds simple, and you might immediately say, “Then I’ll just stay in the most familiar environment! I’ll have zero surprise!” Actually, that doesn’t work because you need to forage, reproduce, and grow. Moreover, even if you don’t change, the environment might. So you must proactively go out and explore.

You don’t predict for the sake of “prediction”; you predict for survival.

So you shouldn’t seek to turn surprise directly to zero. To avoid being caught off guard by future surprises, you’d better have some limited, small surprises now. The entire art lies in the “degree” of this surprise: too much and you collapse, too little and you become dull.

In my view, the secret to operating the Free Energy Principle lies in how you handle the current level of surprise.

Take kindergarten children, for instance. They are just starting to encounter the world and find everything new. If you overwhelm them with too many new things every day, they will be at a loss and learn nothing. It’s best to keep the environment stable and introduce just a little something new each day so they can learn. Some have analyzed Montessori education [3] using the Free Energy Principle and found that they carefully design a low-entropy classroom environment where even every teaching tool has a fixed place. This reduces unproductive surprise during a child’s exploration, allowing them to focus on learning.

Just the right amount of surprise is the “learning zone” (as opposed to the comfort zone and panic zone) mentioned in “Deliberate Practice” theory. It’s the “sense of competence” in “Flow” and “Self-Determination Theory.” It’s the “Likable = Familiar + Unexpected” we’ve repeatedly discussed in the Elite Daily column, and the “appropriate amount of uncertainty” we just mentioned. This is a fundamental property of life!

We need to reduce surprise, but we also need a little bit of it. By having small, controllable surprises normally, you can align ahead of time and avoid facing catastrophic, large surprises.

This is also the root of “curiosity.” Curiosity isn’t an optional emotional seasoning but a strategic setting for intelligence [4].

Current mainstream AI is mostly based on reinforcement learning and lacks sufficient acceptance of curiosity. Karl Friston himself has joined an AI startup called VERSES AI [5], intending to develop new types of neural networks based on the ideas of the Free Energy Principle.

True stability isn’t about eliminating change but about incorporating change into the model.

✵

Let’s look at some daily applications of the Free Energy Principle.

The most basic use is: when you encounter any unacceptable surprise, think about what should align with what. Is your internal model wrong, or is the external environment wrong? If you find all your colleagues are incompetent, the environment might be wrong and you need to change it. But if you find society is unfair and the whole world is wrong, you might need to update your cognitive model.

A better use is to actively seek “small surprises.” For example, when reviewing for an exam, most people like to look at notes or even read the textbook repeatedly because it minimizes surprise and feels most comfortable. But if you aren’t surprised now, you will face a huge surprise during the exam. The correct approach is active inference: find some practice questions to create a moderate amount of surprise and expose the holes in your model, thereby suppressing uncertainty.

The most common mistake people make is indulging in current non-surprise out of fear of large surprises.

Take “procrastination,” for example. You clearly have a paper to write, but you don’t write it. This isn’t actually laziness; it’s you avoiding an expected high surprise. You correctly predict that writing the paper will involve many difficulties and potential failure, so you automatically pivot to a path that minimizes expected free energy—lying in bed and scrolling through your phone.

But you will eventually face the deadline. The Free Energy Principle’s advice is to change macro-predictions to micro-predictions: “I just want to open the computer, create a blank document, and write a title.” You can surely do that. By doing it, you make the prediction successful, gain a tiny bit of dopamine, and build momentum.

Much suffering isn’t because your emotions are too strong, but because your model is too rigid. There is a framework from the perspective of computational psychiatry that suggests “depression” occurs because patients hold extremely negative prior beliefs about the future [6], such as “Everything I do will fail.” Consequently, even when the outside world provides positive feedback—like family encouragement or small successes—you see the error in your prediction but ignore it as low-precision noise. Your model cannot update, trapping you in a self-validating negative cycle.

Perhaps depression is because the Markov Blanket filters out good information. Can you be more sensitive to good information? If you predict that they won’t reply to your email, but it turns out they do, can you modify your model and cheer up a bit?

According to the Free Energy Principle, your personality, preferences, and values—your “core self”—are essentially the highly precise predictive models you solidified in the past to adapt to your environment. Your habits are the automatic execution of these models, saving the energy cost of calculating free energy. Why is it so hard to change habits? Because change means dismantling the model, temporarily exposing the system to anxiety and uncertainty—that is, high free energy.

You find it hard to change habits because you are unwilling to pay the cost of a short-term rise in free energy.

Creating an action so small that it’s almost impossible to fail, making a positive prediction come true, is a universally good way to actively improve your model.

✵

In summary, living is alignment: either making your understanding align with the world, or making the world align with you. We cannot just passively rely on perceptual inference; we must go out and explore and act. This is the fundamental obligation of an independent individual to maintain themselves against collapse.

Because you don’t want to be a pile of water in water; you insist on being a fish.

You must be distinct from the environment, yet live in harmony with it.

I asked an AI if it had any deeper insights into the Free Energy Principle, and Gemini said something particularly romantic: to love someone is to include them within your own Markov Blanket.

By spending time together, you learn each other’s generative models, ultimately minimizing the mutual prediction error. You understand them without words; they understand you without explanation. You form a shared low-entropy state.

In this chaotic universe, finding someone with whom you can jointly minimize free energy and building a super-organism to withstand the storms together—isn’t that the most scientific form of love?

Notes

[1] Friston, Karl. “The Free-Energy Principle: A Unified Brain Theory?” Nature Reviews Neuroscience 11, no. 2 (2010): 127–138.

[2] The strict statement of this is: variational free energy is a computable upper bound on “surprisal” (unexpectedness). But don’t let that bother you.

[3] Laura Desirèe Di Paolo, Ben White, Avel Guénin-Carlut, Axel Constant, Andy Clark; Active inference goes to school: the importance of active learning in the age of large language models. Philos Trans R Soc Lond B Biol Sci 7 October 2024; 379 (1911): 20230148.

[4] Elite Daily Season 6, “A Brief History of Intelligence” 3: The Learning Revolution.

[5] Denise Holt, Friston’s AI Law is Proven: FEP Explains How Neurons Learn, Medium, Aug 15, 2023. https://medium.com/@deniseholt1/fristons-ai-law-is-proven-fep-explains-how-neurons-learn-3718d3be69ac

[6] Chekroud, Adam M. “Unifying Treatments for Depression: An Application of the Free Energy Principle.” Frontiers in Psychology 6 (2015): 153.