Andrey Vlasov Predictive Minds Under Pressure

Dialogue 8. Free Energy and Predictive Coding

Teacher: Now we can turn to a theory that has become genuinely central in computational neuroscience — and also, it turns out, a surprisingly good description of what it feels like to be wrong in a stressful situation. This is the Free Energy Principle (FEP) and its implementation through predictive coding.

The core claim is simple and radical at once: a brain can be modeled as a system that minimizes a quantity related to surprise about its sensory inputs.

Student: So, the brain is basically a Bayesian nerd?

Teacher: That is one good way to say it — the brain might object to the word «nerd», but the equations would quietly agree. A Bayesian reasoner constantly updates beliefs based on new evidence, always asking: given what I am sensing, what is most probably causing it, and what should I predict next? In predictive coding schemes, higher levels of the cortex send predictions downward; lower levels send back prediction errors — the gap between predicted and actual input. The system continuously adjusts its internal model to reduce these errors.

In this picture, perception is not passive reception. It is active inference: the brain is always guessing what is out there and checking those guesses against incoming data.

Student: And conscious experience appears when… what exactly happens?

Teacher: One intuitive proposal is that a state becomes conscious when prediction errors at relatively high levels of the hierarchy cannot be quickly suppressed — the model has to reorganize itself significantly. During that reorganization, the content becomes globally available and experienced.

Notice the thread back to Anokhin: his «acceptor of the result of action» is exactly a predecessor of the generative model. The mismatch he described is what we now call prediction error. The terminology changed; the insight stayed.

Correction of behavior is model updating. The Free Energy framework generalizes this logic to perception, action, and learning in a unified way.

In the ontology of this book, FEP is one of the main tools for describing how a predictive mind stays coherent in a noisy world — or fails to.

Dialogue 9. Integrated Information and Computational Mass

Student: Where does Integrated Information Theory fit into this? You said we would not treat it as a rival dogma.

Teacher: IIT comes at the problem from the opposite direction — which is exactly what makes it a useful companion to FEP, not a rival. Instead of starting with neurons and computation, it starts with the structure of experience itself and asks: what must any system have if it is going to have experience at all?

Giulio Tononi proposed a set of axioms: experience exists, it is structured, specific, integrated, and exclusive. From these, he constructed a formal quantity, Φ (phi), which measures how much information a system carries as a whole, over and above the sum of its parts. A system with high Φ is said to have a rich, integrated state.

In this book, we do not take the slogan «consciousness is Φ» literally — it is a tempting slogan but an overcommitment. Instead, we treat Φ as a candidate for a computational mass index: a way to ask how much structured, integrated «stuff» a system can carry in principle.

Student: And how does that relate to FEP?

Teacher: FEP tells us how systems maintain themselves over time by updating models and reducing surprise. IIT tells us something about how densely structured their internal states can be. They point to different, but potentially complementary aspects of conscious systems.

You could say, playfully, that FEP describes how a system moves through its state space, while IIT hints at how «heavy» or «thick» any given state is — how much is packed inside the moment. One tracks the journey; the other tracks the density of the traveller.

The deeper theory — if it ever arrives — might unify these. For now, we use both as tools. Neither breaks if you drop the other.

Dialogue 10. A Brief Truce

Student: So, we have gamma synchrony, thalamocortical loops, functional systems, significance detectors, predictive coding, and integrated information. That is a lot of moving parts.

Teacher: It is — and if your head is slightly overloaded right now, that is a sign the material has actually been landing. But notice a pattern before we move on.

— Neural correlates (NCC) tell us where and when conscious contents appear in the brain.

— Functional systems and predictive coding tell us how the brain anticipates and corrects its own states.

— Significance detectors and global workspaces tell us which contents make it into the spotlight.

— IIT and related measures hint at how structured and unified those states can be.

We do not need to collapse this into one slogan — and we should probably be suspicious of anyone who tries. For our purposes, it is enough to treat these as parts of a single conceptual toolbox, each with its own job.

In the next chapter, we will leave the brain for a moment and look at consciousness from another angle: as activity — as something that is produced in doing, in social interaction, and in meaning. This is where the Soviet tradition and modern enactive approaches have a lot to say.

Конец ознакомительного фрагмента.

Текст предоставлен ООО «Литрес».

Прочитайте эту книгу целиком, купив полную легальную версию на Литрес.

Безопасно оплатить книгу можно банковской картой Visa, MasterCard, Maestro, со счета мобильного телефона, с платежного терминала, в салоне МТС или Связной, через PayPal, WebMoney, Яндекс.Деньги, QIWI Кошелек, бонусными картами или другим удобным Вам способом.