Conifold Theory Basics, Part 3: How does neural computation generate consciousness?

Neuroscience tells us that perceptual experience and volitional action are tied to neuronal activity in the cerebral cortex. But how does neural computation produce these key features of consciousness?

Consciousness is characterized by a cohesive stream of rich, qualitative perceptual experience; the emergence of cognitive models of the world, centered on the self; and the ability to initiate volitional action. Neuroscience tells us that all three of these key features of consciousness are tied to neuronal activity in the cerebral cortex.

The qualitative nature of perceptual experience is limited by the range and sensitivity of the sensory apparatus, and is dependent on the encoding of sensory stimuli in the cerebral cortex. Experiences prompt synaptic remodeling and the formation of memories, molding perceptual expectations in relevant contexts. These memories and expectations are then used to select behavior within the perceived context.

How neurons encode information

To compute the most likely state of the surrounding environment, a cortical neural network must select an optimal system state in the present context, from a large probability distribution. It’s important to note that each neuron is not in a defined ‘0’ or ‘1’ state – each neuron has some probability of switching from a ‘0’ to a ‘1’ state, based on the contribution of both upstream signals and random electrical noise. So each neuron is performing a computation – by adjusting its voltage - but each ion in the system is contributing disorder or randomness, rendering the state of each neuron uncertain.

In this view, the brain encodes information in a very physical way. The entire neural network is a macrostate, or the sum of all component microstates. Each neuron is a microstate, with some probability of firing or not at a given moment. The distribution of all component microstates is the ‘information’ encoded by the system.

Information is a physical quantity

Energy must be expended to generate information. This physical process means that caloric energy, used by the brain, is actually producing a physical quantity of information.

The information encoded by a neuron is represented by its voltage. The voltage is the difference in electrical potential between the inside of the neuron and the outside. That is an electrochemical gradient - with many positively charged ions outside the cell. In a spinal neuron, it makes no difference if some ions leak across the neuronal membrane. But in a cortical neuron, it makes all the difference - these tiny little 'leak currents' can actually affect the probability of a neuron firing.

So neuroscientists have the hard job of describing how this random electrical noise contributes to cortical neuron signaling outcomes. It doesn't make any sense - if spinal reflex circuits are clean digital circuits, robust to random electrical noise, why would cortical neurons evolve to be highly sensitive to random electrical noise, even allowing this noise to affect signaling outcomes?

A closer look at information processing in cortical neurons

To consider how random electrical noise affects the likelihood of a cortical neuron firing a signal, we have to look more closely at the ions (charged atoms) which are floating around. Each ion has no defined state in the present moment – the ion is instead a distribution of probability amplitudes across the x, y, z, and time axes. There is a fundamental uncertainty in the position and momentum of that atom.

So, in the present moment, an ion is not observable because it is not yet defined. But these ions must continue to exist – the first law of thermodynamics says that energy and matter cannot be created or destroyed, only converted. So any matter that is converted into a distribution of possible states must continue to exist. This distribution of component microstates – or information – must continue to exist as a physical quantity. This physical quantity is entropy, or a thermodynamic quantity of information.

Most of the time, entropy just dissipates. But a very specialized system - like the brain, which traps energy to do computational work - may be able to parse this information, and extract a signal from the noisy data. The signature of thermodynamic computation is extreme energy efficiency, combined with large amounts of computational power and flexible (non-reflexive) decision-making in a variety of contexts.

This computational process drives non-deterministic signaling outcomes

As probability amplitudes constructively and destructively interfere, resolving the system state in the present moment, information is physically compressed. And as information is compressed, a proportional amount of energy is returned to the system.

This energy is then available to do work, and that work involves encoding the meaning that was extracted during this computational process. This inherently probabilistic computational process results in an interesting phenomenon – synchronous but statistically random firing of sparsely-distributed neurons across the cerebral cortex. This so-called ‘populational coding’ is paired with a multi-sensory percept. And this coordinated neuronal firing drives coordinated behavior.

This new theoretical framework is completely compatible with modern neuroscience – it merely adds a mechanism to connect the noisy coding of cortical neurons to the system-wide synchronous firing that is observed periodically across the network.

The computational process produces perceivable information content

There is one additional feature associated with this process of non-deterministic computation, which is not associated with other types of computation.

As probability amplitudes constructively and destructively interfere, resolving the system state in the present moment, information is physically compressed and each neuron has some non-deterministic signaling outcome – it fires or it doesn’t. But this physical process has another facet. If the membrane of each computational unit also meets the criteria of a holographic recording surface - as biological neurons do - then any information that is physically encoded by interfering probability amplitudes will be paired with a holographic projection of that encoded information content. This perceptual content is exclusively accessed by the encoding structure and is representative of incoming sensory data in all available modalities.

In other words, consciousness is not raw, unfiltered access to reality. It is a reconstruction of reality, encoded by neurons in the cerebral cortex. And the very same process that generates perceptual content also yields non-deterministic signaling outcomes, which direct the action of the body within the perceived context.

Cortical neuron computation underlies the key features of consciousness

As we process information, our neural networks grow more ordered over time - as energy is trapped to do computational work, and lived experience is encoded into the neural network. And that's what thermodynamic computation is - it's about distributing time and energy resources to do the work of information processing, and then becoming an organized system with that gained understanding.

In this neuroscientific view, our brains generate both representative information content and volitional action, through a process of non-deterministic computation. Importantly, these phenomena only occur in neural networks which retain sensitivity to random electrical noise (such as cortex), but not in neural networks that are robust to random electrical noise when gating the action potential (such as spinal reflex circuits). The latter type of neural circuitry supports only deterministic computations, whereas the former type of neural circuitry supports non-deterministic computations, naturally paired with perceptual content.

Clinical applications of this research

If neuroscience can explain the emergence of perceptual content and non-deterministic outcomes from cortical neural activity, then we have the opportunity to better describe the relationship between the brain and the mind. This theoretical framework allows us to consider exactly what happens when critical steps in information processing go awry, how changes in computational processing correspond to mental illness, and what interventions are needed to support healthy neural computation. This concrete theory of mind provides a more neuroscientifically-grounded perspective for psychiatry, lends enormous support and justification for the efforts of mental health professionals, and offers new directions for the treatment of patients.