Neural Annealing

Neural Annealing is a proposal for how the human brain self-organizes and integrates information. It was developed at QRI, crystallized by Michael Edward Johnson in 2019, and has become one of the most foundational components of QRI's model of human consciousness.

For annealing to take place, the brain must enter a high-energy state to increase its plasticity and permit internal reorganization. This reorganized structure can be maintained once the brain returns to a low-energy state, provided the cooling proceeds sufficiently slowly. The frequency and depth with which a person anneals are proposed to be key variables for healthy brain function.

Energy in the brain can be increased by amplifying energy sources or by disabling energy sinks. If sources are amplified, this usually comes with semantic content that introduces a "bias" into the reorganization process. If the content is destructive (e.g., if it is psychologically damaging or comes with false factual assumptions), annealing has the potential for profoundly negative consequences. It is even possible for such annealing to be strategically utilized, e.g., to create emotional attachment to a group or aesthetic in another person. Conversely, positive content (consider metta meditation) can help steer the brain in a healthy direction.

If energy sinks are disabled, annealing tends to be semantically neutral. This type of annealing is primarily achieved through concentration meditation (where sustained attention is placed on a neutral sensory object) and is generally considered safer.

General Principles

An example atom structure called body-centered cubic (BCC) that's common in, e.g., iron at room temperature. (Note that the graphic only shows a cross-section of a pattern that extends in all spatial directions.) Structural damage to the material manifests as displacements of individual atoms from their ideal state, which can be repaired through annealing.

The Analogy to Material Annealing

Neural annealing is named after the process of annealing from the material sciences. In its healthy state, the atoms of most materials are arranged in a symmetrical lattice structure such as the one shown in the graphic to the right. Over time, this structure can deteriorate, e.g., due to the mechanical stress put onto a drill head during regular usage. On the macro level, such deterioration corresponds to a degradation of structural properties like decreased ductility and increased brittleness; on the atomic level, it corresponds to minor spatial displacements of atoms from their ideal positions in the grid. Material annealing is the process of temporarily heating the material to correct such dislocations and thereby repair structural damage.

At the level of particles, this increase in temperature leads to small, functionally random movements of each atom. Because the lattice structure constitutes a local minimum of the potential energy of the system, it is the equilibrium state that is naturally reached as a result of this process. Note that this makes material annealing an example of nonstandard computation, in which an optimization problem is solved as the natural result of numerous local forces acting in parallel.

While the introduction of energy, increase in plasticity, self-reorganization, and the subsequent cooling process are shared between both types of annealing, there are also two major disanalogies, which will be elaborated on in the next two sections.

Self-Organization and Semantic Content

Each material usually has a unique optimal atom structure at room temperature, so material annealing will converge to the same equilibrium every time. Conversely, the brain has many degrees of freedom in how it self-organizes. This additional dimension to neural annealing is how it can modify a person's belief network (see next section), which can encode personality traits such as one's aesthetic. Semantically loaded annealing can be considered a type of annealing in which the brain is given additional "instructions" on how to reorganize.

Energy in Neural Annealing

In material annealing, the energy needed to start the process is synonymous with heat. Conversely, the concept of energy in neural annealing is not reducible to a single, measurable quantity, but there are a number of correlates:

• total neural activity in the brain
• total activity in the brain's eigenmodes
• energy as defined in the Free Energy Principle (FEP)

Because of the FEP, sensory inputs generally act as energy sources (free energy), whereas top-down predictions act as energy sinks. Under normal conditions, this dynamic will result in a perpetual low-energy state since most sensory inputs are regular and easy to predict. Energy can be raised in a number of ways:

• Prediction errors (i.e., surprises) can increase energy since top-down models will fail to act as energy sinks. A large number of prediction errors may accumulate in an unfamiliar context, such as during a convention or a visit to a foreign country.
• Intense emotional experiences can increase energy for a similar reason. These can be romantic, sexual, intellectual, religious, spiritual, or even based on something mundane like watching a horror movie or playing an emotionally taxing game.
• Concentration meditation (i.e., sustained attention on sensory inputs despite their lack of surprise) will increase energy by disabling energy sinks. Meditation is the only mechanism which, at least for experienced meditators, is both safe (due to the lack of semantic content) and available on demand.
• Valence is an energy source; this can be utilized during meditation.

Furthermore, psychedelic substances can drastically increase energy by interfering directly on the physical level. Unlike meditation, the effects of psychedelics are highly volatile, with the potential for profoundly positive and profoundly negative consequences.

Annealing as Non-Local Belief Network Updates

An animation of a solution to the traveling salesman problem computed with simulated annealing

One of the models of human consciousness that QRI takes seriously is the Resonance Hierarchy. Under this model, consciousness acts like a hierarchically structured network in which feature recognition is implemented via physical resonance between different components.

At its core, the resonance hierarchy is a proposal for how an inherent drive toward higher resonance (or equivalently, to higher valence) can be utilized for computation. Within room-temperature consciousness, this drive typically acts locally, leading to valence gradients that can be utilized for feature recognition but will leave the network's global structure intact. In this setting, annealing is the principled mechanism to increase plasticity and thus allow for more global changes. The more deeply one anneals, the more drastic of a network overhaul is possible. This type of restructuring is loosely analogous to simulated annealing, which is an optimization algorithm (itself inspired by material annealing) that can be utilized as an alternative to gradient descent. In simulated annealing, the system repeatedly chooses a random neighboring state in the solution space and then either switches to it or not, where the new solution being superior increases the probability for a switch but does not guarantee it, thus mimicking the random movement of atoms in material annealing.

Note that while optimizing the resonance hierarchy is generally beneficial, repeated and excessive use of psychedelics may gradually anneal a worldview that either over- or underfits reality, since specific substances will bias the belief network toward higher (e.g., DMT) or lower (e.g., 5-MeO-DMT) complexity. In general, anything with the potential to significantly alter one's belief network should be recognized as a highly impactful intervention and be treated with appropriate caution.

Annealing as a Mechanism to Escape Negative Attractor States

As another application of this model, various adverse psychological conditions (trauma, depression, addiction, PTSD, etc.) can be modeled as sticky states, either of parts of the network or the network as a whole. These states are "sticky" or "attractors" in the sense that they are self-sustaining: once attained, they are difficult to escape. This may be true both due to low-level properties of how the condition is instantiated in the network, and due to high-level behavioral change. For instance, depression tends to lower the motivation to explore and enter new contexts, thus protecting itself by keeping the brain in a perpetual low-energy state.

Since escaping such states requires a significant overhaul of the network, annealing is a helpful and perhaps necessary step. Thus, the model provides an explanation for the success of psychedelic therapy.

That said, note that any specific condition may also have biological aspects and manifestations not captured by the neural annealing framework.

Practical Implications

The most straight-forward intervention suggested by the paradigm is to set up opportunities for neural annealing in one's personal life. This could mean an hour of meditation once per week (especially absorption practice, where attention is kept on a single object) or a meditation retreat once or twice per year. If it is available, semantically loaded annealing can serve the same purpose, provided the content is healthy.

Note that because the subsequent cooling should proceed slowly, activating too many energy sinks immediately after the meditation session might ruin the benefit. Thus, going to sleep or for a walk after a meditation session is recommended, whereas scrolling social media or engaging with intellectual work is not.

A different class of implications concerns theory of mind and social change. Identifying neural annealing as a mechanism to make concepts emotionally salient can explain the use of ritual, ceremony, and symbolism that is common in many cultures. Under the neural annealing framework, their primary function would be to evoke emotional responses in people, thus triggering a semantically loaded annealing process whose content is structured around memes or aesthetics relevant to the group or culture. Note that this explanation does not require intention; it simply predicts that memes that utilize neural annealing will tend to come out on top in an evolutionary process. The fact that such "malicious" forms of annealing occur absent of an intellectual understanding of annealing is also why discussing the topic is not considered an info hazard.

Neural Field Annealing

Neural Field Annealing is the latest iteration of the framework developed at QRI. Its core claim is that many of the phenomena of annealing can be understood as field effects and, furthermore, that one can train the nervous system to behave more as a field by treating it as one. E.g., if one has a low-valence internal representation, it may be helpful to treat it as a dissonant structure in the field of consciousness. Paying attention to such representations with a loving or caring attitude will send it vibratory patterns which, over time, can re-anneal them into a more consonant part of the field. (See the second link in the Resources section for details.)

Annealing and Sleep

Speculatively, sleep may be a form of annealing. In his original post, Mike Johnson wrote:

Is sleep a natural annealing process? If so, this could cleanly explain the connection between depression and chronic sleep disturbances — poor sleep as both a cause and effect of infrequent annealing. And it would indicate a treatment path: a restoration of normal annealing patterns may help improve both mood and sleep. I hold the following lightly, but we might model nREM as a ‘semantically-neutral’ high-energy state (pumping power into harmonics to perform biological housekeeping tasks, perhaps breaking up brain plaques) and REM as ‘randomized annealing’ as predictive models are switched back on, leading to both a semantic interpretation of this energy (dreaming) and the same sort of ‘neural search -> annealing’ consolidation and optimization process which happens after wakeful high-energy states. This would suggest that (1) nREM disruption might lead to primarily biological deficits, and REM disruption cognitive/emotional deficits, and that (2) if one anneals very regularly they may need somewhat less sleep, in particular REM.

One reason this result is tempting is that the purpose of sleep is still not well understood in biology and neuroscience, suggesting that a novel paradigm may be required. There are also examples of advanced meditators who haves suggested that meditation can substitute for sleep. That said, note that this substitutive role of meditation is likely limited for all but the most advanced meditators. Furthermore, the fact that increased sleep cannot substitute for meditation implies that even if sleep is a type of annealing, there are substantial differences between the two.

Resources

• Neural Annealing: Toward a Neural Theory of Everything – the original post introducing neural annealing, by Mike Johnson.
• Neural Field Annealing and Psychedelic Thermodynamics – the presentation introducting Neural Field Annealing, by QRI co-founder and president Andrés Gómez Emilsson.
• Neural Annealing (Presented to Dr. Karl Friston @ the Wellcome Centre for Human Neuroimaging @ UCL) – a presentation by former QRI member Quintin Frerichs.
• Healing Trauma with Neural Annealing – a blog post on the therapeutic potential of neural annealing, by Andrés.