Beautiful Loop Theory

Introduction

I've been interested in theories of neural correlates of consciousness for some time now. I've also been interested in altered states of consciousness (especially those produced by meditation).
Recently, I came across a paper that proposes a theory of neural correlates of consciousness that claims to provide a unified framework to explain the altered states of consciousness of meditation, psychedelics and dreams. Obviously, it's not the first theory to take an interest in them, but it's one of the few to give them such a central place.
Naturally, it sparked my curiosity.

The theory is called Beautiful Loop Theory (from now on, abbreviated as BLT), and is based on active inference.

The authors are Ruben Laukkonen, Shamil Chandaria and, in the second version of the paper, the famous Karl Friston.

Laukkonen: “I’m a Senior Lecturer (US: Associate Professor) in cognitive science and computational neuroscience. My research is theoretically driven and traverses many levels, from neurons to psychology. Key topics include: meditation, insight, consciousness, and artificial minds. I’m also a chief investigator on Australia’s largest psychedelic clinical trial, supported by ~$4M from the Medical Research Future Fund (MRFF) and the Peregrine Foundation.” His website and his Substack.
Shamil Chandaria: “Dr. Shamil Chandaria is a distinguished expert in artificial intelligence, computational neuroscience, and mathematical modeling, with over 30 years of experience spanning multiple fields. His academic journey began at Cambridge University, where he studied Natural Sciences and Economics, followed by a PhD from the London School of Economics, specializing in mathematical modeling of financial systems using stochastic differential equations and optimal control theory. He also holds an MA in Philosophy from University College London, where he developed a keen interest in the philosophy of science, biology, and neuroscience. Beyond academia, Dr. Chandaria is an entrepreneur, philanthropist, and dedicated meditator. He co-founded the Centre for Psychedelic Research at Imperial College London, the world's first psychedelic research center, and actively supports the Global Priorities Institute at Oxford University. He also funds groundbreaking research on the neuroscience of meditation at Harvard University and UC Berkeley. Recognized for his contributions to Science and Technology, Finance, and Philanthropy, he was awarded an OBE in 2022.” His Youtube channel.
Karl Friston: His Wikipedia page.

In this post, I'm going to briefly introduce BLT by making some simplifications (sorry in advance if I make some mistakes). I will then present how it accounts for altered states of consciousness in meditation.

I've relied mainly on this paper. But reading some other papers by Laukkonen or Friston was also useful. You can also read Laukkonen's own Substack post on BLT.

BLT

There are three central elements to BLT:
1. A hierarchical active inference system
2. A Reality Model
3. Recursivity

I'm going to quickly go over what a hierarchical active inference system is, as it's complex and fairly well-known. Those who want more details can consult the paper (or this one) that gives a more detailed summary.

But roughly speaking, the idea is that we have a system that is constantly formulating theories to explain our sensory data, based on something like its “beliefs” (priors). It tries to minimize the discrepancy between its predictions and the sensory data; when there's a discrepancy, it either modifies its beliefs, or it modifies the organism's behavior to obtain sensory data that match (hence the “active”).
This system is also hierarchical; it contains several “layers”, the higher a layer is, the more abstract and temporally extended structures it manages, and the lower a layer is, the more simple and immediate features it manages.
Finally, this hierarchy is precision-weighted. Roughly speaking, it's as if each layer were assigned an “importance”/“salience”/“confidence”; the more precision a layer has, the more it weighs heavily in the system and can override the prediction errors reported by the other layers and impose its own theories. The distribution of this “confidence” can change according to a number of factors.
Perception, in this framework, is not a passive phenomenon; it's the construction of explanatory theories of our sensory data. Two people whose hierarchical active inference system does not have the same “beliefs” or the same “confidence” (precision) distribution may not perceive the same thing even though they are looking at the same thing.
Also, perception is a patchwork of theories that come from many layers (the hierarchical active inference system produces a coherent and “integrated” multi-level account); if we were to remove the top layers, our experience would become less and less “thick”, revealing a less and less fabricated experience on which fewer and fewer abstract and temporally extended concepts are imposed, as if we were becoming a newborn baby, or worse.

Active inference is a fascinating model. It claims to unite under a single framework such things as perception, action, attention, illusions, learning, certain mental illnesses, etc.

But the paper's authors note that active inference is rarely seen as a theory of consciousness per se. They would like to extend active inference into a theory of consciousness. To do this, they add 2 things to active inference: the reality model and recursivity.

The idea of the reality model, to put it simply, is that in the hierarchical active inference system, there is competition between different theories to explain our sensory data, and the best or good enough theories are “sent” into the reality model. The reality model contains, so to speak, the theories currently “accepted by the brain” about “what's going on in the world”.
(This is perhaps somewhat recalling elements of GNWT, where crossing a certain threshold makes a piece of information globally accessible. It's not the same, but there's a similarity.)

Recursivity is the most important point; according to the authors, it's the key to consciousness.
The idea is that the content of the reality model, although an output of the hierarchical active inference system, is fed back as an input into the hierarchical active inference system. (For example, if a theory from the hierarchical active inference system is to win the competition, it must be consistent with the actual content of the reality model, hence the usefulness of a form of recursivity.)

But recursivity is a bit “fluid”. Sometimes recursivity can be “stronger” or “weaker”. Some elements of the reality model have stronger recursivity than others, and which elements get more recursivity changes according to several factors.

According to the paper's authors, we are aware of something in the reality model when that thing is in a recursive loop with the hierarchical active inference system, and awareness is proportional to the strength of the recursion; the “clarity” and “scope” of awareness are variable.
(This is perhaps somewhat recalling elements of HOT. It's not the same, but there's a similarity.)

The authors introduce the term “epistemic depth” to describe the level of intensity of the recursivity. Since consciousness depends on recursivity, the higher the epistemic depth, the more aware we are of the content of the reality model.

As a result, according to BLT, each conscious state can be described by two (or three) aspects: the distribution of precision in the hierarchy, and epistemic depth.
For example, during sleep with few/no dreams, precision drops throughout the hierarchical active inference system, resulting in a very poor or even empty reality model. On top of this, epistemic depth is drastically reduced, resulting in minimal/absent awareness.
In dream sleep, on the other hand, the hierarchical active inference system and epistemic depth are more active, giving a richer content to the reality model and a reflexivity that allows for some consciousness.

Altered states of consciousness in meditation

A well-known altered state of consciousness in meditative practice is minimal phenomenal awareness. This is a state where the “content” of consciousness is minimal, but where awareness remains very present. It's a state described as “consciousness that is only aware of itself”, as a state of “luminous purity”.

According to BLT, advanced meditators can “control” their precision distribution. To enter a state of minimal phenomenal awareness, they start by reducing the precision of the upper layers of the hierarchy, so as to have an almost raw experience. Then, precision is also withdrawn from the lower layers, to have an almost empty reality model. But the meditator retains a high level of epistemic depth (recursivity). This leads to a kind of weird loop where the only signal present is the recursion signal itself.

Another well known altered state of consciousness in meditative practice is non-dual awareness.
Non-dual awareness can be used to describe a number of things. To name the main ones; an experience without subject-object duality, a minimally conceptual experience, and a conceptual experience but without reification (non-dual awareness can sometimes even encompass minimal phenomenal awareness).

According to BLT, conceptual experience without reification comes from a modification of beliefs in the upper layers of the hierarchy. According to BLT, this modification comes mainly from introspection during an increase in epistemic depth:

By making contents of the reality model an object of awareness (epistemic depth) and inquiring into the three characteristics (Burbea, 2014), one’s priors, which influence experience, may begin to restructure (i.e., insight) […]
Thus, high levels of epistemic depth increase the probability, especially for advanced meditators, that phenomena will be perceived as mental constructions and therefore the commonsense phenomenology of naïve realism dissolves. When phenomena are so perceived, they are said in Buddhist terms to be empty (śūnyatā, Burbea 2014).

Regarding minimally conceptual experience and experience without subject-object duality, it seems to me that the paper doesn't mention them (I may have missed that bit), but I think it's fairly safe to guess their account under BLT.
Minimally conceptual experience is to be explained by a reduction in the precision of the upper layers of the hierarchy, giving minimally fabricated experience (this is kind of the first step in achieving minimal phenomenal awareness).
Experience without subject-object duality must be explained by a modification of the belief in a self of a higher layer of the hierarchy achieved by different types of meditation, or by the reduction of the precision of this layer during meditation.

Discussion

BLT is fascinating and beautiful. Still, I wonder how it can explain other qualities frequently reported in meditative practices; unifying quality (the experience of the multiplicity of objects of perception as nevertheless united; everything is in fact perceived as “One”), inner subjective quality (the perception of an inner subjectivity to all things, even those usually experienced in purely material forms), noetic quality (the experience as a strong source of valid knowledge; it seems that the world and its contents are experienced as they really are, without distortion), ineffability, religious quality (the intrinsic sacredness of the experience; this includes feelings of mystery, awe, and reverence that may nevertheless be expressed independently of traditional religious language), etc.
(Maybe I missed the bits about it in the paper.)

Perhaps unifying quality and inner subjective quality can be explained by a change in the beliefs of the upper layers (but then it's strange that you can enter and leave a state with these qualities so easily and quickly) or a decrease in their precision (but then it's strange because it looks more like a change in the beliefs of the upper layers than a reduction in their importance).
And perhaps ineffability can be explained by an extremely different reality model due to a change in the beliefs of the upper layers or a decrease in their precision. Or maybe by making experience minimally conceptual, we get this sense of ineffability because language relies on concepts.

But I'm not sure how BLT explains religious quality.

The paper discusses noetic quality. It explains it as being due to the increase in epistemic depth which creates both noetic quality and the feeling of “expanded consciousness”. I don't know what to think of this explanation. I mean, can't derealization occur at any level of epistemic depth? And isn't derealization a bit opposed to noetic quality?
Perhaps BLT can explain derealization by relying on a theory such as this, which holds that it is due to a decrease in interoceptive prediction precision. Derealization would be a special case that doesn't threaten the theory. But still, it seems that this may suggest that noetic quality is more complex than a simple increase in epistemic depth. Perhaps an increase in epistemic depth can explain the feeling of expanded consciousness, whereas noetic quality is multifactorial.

Another question I have is how BTL explains that advanced meditators can move very quickly from a normal state of consciousness to an altered one, or very quickly from an altered state to a normal one.
Some altered states of consciousness are explained by BTL as a change in precision distribution, with advanced meditators having great control over this. But other altered states of consciousness are explained by a change in the beliefs of the higher layers of the hierarchy.
Perhaps BLT should consider that not only do advanced meditators have control over their precision distribution, but also have, to some extent, control over the priors of their higher layers.
Alternatively, one might consider that BLT implies that, in fact, meditators cannot freely enter and exit any altered state of consciousness; only those states induced by a change in the distribution of precision, not those induced by a change in the beliefs of the higher layers. It's a nice prediction, and one that may already be confirmed. Meditators often say that you can't really get out of certain states of enlightenment, unlike stream entry.

In any case, BLT is fascinating and splendid, and I can attribute my wonderings to errors on my part.