It’s a truism that everyone has a different idea of what they’re talking about when they refer to “consciousness”, and so there are very many different theories of consciousness. Psychule Theory proposes that (almost) all of these theories have the same fundamental basis, namely, pattern recognition (well-defined, see below). Thus, Psychule Theory is a “minimal unifying model” as described by Wanja Wiese. The term “psychule” is meant to indicate the minimal unit of consciousness, just as a molecule is the minimal unit of a substance.
Psychule Theory claims that consciousness can be entirely explained using only what we know about physics, but that explanation also requires an understanding of properties which are not physically measurable but nevertheless derive from physics. These properties are goals, information, and communication. This essay will provide a brief description of the natural physical derivation of these properties. It will then describe the impetus and requirements for the instantiation of “pattern recognition” as a base unit of consciousness, and something of the natural history of complexification by reiteration of this base unit up to a general structure applicable to the human brain. The essay will then show how this theory applies to and/or explains some other major theories of consciousness, and finally will touch on more exotic examples of consciousness, like AI.
Measurement as pattern detection
When physical entities interact they follow specific patterns which we call the laws of physics. Thus every physical interaction results in a pattern realization, the pattern being the combination of (measurable) prior states plus the (measurable) resulting states. The pattern is “real” because the resulting states are in fact measurable.
Goals
A system can be said to have a goal, by definition here, if
1. there is a specific (measurable) state of the world called the goal state,
2. the system tends to select mechanisms which can measure (detect) the current state of the world,
3. the system tends to select mechanisms which can move the world toward the goal state, and
4. the system tends to coordinate the detection of a specific current state with a mechanism which moves the world toward the goal state.
Another way to say this is that the system tends to generate mechanisms which detect a discrepancy from the goal state and respond with mechanisms which move the world toward the goal state. The first such system we will be concerned with is the one we call Nature, the system which is responsible for Life.
Information
Because physical interactions follow laws (patterns), any given physical state, detectable by measurement, is the result of prior interactions of entities in different (from the current) measurable states. Every physical interaction we care about involves a physical change. Because only certain prior (measurable) states could possibly lead to the current (measurable) state, measuring a current state potentially provides us with indications of what the prior states were. These indications can be thought of as counterfactuals with associated probabilities. Eg., if we measure X at time 1, there is a probability that at time 0 we would have measured Y. We call this property, i.e., the property of indications resulting from measurements, mutual information. We say X shares mutual information with Y. This property becomes an affordance for a system with one or more goals.
Communication
As stated above, a system with a goal can further that goal by selecting and coordinating two mechanisms, a detection mechanism and a response mechanism. In some cases the response can be directly linked to the detection mechanism. A classic example of this is the Watt governor (google it) or analog thermostat. In other cases, the response mechanism will necessarily be at some distance from the detection mechanism. A classic example would be in bacterial chemotaxis, wherein a cell surface receptor must detect a specific molecule (sugar?), but that detection has to be conveyed to a mechanism on the other side of the cell, the flagellum. In this case the detection mechanism can generate a signal, such as a specific molecule. As described above, this signal molecule will bear mutual information with respect to the pattern detected. This signal can then be transported to the response mechanism, where it can be detected and responded to. Note that the physical structure of this signal is arbitrary to the extent that it can be produced by a detection mechanism and transported to a response mechanism. The key is that the mutual information with respect to the detected pattern is conveyed to the appropriate response mechanism. This combination of pattern detection mechanism, signal medium, and response mechanism is what I call pattern recognition and designate a specific example as the psychule, the minimal unit of consciousness.
Minimal Unifying Model
In a recent paper, Wanja Wiese identified the need for a Minimal Unifying Model (MUM) (https://academic.oup.com/nc/article/2020/1/niaa013/5870169). Psychule Theory purports to provide such a model, designating the psychule as the minimal unit of consciousness. As explained elsewhere, several current theories can be unified by explaining how they are based on this form of pattern recognition (psychules). But here I will describe a very abbreviated natural history leading from the simplest psychules to the complex varieties in the current human brain.
Evolution
The first psychules were probably associated with cell-surface receptors in single cell organisms like the chemotaxis receptor mentioned above. Another example of early psychules involves cells communicating with each other. Note there are two kinds of such communication to consider: loud (e.g. releasing a molecule outside the cell which goes to multiple targets) and quiet (releasing molecules directly into single targets via a gap junction).
Next is the development of cells whose sole general purpose is to perform these detections/communications: aka neurons. Via neurons, communication can go to very specific distant cells. And once you have neurons, you can develop very sophisticated communications with other neurons, giving new, useful, and complex information processing (Neural nets) all based on psychule-type pattern recognition. These neural nets allow for higher order pattern recognitions, i.e., recognitions of recognitions (… of recognitions … and so on) as well as feedforward and feedback (prediction) signals.
Now I will propose two structures which I think are significant for how consciousness functions in advanced (human) brains. The first is a general purpose pattern recognizer. In her book “Beyond Concepts”, Ruth Millikan introduces the idea of Unicepts and Unitrackers. In short, Unitrackers are general purpose pattern recognition mechanisms. They learn to track a single concept (thus unicept), although the exact pattern they track is plastic in that it can shift or even disappear. (See here for an online description: https://philosophyofbrains.com/2025/09/15/ruth-millikan-unicepts.aspx) I currently postulate that the role of unitracker in the human cortex is played by the L4 pyramidal cells and the L2,3 cells associated with each. As the L4 cells receive the main inputs from the thalamus, I suspect they are key constituents of the unitrackers. Note, however, that a single unitracker might be constituted by a set of L4’s (and related L2,3), and also that a single L4 may be part of more than one unitracker. If I were a neuroscientist, this is the area I would explore first.
The second significant neural structure is the semantic pointer which is described by Chris Eliasmith’s Semantic Pointer Architecture (SPA). A semantic pointer is essentially a convolutional neural network. This is the kind of network where you can famously perform operations like “King – Man + Woman” and the network would output “Queen”. The gist of the semantic pointer is that you can apply multiple input vectors to a network from individual sources, then perform operations on those vectors individually or as a group, and then input a query which will return one of, or a set of, those input vectors. The outputs are “semantic pointers” in that they can (top down) activate the sources of the appropriate inputs. It’s also important that semantic pointers can be used to put specific inputs into a context. For example, if the context is “size”, each item can be entered as “thing+size”, such as “dog+1” for Chihuahua, “dog+8” for Great Dane, etc. This context could potentially be important for which unitrackers in the “audience” of this pointer get activated.
So how do these structures apply to psychules? Obviously the unitrackers are acting as pattern detectors, but from where are the patterns that they detect coming? In some cases they’re coming directly from other unitrackers (L4->L2,3 —>L4), but in other cases they’re coming from intermediate networks acting as semantic pointers, such as an area of the thalamus. A semantic pointer network (eg. one of several in thalamus) would act as a viewing screen or a broadcasting workspace for an audience of unitrackers. Those unitrackers which are activated by this screen/workspace could then generate a combination of responses which could include feedforward to the next set of unitrackers in a hierarchy, feedback to the screen or prior unitrackers, and some additional action(s) for control purposes (muscular, endocrine, attention, etc.)
Other Theories of Consciousness
How does Psychule Theory explain consciousness? To my understanding there are three main categories of consciousness theories: top down, bottom up, and out of nowhere. Top down theories essentially start with the most complicated form of consciousness, human, and try explain the phenomena they find. The obvious examples are Global Neuronal Workspace, Predictive Processing, Attention Schema. Each of these can be explained by layers of neural networks, some of which constitute an audience of unitrackers while others constitute semantic pointers. The semantic pointers are “workspaces” which get inputs from lower unitrackers and broadcast to an audience of higher unitrackers. The outputs from unitrackers include feedforward as well as feedback (prediction) to lower layers. And it’s possible to control attention by activating some unitrackers from the top (as opposed to bottom up) and suppressing others, thus altering their reactivity patterns.
The major bottom up theory is Integrated Information Theory (IIT). Psychule Theory (also a bottom up theory) largely agrees with IIT, but without the more mind-bending parts. As far as psychule theory is concerned, IIT’s value of Phi is a perfectly good way to identify an instance of a given pattern recognition and quantify the mutual information that detection has with respect to a pattern detected. Nevertheless, I think recent work under the title of Integrated Information Decomposition does a better job of that.
Another type of bottom up-ish theories are those which emphasize “feelings” and “valence”. Bottom up-*ish* because they actually start in the middle, but then aim up. The prime example is Mark Solm’s theory, and the primary indicator is the example of anencephalic patients who essentially lack a cerebral cortex and yet show indications of consciousness. Psychule theory describes the responsible psychules as “loud” because the pattern detections happen in the lower or mid brain but tend to generate their effect via systemic signals like neuroregulators, hormones, etc.
“What it’s like to be something”
Finally, I will try to address the proposition that something is conscious if there it is something it is like to be that thing.
The hard part of explaining consciousness is explaining David Chalmers’ “Hard Problem”, aka that feature of “what it’s like” which is also referred to as phenomenal consciousness. Psychule Theory says phenomenal consciousness simply is the first person perspective of a system that recognizes patterns. Actually, to know what something “is like” assumes a capability to compare one “thing” with another “thing” in a context. So a system with just simple psychules would have no means to compare them per se. The simple example is the bacterium which has psychules only for attraction (chemotaxis). There is no sense in which this bacterium can compare this Psychule with another. However, if this bacterium has an additional, separate Psychule for repulsion (toxins), and there is a mechanism which determines which signal is “stronger” so as to determine the final outcome, one could argue this outcome constitutes such a comparison. “All sugar and no toxin” is more like “lots of sugar and a little toxin”, in that the outcome is similar, as opposed to “no sugar and lots of toxin”. But again, in the bacterium there is no higher-order mechanism to recognize this difference. It simply experiences, or doesn’t.
At the other end of the complexity spectrum, people do have this capability of putting prior recognitions into context, and note this capability is a feature of semantic pointers. The judgment of being more or less alike can be performed by the unitrackers responding to those semantic pointers. As such, the question of “what it’s like” assumes a very high-level complex Psychule structure. Also notice that for any such structure with multiple layers of unitrackers and semantic pointers, only those patterns which reach the highest level so as to be available for action, planning, memory, and report are the patterns being called “phenomenally conscious”, whereas the lower level patterns which do not get propagated to the top are often described as “unconscious” even though “subconscious” might be more accurate.
“What it’s like” to be an Artificial Consciousness
Consider a toy system that has a camera, can recognize just a few (say ten) things, and responds to those recognitions by sending tokens to an LLM, also part of the system, which LLM has been trained on those tokens. The LLM has its own perspective on those recognitions. It doesn’t know anything about the tokens. All it can say is that one or more of those ten recognizable things are recognized, or not. From its perspective the recognitions are just given. The whole system’s Umwelt is those ten things (plus words presumably). Everything it can say about those recognitions would line up with what you say about your phenomenal experience. It has them, it knows it has them, but it can’t tell you what it is that it has. This is exactly what people say about their own conscious experiences. Psychule Theory says that for the system to explain (via the LLM’s perspective) what is happening, it would use language that tracks exactly how people describe their own experience, and this theory hypothesizes the system would do so for the exact same reasons that people do.
Simulation Argument
Some who argue that AI’s aren’t/won’t be conscious bring up the “simulated rainstorm doesn’t make anything wet” argument. (I’ll avoid Bernardo Kastrup’s simulated kidneys). The computational functionalist (well, my) response is that yes in fact, simulated rain does make things wet … in the simulation. The simulated umbrella gets simulated wet. The simulated Jupiter would revolve around the simulated sun because of the simulated gravity. And a perfectly simulated person would make all the same arguments that he is conscious, and for the exact same reasons that a non-simulated person would argue that they are conscious. I’ll say it again … the exact same reasons. We could be in a simulation now. How would you know? Answer is you cannot know (unless the simulator somehow informs you). But whether we’re in a simulation does not impact whether we are conscious.
And note that if the input is not simulated, if the input is real, as in the example of the LLM with a camera given above, then the associated psychules are also not simulated. They’re just psychules.
That’s the short form of Psychule Theory. Some day I may generate a longer form. Hopefully.
