NB: Originally published on Map and Territory on Medium. This is an old post originally published on 2016-09-14. It was never previously cross-posted or linked on LessWrong, so I'm adding it now for posterity. It's old enough that I can no longer confidently endorse it, and I won't bother trying to defend it if you find something wrong, but it might still be interesting.
In a recent post Scott Alexander gives a review of some recent results in neurobiology that suggest a powerful, unifying set of mechanisms for how information is integrated in the brain. I recommend you read his article and the original research if you can, but I’ll summarize it briefly.
There are various chemicals regulating activity in the brain. There is now evidence that these chemicals act in coordinating an information pump in the brain. Change the chemicals and you change the parameters of the information pump. It seems specifically the information pump in play fits the Bayesian model in that certain chemicals regulate the presentation of prior evidence, others new evidence, and yet others confidence in those evidences.
What I find compelling is that the model described provides a plausible mechanism by which the theory of a 2-part psyche might work. There are several two-part theories of psyche, that is to say theories of how mental processes are organized. My preferred one is near/far construal theory, but there is also the S1/S2 distinction, the fast/slow distinction, in Chinese philosophy yin and yang, and even the hot/cold blood model from medieval European thought. Each of these acts as a way of classifying thoughts and behaviors along a spectrum between two extremes.
The interesting thing about the 2-part psyche theories, and why I prefer the near/far distinction, is that they all seem to operate along the same dimension. Near/far uses the metaphor of distance (because it happens we use similar reasoning patterns when working with things that are physically near versus far) to differentiate between things that are heavy on details and light on patterns versus those that are heavy on patterns and light on details. S1/S2 uses basically the same dimension, as does fast/slow, yin/yang, and hot/cold: stuff with lots of details is near, fast, yin, hot, and part of S1 while stuff with less details and stronger patterns are far, slow, yang, cold, and part of S2. This suggests that they are all pointing at the same sort of thing, though in slightly different ways.
And, as it happens, this is basically the same dimension along with chemicals in the brain seem to affect cognition, balancing between how much to weigh new evidence (details) against prior evidence (patterns). So it seems that we now have a plausible biological basis for the two-part psyche we’ve reasoned exists and find useful, whereas before it was just a pattern that worked without strong evidence of a mechanism.
The 3-Part Psyche
So that takes care of the 2-part psyche, but what about the arguably more popular 3-part psyche model. The 3-part model dates back at least to Aristotle in the West and the gunas in India, was revitalized by Freud, and has bloomed into various descendent theories in modern psychology such as Internal Family Systems. Each version has different boundaries and explanations, so for simplicity I’ll use Freud’s well-known terminology.
Briefly, these theories all see roughly the same three parts in the psyche: the id, the ego, and the superego. The id is the part that acts and responds “on instinct”, the ego is the part that is “rational” and integrates the other two, and the superego is the part that operates on “moral” grounds. These parts are viewed as working in relation to one another, frequently in opposition, with what someone does and thinks arising from their interaction.
These theories connect with the 2-part psyche model in that near corresponds to aspects of the id and ego and far corresponds to aspects of ego and superego. When we see this kind of correspondence with superposition, it suggests both are mixing up the underlying reality in different ways. We can use this to try to pick apart what’s really going on.
The commonalities of id and ego seem to be an inclusion of details, same as for near. What’s different is that ego has a concept of integration with patterns whereas near and id do not.
The commonalities of ego and superego are just the opposite: inclusion of patterns. Same goes for far. The differences are that ego includes details while far and superego do not.
I propose from this that if we separate out details and patterns onto separate dimensions we can get a 2-dimensional model that captures both the 2-part and 3-part psyche models and even suggests a 4-part psyche model.
Now the 2-part model corresponds to the line between the more details, less patterns corner of the space and the less details, more patterns corner with near and far as their division down the middle, respectively. This is drawn as a dotted line in the above chart.
The 3-part models corresponds to 3 of the 4 quadrants formed around the middle of the 2-dimensional space: id is more details with less patterns, ego is more details with more patterns, and superego is more patterns with less details. This also leaves a suspiciously empty 4th quadrant to be part of the psyche with less details and less patterns.
And, to make things even better, this fits with the biological model Scott summarizes: there are chemicals regulating how much to favor details and how much to favor patterns. Normal thinking and behavior fall in the ego quadrant or at least near the center while mental disorders appear when the chemical regulation of detail and pattern strength are out of their typical balance.
So for all their faults in the past, maybe our theories of the psyche have been pointing us in the right direction all along, just in a confused way.
The 4th Quadrant
This still leaves us with the fourth quadrant that’s gone unaddressed. Here I’ll offer some brief speculation on what it might be before wrapping up.
Since this theory points to something that will feel qualitatively different to us from the inside the way id, ego, and superego do when both details and patterns are weak, we should go looking for things we consider mental states that don’t fit well in the existing 2-part or 3-part models. One immediately comes to mind: dreams.
Dreams are, among other things, a time when you have low sensory information and seem to have trouble completing patterns. We talk about “dream logic” because in dreams you can jump between fitting patterns to limited data that often violate the causal narrative we expect to find in our thinking. And dreams seem to incorporate memories, often recent and important memories, in place of outside sensory data. This is by no means a slam dunk, but it does weakly fit the evidence.
Which is the point on which I wish to end: all of this is based on fairly weak evidence. Although the 2-part and 3-part psyche models are fairly robust, they have always had problems because they readily fall apart upon rigorous inspection and have not had a clear biological basis so are subject to introspection bias. Additionally, my new evidence from Scott is an interpretation of an interpretation of recent findings and stretches well beyond what we can safely conclude.
At the same time, these ideas are exciting and, I think, worth exploring because they give us a potential model for better understanding human thoughts and behavior. I fully expect this 2-dimensional, 4-part psyche of details and patterns to make wrong predictions, but I’m hopeful it makes more right predictions and fewer wrong predictions than either 2-part or 3-part psyche models do. I look forward to testing them and seeing what more we can learn about our messy selves.