The Quaternion Process Theory of Human Cognition
I’ve always believed that the quirks in the behavior of the human brain were clues to its working. Daniel Kahneman’s observations about thinking fast and slow (i.e. Dual-Process theory) kicked started my journey on this research program. Kahneman is a behavioral psychologist that works on judgment and decision-making. System 1 and System 2 are Kahneman’s categorization of behavior that he has observed, they are not mapped to brain regions.
I’ve come to the belief that system 2 doesn’t exist, but rather it’s an emulation built on top of system 1. Thus deliberate thought is a consequence of the intuitive system of system 1. Evidence of this comes from AI research specifically AlphaZero. (see: AlphaZero: How Intuition Demolished Logic)
This differs from Ian McGilchrist a clinical psychologist who has explored the difference in function between the two brain hemispheres. McGilchrist separates brain function into two parts. The left hemisphere tends to see isolated elements, isolated fragments, but the right hemisphere tends to see a picture in which everything ultimately is connected and nothing is seen out of its context.
The left hemisphere of you being very narrow sheers off the context. The left aims to capture. It takes snapshots that are very thin in space and time. The right is seeing a flowing picture, which is moving all the time and changing.
The right understands implicit information, meaning body language, reading the face, metaphors, sense of humor, puns, and irony. The left has a good lexicon and some rules of grammar for creating sentences. The right hemisphere tends to see things as animate. The left hemisphere seems to prefer machines, tools and things that are inanimate.
The right hemisphere, presences the world and the left creates a map of the world. McGilchrist says “you would be a bit of an idiot if you mistook the map for the place that is mapped and tried to live in the map.” The left hemisphere has extreme optimism, and the left hemisphere is full of BS. The right hemisphere is more pessimistic than it needs to be. They’ve evolved to compliment one another, but their relationship has a problem.
The left hemisphere knows less. It’s limited in its understanding. It believes that it knows everything. The right hemisphere, knowing far more doubts that it knows everything. I guess paralleling the culture of certainty and the culture of doubt (i.e. religion vs science). As long as the right hemisphere is willing to be guided by the left that sees more, then things work well.
There’s some overlap between Kahnman’s dual-process theory but with distinct differences.
Given McGilchrist’s observations about the human mind, I conjecture that there are *two* kinds of system 1s. The left and the right hemispheres kind. There are also two kinds of system 2 slow thinkers. This new framing reveals two new questions. What is a left hemisphere intuitive thinker? What’s a right hemisphere slow deliberate thinker? The first kind reveals itself in people like Rain Man who can perform calculations in their heads very rapidly. The second kind reveals itself in creativity that takes days to stew.
McGilchrist offers four ways that we discover truth. These are reason, science, intuition, and imagination. In this framing, we see that reason and intuition are System 1, and science and imagination are System 2. Alternatively might also label the two systems as one that employs empathy and the other that employs fluency. These are complementary modes of thinking that are required to achieve human complete intelligence.
I call this the Quaternion Process Theory (QPT) of cognition. To pay homage to Kahneman’s dual-process theory and quaternion emphasized four better than quadratic. Besides, the most fundamental thing that acts like consciousness is the quantum spin and it's also expressed in Quaternions. ;-) Hence here are the four kinds of thinking:
Via participatory knowing. Discovering affordances of things and blending concepts to arrive at new things with new affordances.
Via procedural knowing. Discovering affordances of things by seeking language expressions of description and that language has its own affordances.
Via perspectival knowing. Discovering to the expression of instruction and shapes the instruction of imagination. That expressions convey the coordination of shared intentions.
Via propositional knowing. Discovering that different expressions have commonly shared principles. That these principles lead to new questions and thus new hypotheses.
Both hemispheres follow this staircase model of competence.
Observe that System 1 thinking is at the level of competence. Think of it as amortized inference or habits. We become habituated to certain kinds of empathy or fluent thinking.
Unlike the dual-process theory, this theory fills a gap that has been nagging at me when I wondered where ingenuity would emerge from deep learning.
Where is the Artificial Ingenuity in Deep Learning?
What is ingenuity and where does it originate from?
Thus the Quaternion process theory consists of four systems. Consisting of two massively parallel processes focused on empathy and fluency. Each process exhibits unconscious and reflective modes.
General intelligence is likely to be a consequence of complementary thinking modes. At different levels of complexity, you have complimentary processes that coordinate to solve a problem. These combine to construct a new process that also coordinates with another process.
So what we have are rich shared conversational spaces in minds. The notion of strict protocols between the mind’s modules needs to be replaced by the idea of loosely coupled conversational protocols between adaptive modules.
In our inclination towards reductionist ideas, we prematurely categorize the modules of a mind. As a consequence, we end up with inflexible and brittle systems. This has been a design flaw of AI systems from the very beginning.
It’s the lack of recognition that simpler solutions can be discovered by being there (i.e. subjective framing) than a universal generalized solution. Reality is sufficiently messy that only a patchwork of complementary solutions will suffice.
Hence we have brains that have two hemispheres with complementary functionality. All general intelligence is a consequence of collections of intelligence. We are who we are because we are in constant collaboration.
System 2 processes are conscious processes, hence we might study the two ways that consciousness research has described consciousness. A-Consciousness is representational, and P-Consciousness is phenomenal. A-Consciousness is functional, and P-Consciousness is non-functional. (see: https://philosophy.tamucc.edu/notes/p-and-a-consciousness…)
It seems to me that consciousness is whatever that is remembered in the right or left brain. If you can’t remember something, you can’t be aware of that something and thus are not conscious of it. Said differently, processes that have as a side-effect a generation of a representation that can be subsequently inspected in episodic memory are conscious System 2 processes.
But what about computers that remember everything, are they conscious? Not in the sense of A-consciousness and P-consciousness because they operate very differently from human brains. We have yet to build cognitive technologies that are architected like biological brains. Deep Learning systems have their own unique dynamics that may not operate like brains. Furthermore, their lack of a body implies an absence of P-consciousness.
Let’s ponder the subject of consciousness and intelligence. These two subjects are often conflated. Many assume that human consciousness leads to human intelligence. Is this true or can we achieve human intelligence without human consciousness?
When we discuss consciousness there are two kinds P-consciousness ( how it feels to be ) and A-consciousness ( awareness, attention). But when we speak of human intelligence we should include all the cognitive characteristics that make up the human mind. Consciousness is not everything about the mind. It is only the part that our minds can introspect about themselves. It’s the reflective and deliberate part of the mind. The slow part of the mind, the system 2 in Kahneman’s dual-process theory. It’s a subset of who we are.
Human intelligence is uniquely tuned toward shared intentional behavior (see: Tomaselo). We are a cooperative species and it’s through language that we coordinate in highly sophisticated ways. Human intelligence is a consequence of our linguistic bodies. The question then is reframed, can we have shared intentions without consciousness? Can we recognize another party’s attention without A-consciousness? Can we have empathy without P-consciousness?
The key characteristic of consciousness is its reflective and introspective quality. An agent that is absent of these capabilities will be encumbered. But an agent that does may not be at the same level of the spectrum found in humans. But how do we characterize degrees of consciousness? Perhaps degrees of self-referentiality. Which strongly hints at the employment of code duality principles in achieving this.