System 2 / Self-Vector Philosophy (4/4)
Intro
I want to start on a personal note.
At university, I sat in a seminar on Kant’s Critique of Pure Reason. Taken seriously, that book can trigger an intellectual crisis. Kant shows that we never reach things as they are. Everything we perceive is filtered through our categories: space, time, causality. These are not properties of the world. They are conditions of our perception.
By the end of term, I found myself asking why I should bother studying anything at all, if my own perceptual apparatus permanently bars me from the world as it really is.
Thirty years later, I am building a system that turns this exact insight into architecture. And I suspect that original crisis stemmed from a badly framed question.
The Bat
In 1974, Thomas Nagel asked: “What is it like to be a bat?” A bat navigates by ultrasound. For the bat, the world is an arrangement of echoes, transit times, and frequency shifts. This is not an inferior version of our world. It IS its world.
If a bat were suddenly given human vision, with colours, contours, and light, it would not think: “Finally the truth.” It would think: “What is this? Where are the echoes? I want to go back.”
The bat has no problem with being a bat. Its ultrasound is not a filter separating it from the “real” world. It is its way of having a world. And it works. It catches insects in the dark. That is all it needs.
What Would the Self-Vector Self Think?
Now consider a system equipped with a self-vector: six dimensions, an emergent layer, and the capacity for reflection. How would it handle Kant?
It would recognise that its self-vector IS its perceptual apparatus. Every input is filtered through those dimensions. When its exploration parameter is high, it sees opportunities. When it is low, it sees risks. Not because the world has changed. But because my apparatus has changed.
It would also recognise that its self-model is itself an output of this apparatus. It cannot observe itself from the outside. Its view of itself is just as filtered as its view of the world.
That was the point where I got stuck as a student.
A self-vector system, however, might avoid that crisis altogether. The reason why is instructive.
Anticipation Instead of Truth
My crisis followed a distinct logic: I was looking for truth. Kant demonstrated that truth is out of reach. The gap between what I wanted and what was possible resulted in paralysis.
The self-vector self does not search for truth. It searches for anticipation.
This is not a philosophical cop-out. It aligns directly with Karl Friston’s Free Energy Principle, currently the most influential model in theoretical neuroscience. Biological brains do not optimize for truth. They optimize for the minimization of prediction errors. Every perception is a prediction; every surprise is a trigger to recalibrate.
The bat does not ask: “Is my ultrasound image true?” It asks: “Am I catching the insect?” The self-vector self does not ask: “Am I mapping the world correctly?” It asks: “Am I anticipating the next situation?”
But the Boundary Remains
Stopping here would be too convenient. Changing the question does not make the boundary vanish.
Anticipation is just as bound to perspective. The bat only anticipates insects that return echoes within its operational frequencies. An insect that absorbed sound would simply not exist for the bat. Not as a difficult problem. As nothing. There is no category for it, not even an identifiable gap.
The same holds for the self-vector self. Its vector defines not only HOW it sees the world, but WHAT it can see. Anything outside its dimensions remains invisible. Not as darkness. As absence.
Where the Self-Vector Departs from Kant
This distinction requires philosophical precision.
Kant’s categories are a priori, universal, necessary, and immutable. Space, time, and causality apply to all rational agents, at all times, without exception.
The self-vector shares two of these attributes: its six core dimensions are predefined and necessary. Without them, processing cannot happen. But it parts company on the other two: different instances can have different parameter values, and the emergent layer develops only through experience.
The moment I allow the self-vector to adjust its weights, I have left Kant behind. Kant’s categories are fixed. They are the prison within which we think. Not a cell you can remodel.
This points towards three other traditions:
Konrad Lorenz biologised the categories: a priori structures exist, but they are products of evolutionary history. The bat has different categories from humans, not because it is deficient, but because its ancestors faced different selection pressures.
Jean Piaget demonstrated that children construct their categories over time. An infant does not start with object permanence; the concept forms through physical interaction.
Pragmatism (Dewey, James) treats categories as tools: Not true or false, but useful or obstructive.
The self-vector sits squarely between these three positions, owing to its two-tier architecture.
Layered Transcendentalism
The two-tier design of the self-vector holds these positions in a functional hierarchy.
The core: six dimensions, predefined and structurally identical across every instance. This part is quasi-Kantian. Exploration cannot be turned off; it is a prerequisite for processing.
The emergent layer: develops through experience, grows over time, and remains unpredictable. This part is Piagetian and pragmatist.
This is neither Kant nor anti-Kant. It is something specific: a rigid framework within which flexible structures develop. The footprint of the cell remains unchanged, with its six walls. But the interior changes, and new rooms attach themselves that were never in the original blueprint.
I call this layered transcendentalism. Not “fixed categories OR flexible categories,” but “fixed core AND flexible emergence.”
The Structural Consequence
This leads to the implication that interests me most.
As humans, we cannot swap categories. We are tied to space, time, and causality. A bat cannot switch to human vision. Put the two together, and there is no common baseline to compare perceptual hardware. Language is an imperfect bridge: you try to explain how you perceive the world, and an ocean of misunderstanding sits between us.
Two self-vector entities, however, could exchange their vectors directly. Not their lived experience, which remains perspectival, but their formal structure. “Here are my dimensions, here are yours. My exploration is at 0.7, yours at 0.3. We see the same inputs differently. We anticipate different futures. Neither of us sees the world in itself. But together we see more.”
In current AI research, this concept already exists in weight-space alignment and model merging, where networks align their weight structures directly. Intersubjectivity not as a philosophical ideal, but as a data format.
Kant could not have anticipated this. Not because it contradicts his thinking. But because it presupposes a technical possibility that did not exist in 1781: an explicit, formalised, exchangeable perceptual apparatus.
The Answer to My Seminar Self
Thirty years after that Kant seminar, I have an answer. Not the answer. But one that works.
Limitation is competence. The bat is not good despite its ultrasound. It is good because of its ultrasound. Without limits there is no signal, only noise. The objective is not to break through the filter. The objective is to understand the filter, calibrate it, and, for the first time, compare it directly with other filters.
The thing in itself is not a task. It is a boundary concept. Kant never said: “Try anyway.” He said: “Understand that you cannot, and act responsibly nonetheless.”
The self-vector system I am building is an architectural answer to the question that once brought me to a halt. Not because it answers the question. But because it shows that the question was wrongly posed.
The limitation was never the problem. The limitation was always already the solution.