A device that produces philosophy
Finishing Kant’s project and turning the mind into a machine
Perhaps one of the most difficult imaginable challenges in Artificial Intelligence research is to design an AI that could independently derive its own philosophical perspective. Such a task introduces two major hurdles: (1) the AI must be able to somehow build up philosophical concepts autonomously across ethics, metaphysics, etc., and (2) the AI must understand and explain its own outlook. Does the AI in question believe that a priori synthetic knowledge is possible, or that it’s an illusion? Is the AI a materialist or an idealist? Why or why not?
Besides using Large Language Models to ape what others have already written, there isn’t anything close to a theory for how to accomplish this. It is so far beyond current research paradigms that most people who hear the request merely shrug, slightly annoyed that the question was even asked; as though you asked them to pack the cure to all diseases into one pill. Contemporary AI agents are designed to address clearly-defined tasks with accompanying criteria for success: pick up this object, predict this text, maximize this metric, etc. How could you train an AI to perform tasks whose criteria have not even been defined? How can an AI create a philosophical perspective when we don’t even know what the “right” philosophical perspective is?
These are not meant as rhetorical questions. The goal of this post is to discuss a solution.
Let’s go back to the two challenges described in the first paragraph. It doesn’t take much probing to realize that the two are in fact the same task. The self-reflexive work of understanding your own philosophical perspective is identical to coming up with it in the first place. You build your own philosophy through education, self-observation, and a focused analysis of the material that both give you. You are not creating a set of philosophical entities in your mind, then “seeing” them; you are building them at the same time that you are looking for them. Just as inventing a recipe for a casserole is identical to describing that recipe, an understanding of your own philosophy is the same as the presence of that philosophy in your mind. Both are a set of beliefs about the world. So far so good. The question now is: how does this happen?
Everything you believe about phenomenology, ethics, materialism, reality, meaning, etc. comes from your introspective “eye”. This inner “eye” can look inside your mind, understand what it sees, combine it with ideas gained from education, and form beliefs about its own metaphysical or spiritual status. This faculty, then, is the subject of the discussion; what it learns and believes eventually becomes your philosophical perspective.
Few people have delved into this faculty as deeply and competently as Immanuel Kant. To guide this post, therefore, we will start by describing how Kant understood this process, then show its limitations and gaps. At the end of the post we’ll have built a more complete and effective model of philosophical belief, one that has a greater chance at being automated.
Kant was the first metaphysician to try find the walls and functional limits of the human understanding. Before his time, the mind was discussed as though it were essentially divine and limitless, with unfettered access to truth. It was only brought low by its lazy, sinful nature. Kant, on the other hand, began to find “mechanical”, rather than moral, limits on thinking. For example, he argued that you can’t know anything about reality outside of the sensory phenomena to which you have access.
What Kant didn’t do, and couldn’t do, however was to predict exactly what you will think, when, and why — to define the mind deterministically. Your faculty of understanding was still free to spontaneously form its own beliefs about the world, even within the implacable walls of its phenomenological prison. Thus Kant was able to carve out a space for free will even within the mechanical limitations of the brain. Whatever the mysterious force was that drove your mind to create beliefs about the world, he left that to one side.
Kant made a simplification which was necessary to his project. He split cognition into roughly three layers:
- The content that the mind understands (e.g. representations, intuitions, appearances),
- The inner “eye” that understands this content and creates beliefs about it (e.g. judgment, synthesis, analysis, aperception), and
- The motives that drive the whole machinery (mostly ignored).
By segregating these three, he was able to cordon off a person’s desires and motivations from the more intellectual, cognitive side of the mind. He assumed that desires didn’t meaningfully affect his inquiry and could be ignored.
Quarantining desires allowed Kant to redefine the role of the self, or the consciousness, within intellectual cognition. The self was only a construction of its functional parts. In fact, the use of the term “inner eye” throughout this post is a bit misleading, as it implies that Kant believed there was a conscious observer in the Cartesian theatre of the mind. This is untrue. The “eye”, or understanding, was merely the compiler of information into new beliefs or thoughts. Thus he realized his “Copernican revolution” in metaphysics by pushing the ego out of the spotlight, and leaving only a machine of sorts; a project that Hume had begun earlier.
This left Kant with the apparently easier question of how the content of thoughts was connected to the machinery of understanding. For example, how do you formulate a concept of chair so that you can then assert that “all chairs have four legs”?
His approach was to reduce everything intellectual to either content of some sort, or the act of manipulating that content. The goal of the Critique of Pure Reason was in fact to bring metaphysical, non-empirical concepts, like time and the infinite, into the machinery itself as content. He was intrigued by the fact that philosophers were able to explore metaphysical topics seemingly without empirical foundations to rest them on. His thinking went something like this:
- You can create beliefs about metaphysical things, like time, cause, and existence, which don’t come from experience.
- They must come from somewhere.
- Therefore they must be given to the understanding from inside the mind itself.
Of course, he assumed that forming beliefs about metaphysical concepts involves forming beliefs about some “content”, which therefore had to be given to the mind as raw material. The first step was therefore to turn them into objects of the understanding (concepts). He asked himself: “how could a person decide that time was infinite unless they had already represented time and infinity to themselves somehow?”
Besides intuition there is no other kind of cognition than through concepts. Thus the cognition of every, at least human, understanding is a cognition through concepts.
The understanding, namely, demands first that something be given (at least in the concept) in order to be able to determine it in a certain way.
It is never given to us to observe our own mind with any other intuition than that of our inner sense — Kant, Critique of Pure Reason
The introspective “eye” was then like a cook that needed to be handed its ingredients so it can analyze and integrate them. The content had to be presented in a form that it can cook with, generally as an intuition or representation, or some other input to the understanding. So when Kant was faced with the question of explaining how the mind could conceptualize and build theories about space — e.g. that “space has three dimensions” — he was forced to convert space into an input of some kind, as a representation, or a “pure intuition”. This he did using what he called rules and schemas, a topic we won’t dig into here.
Kant’s perspective, incidentally, was in line with 18th century scientific thinking that was in vogue in his time. He wanted truth itself to give him his beliefs, and for the mind to become an impartial storage device and compiler of that knowledge. What psychological “format” these representations of time, mind, or phenomena would need to take on so they could be processed was unknown and irrelevant to his project; as long as they came pre-packaged, abstracting away any details about their origin.
He didn’t entertain the possibility that each representation or concept might be unique, both in how it functions in your mind and in its individual properties. If this happened, of course, the impartiality of his “understanding-machine” would break down, since it would now be making a special case for every new investigation being conducted. When comparing the following beliefs, for example:
- Food gives me energy
- Love gives me energy
Kant didn’t want the understanding to give special attention or treat the synthesis of either belief differently, at least when it came to the mechanism of synthesis. They were both only syntactic cogs in a logical machine. The “eye”, for Kant, remained a dispassionate observer, trying out hypotheses and formulations with whatever substance it was given.
His model of human understanding has since become the model for how modern AI treats declarative knowledge (explicit facts that you learn, as opposed to skills). As nodes in a knowledge graph, all declarative concepts are assumed to be identical in their essential form, and only differ in their properties and relationships to other nodes:
Declarative memory is a long-term store for facts and concepts. It is structured as a persistent graph of symbolic relations — Laird, etc al., A Standard Model of the Mind
Making all concepts homogeneous makes it easier for software to process them using the same operations for all, with none getting special treatment. How these representations came to be in that format is relegated to a separate process of sensory perception or intuition. The brain somehow learns to transform image pixels into a node labelled Harrison Ford. The entities that the understanding works with must arrive prepared in working memory ready to fulfill their tasks. Their connection to the real world is allowed to remain unclear and vague:
This schematism of our understanding with regard to appearances and their mere form is a hidden art in the depths of the human soul, whose true operations we can divine from nature and lay unveiled before our eyes only with difficulty. — Kant, Critique
But unlike the rules of grammar, which can be applied independent of content, meaning cannot. Meaning is by definition motivated, subjective, and preferential. Concepts that only connect to one another in a knowledge graph have no grounding in reality. It is what you can do with them, and, as we’ll see, how they satisfy your goals that gives each of them its meaning. By trying to make the understanding impartial, Kant’s model implicitly erased or ignored the motives that compelled the “eye” to create its beliefs. He never bothered with the simple question of why: why did the mind decide, at a given time, to form a particular belief? Why now? Why this? Did it just happen, with no discernible reason behind it?
Nietzsche highlighted this oversight at length, by describing how the will and personal motives coloured one’s judgments. Yet neither he nor Kant could explain the mechanism by which the motive for engaging in an inquiry influences the resulting beliefs. And is does influence them. Just as a person who is looking for a fight actually creates one in the act of looking, the abstract concepts that the understanding perceives — that is, the ingredients that the cook uses — are generated at the same time that you begin to look for them.
For example, a mind that is looking for something to slake its hunger gradually creates concepts like food, taste, even hunger itself (as a thing it should avoid). It also creates useful beliefs about the world, e.g. apples can be eaten. Your subjective attempt to understand your environment is identical with the act of creating those concepts. Food gets attached to everything that solves the problem of hunger, no matter how odd it may look (e.g. intravenous drips, pixie sticks, grasshoppers). And like food, the very existence and meaning of concepts like mind, causality, and even philosophy are the consequences of trying to interact usefully with life’s experiences. Their properties are not “lying there” to be picked up as probabilistic patterns in data; they are projected out of the desires of the searching “eye”.
For Kant such a suggestion would have been heretical, since it meant the ostensibly mechanical “eye” would bring a bias or prejudice to the inquiry, and thus fail to be impartial. As an analogy, a criminal investigator that desperately wanted to find the perpetrator of a crime may be tempted to project that propensity into every dark corner, even to misplace it onto a bystander. And yet… an investigator that didn’t want to find the culprit, perhaps didn’t even care to see that a crime had occurred would be useless to the cause. The motives you bring necessarily influence what you see, even during your most stringent attempts to be objective¹. They focus your attention and shape every interpretation of what you experience. It requires maturity and self-restraint to ensure you don’t, like a child, accidentally believe your own delusions of what isn’t reasonably there.
There are no simple, atomic concepts to feed into Kant’s “understanding-machine”. Every act of perceiving a concept involves a selective act of interpretation. When someone points and asks you — “what is that?” your answer necessarily connects the direct experience of that stimuli to something else, like a word-sound (“chair”) or a visual thought (an image of a crime being committed). This is the same process by which you create beliefs about experiences such as “the chair is brown”. Both are interpretations. So the abstraction layer in Kant’s model that was key to separating perception from reasoning is illusory. The process that brings concepts into the mind is the same process that determines the beliefs you have about the world. A “concept” is, in essence, a heap of beliefs.
By trying to separate the concepts from the act of reasoning, and turning the former into symbolic variables, Kant removed the top-down meaning of the content, which made it impossible to effectively reason about. You could no longer perform any logical operations on such concepts; even something as simple as finding contradictions isn’t possible unless you know the intent and meaning behind how the variables should be interpreted, e.g.:
- My head is part of my body
- My nose is on my head, not on my body
Without a top-down interpretation of what you mean when you say “body”, there is no way to resolve the above contradiction. The process by which conceptual content is generated (perception) is inseparable from how it is interpreted and reasoned with.
This point is worth restating more than once. Kant hoped that representing such things as time as abstract concepts could hide the messy details of how you originally came to perceive them, in the same way a cook doesn’t care to know how apples or zucchini grow from seed. But concepts like time and cause aren’t actually present as entities or “objects” in the mind — Kant himself said as much. If time is truly the shape of inner experience itself, as he proposed, then how can you judge if it goes in one direction, or even what “one direction” means for time? According to Kant, it happens by analogy to an object moving in space. That implies you were able to compare your representation of time with a completely different a priori concept, space. And that means you must be able to access something behind both abstractions which, according to Kant, presumably doesn’t exist. Kant fully recognized that he had not answered all such questions, but he didn’t realize that answering these questions would cause him to rethink the assumptions of his entire system.
Kant may have felt, perhaps, that he didn’t have many options available to him. During his time philosophers of mind assumed there were two sources of information from which you formed your beliefs: (1) from experience, and (2) from something built-in or structural within the mind. But there is a third alternative that was rarely considered: that they are imaginary. “Imaginary” doesn’t mean inaccurate or fantastical, rather it means artificial or fabricated. You can invent an understanding out of thin air — such as Bohr’s invention of the atomic model, or superstring theory — then apply it to reality, and still find it somehow useful. There is nothing in your sensory experiences of everyday objects themselves that forces you to think everything is made of superstrings. This is a “fantastical” thought, which only escapes being derided as lunacy because it is useful in many contexts.
It helps if we go back to the example of the criminal investigator. There is no “crime” or “perpetrator” actually present in the sensory experiences themselves. That is an assignment, an interpretation that the investigator projects onto his experiences or his thoughts of those experiences. There are two sides to this interaction: the first is what he actually sees and thinks (the concrete sensory content), and the second is what he came looking for. The first provides the opportunity, the second provides the meaning. Critically, the meaning of the concept comes from the latter², from the fact that the person came looking for a criminal. Criminal is an invention, a tool you use to subjectively label people, to help you avoid or restrain their harmful behaviour.
Neither your motives nor the sensory evidence by themselves directly cause this belief. Even the most sloppy investigator doesn’t just assign “criminal” to any sensory input he comes across. There must be a connection between the specific content and the meaning ascribed to it. It must, in his experience, be useful to interpret a certain set of visual or auditory inputs as a “crime”, and another set as the “perpetrator”. He will label a person a “criminal” because he wants others to arrest and imprison them, to prevent greater harm. Doing so to an innocent bystander will not achieve his goals. The glue for the connection, then, is always outside the concepts themselves, in what he wants to happen. So to see where any concept or belief comes from you should look, not to the material itself, but to the meaningful implications it has for you.
In summary, it is not that criminals commit crimes that hurt us, but rather that people who hurt us can be addressed by calling them criminals. The order of effects is reversed: the effects that a certain experience has on your motives and desires causes you to define its source under a useful concept. The same can be said about any concept, abstract or concrete. Food doesn’t satisfy hunger; food is a concept invented for anything which satisfies hunger.
One counter-intuitive consequence of the above is that the beliefs you form about a concept already exist before you identify the concept, because your beliefs are the precursors that ultimately give rise to the concept. For example, you can feed yourself various fruits, vegetables, etc., and over time form beliefs about which of those experiences lead to satiation, all before you explicitly conceptualize food. You could not have conceptualized food without first having learned how to use it effectively.
Though all this seems straightforward when it comes to food or criminal, it becomes more interesting when applied to the topic with which we began this post: philosophy. We can now look at how everything we’ve discussed so far can be applied to how you create your philosophical beliefs.
The philosophical concepts and beliefs you adopt are always tightly entangled with the motives you bring to the activity. Just as the need to live without fear of harm forced you to invent your notion of crime, and the desire to satiate hunger created the concept of food, the desire to escape suffering creates concepts like nirvana and samsara. Similarly, your desire to blame and punish people for their actions — and receive praise for your own — makes you invent a concept of free will. (See footnote 3 for links to other examples). Such motives and others are what ultimately embed the concepts in your psyche and oblige them to appear — you have little choice in this. The unexpected appearance of a concern, and its eventual resolution will create that space of meaning whether you intended it to or not.
It is not possible to understand what any of the above concepts mean without without having gone looking for something like them. The concept of phenomenon, for example, arises in your mind when you find your own understanding is misaligned with the “objective” or scientific one. Why do you see the colour red when it’s objectively clear these are only wavelengths of light? The need to resolve this “embarrassing” inconsistency between beliefs creates a space of half-truths we call phenomena. It is noteworthy that the word “phenomena” only came into use after Locke proposed that everything you see, hear, and feel is actually distinct from reality itself.
The same applies to the question of how you create philosophical beliefs; they can be reduced useful interpretations of your experiences. For example, consider what happens when you decide to interpret your experiences within a materialist framework. Your beliefs about the material world are a multifaceted fiction you have invented to achieve certain goals, such as the need to secure scarce objects (toys, cake) for yourself, or to predictably obtain and hold onto things you like. This is the root of a set of related notions including substance, matter, and persistence. It is no coincidence that the term “materialist” refers both to a person who believes the world is entirely physical, and also a person who enjoys getting things for themselves.⁴
You don’t actually need to have a direct intuition of matter or substance given to you by your mind to become a materialist; and even if you did, it would not explain where your beliefs about it came from. You can simply add the interpretation of “matter” and all its concomitant thoughts to any experience, as a sort of useful fabrication. This interpretation may come in various forms, such as a word, an image, set of images like a movie, etc. For example, you can see a 3D object in a video game and treat it as though it were matter, by assuming it will remain where you leave it. Being a materialist per se is merely the explicit expression of all of these motives as being the ones you value the most.
Building your philosophical outlook is ultimately based in a variety of fundamental motives, including your need to give yourself stability (metaphysics), to address emergent challenges in social interactions (ethics), to alleviate the causes of psychological suffering (spirituality), to avoid embarrassment due to public errors (logic and ontology), etc. All philosophical beliefs and concepts are useful inventions or interpretations derived from these. They are invented wholesale — like a story, or a detective’s hypothesis — then filtered for utility. They aren’t, as Kant assumed, extracted logically from your intuitions, as something “given” to you.
Kant had projected his simplified model of concepts onto the mind in order to then describe how to synthesize new beliefs from them. The understanding then became a machine, with a perfunctory modicum of “spontaneity” to serve as free will. If he truly wanted to complete his work, however, it would have been necessary to include the deterministic influence that motives have on reasoning. This was the only way to explain how metaphysical concepts and beliefs can arise in a mind.
¹ Objectivity, incidentally is a motive too, and it gives birth to concepts such as reliability, predictability, neutrality, and their associated beliefs.
² This is something of a simplification, since meaning is also indirectly influenced by the patterns of the sensory experience itself.
³ You can read about other examples of this inverted process of concept creation with respect to time, number, self and the world, classification, consciousness, dog and game, and also a discussion of the general case.
⁴ This is not to imply that materialists are all greedy, only that the childhood origin of the concept of the material world is rooted in this need for security and persistence in one’s experiences.
