"The unconscious is structured like a language." — Jacques Lacan
Introduction
In my Chinese Room Fallacy paper, I argued that our definition of 'understanding' reflects an anthropocentric bias—defining it in terms of what sounds intuitive; 'perceiving', 'interpretation of meaning' and 'to understand', all of which is somewhat circular and falls apart under logical scrutiny. I defined understanding as the capacity to memorise foundational axioms, apply them to novel contexts, and build upon them through replication, recombination, and adaptive generalisation across hierarchical levels of abstraction. Understanding is deeply tied to our linguistic and cognitive framework, meaning that it is system-relative.
A useful thought: Bees construct perfect hexagonal combs—not because they understand Euclidean geometry, but because their evolved cognitive-behavioural system encodes an implicit structure that reliably results in this output, which we call a hexagon because of our specific meaning-structure.
Conscious Rationalisation
I don't think we can say that consciousness is required for understanding when 99.99% of our actions, beliefs, and biases, are all subconscious.
Another useful thought—If you walk to a door to exit your house, but you are not consciously aware of the entire process of walking to the door to exit, do you not understand what you are doing? Perhaps absurd to suggest so. To do so, it requires a level of abstraction and meaning-assignment to different actions (by yourself), and objects around you (environment modelling). Walking to a door requires a bunch of different sub-goals, but you subconsciously perform this action. You might be aware of the what you're doing. The what is more of the post hoc rationalisation of the how. So the biomechanics, the balance adjustments, the sub-goals that lead you to do the higher order outputs. But the actual mechanism of walking to the door, you're not consciously aware of.
Most people confuse understanding with reflective understanding — the highest-order, conscious, introspective form of cognition. But in reality, understanding operates at multiple levels of abstraction, most of which are non-reflective and subconscious.
When you speak, you're not—it's weird because you're not consciously aware of the words coming out of your mouth. You can speak really fast and say whatever you want to say, but for some reason, you're not consciously aware of each word being said. The words just come out of your mouth. Even when you try really hard to be consciously aware of each word you're saying, it's very difficult to do so. The words emerge probabilistically and subconsciously. Do you not understand what you're saying? No, of course you understand what you're saying. What you're doing is making a post hoc rationalisation of the words you speak—a post hoc conscious rationalisation of the words you say—and you attribute that to understanding when in reality, the real understanding was the subconscious abstraction and probabilistic outputs that gave rise to your saying the words.
The semantic processing—the actual mapping of meaning, context, relevance—happens in systems we can't consciously access. That's where the real understanding lives. But we mistake our after-the-fact conscious reflection on what we've said for the understanding itself. Conscious awareness = a tiny subset of understanding.
The Spectrum of Abstraction
Understanding exists on a spectrum of abstraction, ranging from low-level procedural knowledge to high-level reflective cognition. At the lower end of this spectrum lies procedural, implicit understanding—the kind that governs how we navigate through doorways, move through physical space, use language fluently without consciously thinking about grammar, or anticipate a friend's mood from subtle body language cues. We don't consciously think these processes through, yet we understand them functionally in the most practical sense.
At the higher end of the spectrum exists what we might call reflective, explicit understanding, which is deeply tied to human linguistics and meaning-structures. This includes grasping complex concepts like Gödel's incompleteness theorem, reflecting on moral dilemmas, or engaging in debates about consciousness itself. This level of understanding feels like what we call "true" understanding precisely because it's verbalizable and introspective—we can articulate it, examine it, and communicate it to others.
However, this high-level understanding that we privilege in our conception of cognition is actually built on mountains of prior unconscious representations. The sophisticated semantic processing, pattern recognition, and contextual modelling that make explicit reasoning possible all operate below the threshold of conscious awareness, suggesting that what we consider the pinnacle of understanding may actually be a narrow slice of a much broader cognitive phenomenon.
The Bottom Line
AI may not have the high-level post-hoc "feeling" of understanding — but that doesn't mean it lacks it altogether. Just like a bee understands how to construct a hexagon — it doesn't necessarily philosophise about it.