Note: This article is written without AI assistance and the opinions expressed in it are my own, not those of my employer.
Generative AI cannot replace you. Yes, it can write a novel. Yes, it can code with less fatigue and more alacrity than you will ever muster. But let’s zoom out for a moment and consider what generative AI actually does. Generative AI predicts tokens. As a review, a token is a unit of data that models analyze or produce. You can think of 1 token as ¾ of a word in text. Generative AI models, given input, predict the next tokens they should write based on their training.
Generative AI cannot generate its input. That’s our role. We train it, we test it, we train it some more, and even then, once the model is trained, we context engineer it, we harness it, we write it guidelines and make skills to fence in our agents. As the CEO of Take-Two noted, AI is essentially derivative. It cannot make truly new things. Given this derivative nature, I’d like to explore what the “thought” of AI might be, and how that can inform our adaptation to it. In the next few paragraphs, I’ll discuss briefly what thinking itself might be, a recent scientific finding that complicates matters in terms of the advent of generative AI, and dwell for a bit on the immense and imminent changes generative AI brings.
Does AI think? Can we eventually model the brain’s tasks of both input-gathering and output creation? And what is thinking in the first place? Unfortunately for our psychological comfort, if you ask philosophers, that question is up for energetic debate.
Does human thought occur solely on the material plane, as Democritus and Hobbes and Daniel Dennett averred? Much of modern behavioral psychology and neuroscience is based on this idea. At one point in my early twenties, I explored getting a degree in neuropsychology. I told the professor in charge of the program that I was interested in where the brain ends and the mind begins, and in response I was told never to ask that question again, and that no serious neuropsychologist would consider that the mind begins or ends anywhere else but in the material of the brain.
Or is there something… else afoot? Is everything in the world immaterial, like Berkely thought? (To which I would say: if everything is immaterial, why does the world suck?) Christian Scientists famously espouse this worldview, often relying on prayer rather than doctors in times of illness.
Are there two completely separate planes, as in Descartes’ philosophy? If you’re having trouble imagining what the implications of this are, The Matrix is a pretty close approximation of this type of world.
Or is there something else happening, something requiring a living being with senses, something more integrated between spiritual and physical planes, as Aristotle and Aquinas suggested? There are a few emergent branches of science that could potentially support this, for example, I’d wonder what proponents of the gut-brain axis would have to say, but there are more scientists having problems directly quantifying the brain. As an aside: what a wonderful problem to have. Being able to think about ourselves, engaging in omphaloskepsis as a population, and doing this from the perspectives of our various branches of science, philosophy, and theology, has got to be one of the most interesting problems out there. In a way, I’m grateful to the growth of AI for stimulating my own ponderings here.
Andrew Knight, a physicist, recently published a paper arguing that human thought cannot be replicated with classical computing. To be frank, this conclusion could be taken to support any of the theories listed above (with the exception to materialist theories that are purely computational or along the stimulus-response lines), but let’s zero in on the argument he presents for a sec. This is what he concludes:
“By calculating the minimum information required to specify distinguishable conscious experiences across a human lifetime and comparing this to established estimates of the brain’s information capacity, we can determine whether a digital computational model of the brain is viable.”
By taking the historicity of the mind into account, that is, the fact that consciousness by necessity arises from sensory memory, he comes to an astonishing conclusion. It’s wild. I plucked this paper off the internet using Elicit, so I barely have any social context, but I imagine the scientific debate around this paper has been furious. I’m summarizing it to the best of my ability, and I do hope I’m not walking into turbulent academic waters, but I’ve got to include the paper in this blog post because it’s so fascinating. Are you ready?
Andrew Knight found that the sum of the parts of the brain is greater than the whole.
Basically, given that consciousness requires historical input, which blows up the calculations a bit, he quantified each sensory modality in bits. Visual, auditory, tactile, olfactory, gustatory, and integrative modalities were included. Then, he compared it to the brain’s current theoretical storage capacity, which is 2.8 × 10^15 bits. The sum of the storage capacity for each sensory modality was 9.46×10^15 bits, much larger. He spelled out three possible conclusions:
One. Consciousness is not entirely produced by the brain;
Two. The brain utilizes non-classical computational mechanisms that transcend traditional information theory constraints; or
Three. Current estimates of neural information capacity are fundamentally inadequate.
If conclusion #1 is true, the materialists could be wrong. They could also be right in a way, but needing some expansion to the brain’s interaction with the senses and the body, e.g., the gut-brain axis I mentioned before. Does AI have a gut-brain axis informing its thinking? Or five senses? Well, nope. Clearly something different going on here. I would be curious to see what the field of robotics has to add.
If conclusion #2 is true, any of the philosophers could be right, and AI cannot replace you, because to model the whole brain, we need quantum computers to be a lot better than they are.
If conclusion #3 is true, well, last I heard, we don’t have enough compute power to model the human brain anyway and this moves the goalpost. AGI is far off.
Given all this, how do we understand AI’s impact on us? Token prediction and generation is far off from our own process of reading and writing. Humans in general learn to read starting with phonemic awareness, which is why auditory and speech difficulties lead to reading difficulties. We might take deaf students as an exception to that particular process, however, the principle remains that we begin reading with our senses. Hearing students don’t break words into tokens, they break them into phonemes, either auditorily, or kinesthetically in the case of deaf students. Then the phonemes are reconstituted to words, which are reconstituted into sentences, which are in turn reconstituted into meaning. There is no “meaning” for the AI agent, and when it comes to AI-generated work, we are the arbiters of meaning.
What does it mean to be an arbiter of meaning? And does it even matter when tens of thousands of us have been laid off, being told by CEOs that AI has replaced us?
The brutal and tragic truth of the matter is that while AI cannot replace us, it can take our jobs. AI seems to have created a good-enough simulacrum of our output to set the capitalist world on a frenetic edge. At the same time, AI does best at acts that don’t involve reasoning, but rather, highly patterned creating, like coding. A function written in code isn’t true or false. Code is strange that way, but it is the realm AI excels in. However, there are so many other realms we excel in, moving from mental room to mental room in our daily work life.
In order to adapt to whatever AI world we’re growing into, we need to get good at critical thinking, or the kind of problem-solving that can be applied to any difficulty, no matter its accidental properties. A good critical thinker is a person that can be dropped into a school, a hospital, a tech startup, and start listening carefully and asking the right questions after gathering the pertinent information.
Another skill we need to grow in is cognitive sovereignty. As far as I can tell, this term was coined by The Artificiality Institute, which has a lot of great resources on this, but basically, it means not outsourcing control of problem-solving to AI. The good news is that there are emotional signs that you’re doing this, and if you learn to monitor yourself, you can prevent yourself from becoming a slop cannon. I have an article on this topic. The technical bits may or may not be outdated at this point, but the portions on self-regulation still apply.
I do want to acknowledge that this is tough. AI is changing how many of us work very quickly. It’s 2026; how we ache for some precedented times! The only comfort I can offer is that while ignoring AI completely may not go well, you can slow down. As DHH himself says, there is not a limited sale on progress. You can take breaks. You can be patient with yourself.
“We should like to skip the intermediate stages. We are impatient of being on the way to something unknown, something new. And yet it is the law of all progress that it is made by passing through some stages of instability and that it may take a very long time.”
— Pierre Teilhard de Chardin