Never Automate What You Do
[The Druids] are said... to learn by heart a great number of verses; accordingly some remain in the course of training twenty years. Nor do they regard it lawful to commit these to writing, though in almost all other matters, in their public and private transactions, they use Greek characters. That practice they seem to me to have adopted for two reasons; because they neither desire their doctrines to be divulged among the mass of the people, nor those who learn, to devote themselves the less to the efforts of memory, relying on writing; since it generally occurs to most men, that, in their dependence on writing, they relax their diligence in learning thoroughly, and their employment of the memory.1
I was and am surprised that the rise of AI / LLM technology has found its main use case in alleviating work from the thinking class. If robots could alleviate work from the working class, they'd be going for it too, but the thinking class should at least recognize that being highly trained in thoughts or in technical knowledge is the one quality which makes them valuable in an economy. I wouldn't doomsay about the rise of AI and implore white-collars to bravely resist its onslaught, I don't believe that, but I would think that white-collars should respond by trying to enhance their own minds in turn instead of offloading cognitive work to these models.
The human mind remains superior to AI agents in every way but two: quickly performing rote data entry, and being filled with knowledge. That is to say, LLMs start off with all the knowledge of the internet available to query, while humans have to spend decades learning to master even a sliver of a section of a field of knowledge ("accordingly some remain in the course of training twenty years"). But once a human is fully trained he outperforms AI, both at complex lengthy tasks and at innovation. Creativity is not a deterministic process. Many of the "discoveries" made by AIs have been solely due to their ability to quickly perform rote tasks (e.g. protein folding). If mankind seems to fall behind, it is more often due to laziness, conceit, misdirected efforts, or some other human trait than it is to lack of creative talent.
Machine learning is a predictive process, while creativity is an explorative process. Creativity takes a thinker in many directions, not all of which are fruitful, all of which must be tested for correctness.2 This is a qualitative difference which at this point and with this technology, I'm certain cannot be remediated, since creativity is a highly imagistic process; the most powerful minds throughout history have consistently stated that their most significant leaps of genius occurred subconsciously and primarily through considering the shapes of ideas. Intuition, feeling, aesthetic — these things take precedence in the process of creation, even though they're typically unvalued during the process of education. These qualities are specifically those which can't (at this point) be replicated by machine's.
A mathematical demonstration is not a simple juxtaposition of syllogisms, it is syllogisms placed in a certain order, and the order in which these elements are placed is much more important than the elements themselves. If I have the feeling, the intuition, so to speak, of this order, so as to perceive at a glance the reasoning as a whole, I need no longer fear lest I forget one of the elements, for each of them will take its allotted place in the array, and that without any effort of memory on my part.3
If Henri Poincaré can believe the above about mathematical reasoning, then what should we believe regarding the ability of AI to intuit? AIs do not have subconscious minds (or minds at all), so while we're all impressed by their ability to recite deep obscure knowledge about highly specialized fields, or to quickly perform simple tasks with moderate reliability, we should not jump to the conclusion that AI will replace Man in his creative role.
To match an intelligent human, an AI model would have to intake information, understand the meaning of that information and not just act as predictive text, then somehow form an image or feeling from that information, and then probe to where that feeling might lead, which is pure subjectivity and aesthetic sensibility. Even then one of these great leaps of genius is not guaranteed, it merely brings this hypothetical AI to its precipice. Few of the examples of genius through history solved their problems by thought alone, it more often took some movement of the body or motion through space to spur on their sudden bursts of knowledge.4 So creativity is quite possibly5 generated in the interface between the subconscious mind and the body itself, in a pre-rational, biological place dominated by impulses, something forever out of reach of AI systems. In any case it is surely a subconscious process, which is also off-limits to machines for the foreseeable future.
We are then faced with a much more pressing problem. Many people especially in the software field make jokes that "if you don't hire junior engineers, you don't get any senior engineers," since every company wants to hire senior engineers who can also do magic and print money instead of training young engineers. This point which the joke makes is obvious, but the unspoken problem attached to it is that if people no longer work directly with the materials of their field, and instead supervise a swarm of AI agents, then innovation dies. It is also said that "innovation happens on the factory floor": the electrical engineer must build circuits and solder; the mechanical engineer must machine parts; the programmer must manage memory and performance in C; do the integrals by hand, not because they can be done faster, but because doing it manually makes you better at it. Since intuition happens in the subsconcious and perhaps even the body, it isn't enough to simply deal with theories about the material world or build CAD models, the young creative engineer must grapple with the world directly, with the body where possible. No one will reach a senior engineer's level of intuition without interacting intimately with every aspect of a system.
As a corollary to this, I must also stress the imperative that the ambitious engineer study deeply. Interacting with the physical world isn't enough to spur creativity — otherwise blue-collars would be driving innovation — the engineer must reflect on his work and understand every aspect of it. A complete knowledge is the goal.
“If you want to become a chemist,” so said Liebig as I started work in his laboratory, “you will have to ruin your health. If you don’t ruin your health studying, you won’t accomplish anything these days in chemistry.” That was forty years ago; is it still valid today? I loyally followed that advice. For many years four or even three hours of sleep were enough for me. One night spent with my books did not count; only if two or three followed each other did I feel I had earned some merit. In those days I had acquired such a fund of knowledge that my friends considered me more reliable than the Jahresbericht.6
Kekulé read and did, he worked in the lab with Liebig and studied hungrily at night. In our culture we seem to bifurcate people into Those Who Work To Make Money in industry and Those Who Study in academia. It would not be hard or very costly for large companies to bump up their R&D budgets and loosen strict expectations in order to give these truly creative minds more leeway to experiment in the field. The right environment would pay off in less time as people think.
In any case, the individual engineer should (must) take it upon himself to study and work in nontraditional ways. The engineer's only edge against a well-informed tool like an AI agent is in his ability to leverage the subconscious mind for nonlinear discovery. An engineer can feel the shapes of things, and demand that an idea or design be beautiful in addition to being true. The slower methods are less efficient, but they may also enable the engineer, if he stops halfway through and reflects, to peer into the true essence of the system he designs and envision how it could be best.
So I say once again that I'm confused why any intellectual professional would willingly offload core work to AI, thus making himself less knowledgable. Every task done by hand contributes to an intuitive understanding of what's going on. The human mind is the singular tool still greater than AI, but we seem intent on using it less and less. If we instead on-load mental work to the mind with the intent of sharpening it, expanding it, making it more flexible, discovering its pathways, then we will have at our disposal a class of mentats far more powerful than AI.
The Druids derided writing because it ruined memory, but why was memory so important? If you can't recall a piece of information at will, then you don't know it. If you can summon it into your mind at any time, then you do know it and remember it. Thus memory and knowledge are the same thing. If writing was bad for true understanding, then how bad must automation tools be? As regards the one specialized skill in which you make your profession, you should be the one doing it, and getting better at it. Never automate what you do, or else you directly subvert the very thing to which you've dedicated your life.
Julius Caesar in De Bello Gallico Book 6, Chapter 14, translated by W. A. McDevitte and W. S. Bohn in 1869. ↩︎
See how circuitous the process of discovery was for James Watson. ↩︎
Henri Poincaré in L’Invention Mathématique. ↩︎
Many of the thinkers I document made discoveries while climbing into busses, carriages, trains, or otherwise moving swiftly over the land, or moving with the body. ↩︎
Forgive the weasel-word "possibly" but I don't like to make claims like "creativity comes from the matrix which lies between mind and body" before my research is concluded. ↩︎
August Kekulé in his speech to the festival at Berlin City Hall, giving advice to young chemists. ↩︎