Intellect requires artifice, and therefore labour. Innate genius can neither be explained nor imparted. This artifice seems autonomous because it arrives removed from its singular, inimitable point of origin. Artifice identifies precisely those aspects of an art subject to be explained, documented and transferred to others. Consequently, intellectual work participates in the history of automation, affecting all trades at some point of its development.
In this book, Dennis Yi Tenen challenges what we mean by "intelligence" and "autonomy" in robotics and AI. The book traces the historical evolution of automation, showing that intellectual work has always been entangled with technological progress.
History gives meaning to such inherited structures. In following the evolution of technology, we are able to make sense of legacy technological debt. Without history, the present becomes invariable, sliding into orthodoxy (it is what it is). History's milestones mark a path into possible futures.
Kircher Versus Kuhlmann on the Nature of Machines
Tenen frames this tension through two 17th-century figures with opposing views on what machines are for.
Kuhlmann believed the path of knowledge should stay difficult, accessible only to those willing to walk it properly. For him, intelligence lives in the human mind, not the machine.
A machine is a tool, a means to an end, and its value is set by the skill and understanding of whoever operates it.
Kircher saw technology differently, as something meant to serve, making difficult knowledge accessible to a wider audience.
Their disagreement plays out as a split between a private, inward experience of understanding and an outward, instrumental effect on the world. Kuhlmann's combinatorial poetry aimed inward, toward reconfiguring the spirit, making new internal connections. Kircher's calculating machines and mechanical poetry aimed outward, toward being useful to others.
This splits between two models of intelligence:
- Platonic: intelligence is the correct internal alignment of thought and feeling with universal truth. It is necessarily private and local and it lives in someone.
- Aristotelian: intelligence is the goal of thought, not the process of achieving it. It's a universal capacity to produce specific results, and it can be borrowed, distributed, or outsourced.
Under the Platonic model, you can only tell if someone understands something by questioning them poignantly, exposing the limits of what they grasp. There's no shortcut to knowledge where change happens through mental toil. This view is inherently human-centered: it assumes the point of learning is personal transformation.
Under the Aristotelian model, the point is not how you arrived at the answer but whether you have it. When we learn, we reach for whatever gets us there, be it books, search engine or other people. Refusing outside help and insisting on total isolation is counterproductive: the goal is the result, not proof of your own unaided capacity.
Why this still matters for AI
This is where the book's two threads meet. Even after the invention of the calculator, we still insist on teaching arithmetic by hand, because the struggle itself carries value independent of the result. Tenen's implicit question for AI is whether we're prepared to make the same argument for writing, coding, or thinking with a language model, or whether we've already adopted the Aristotelian view: that the outcome is what counts, and the labor behind it can stay invisible.
The hand carries the load of value through lived experience. Whether that's still true once the hand can delegate almost everything is the real question this book leaves open.
Where I find myself in this
As someone who uses language models and writes code alongside them, the tension Tenen describes plays out in real time as a daily negotiation with myself.
When I ask a model to explain a concept I half understand, am I learning or am I outsourcing the struggle that would have made it stick? When I generate code that works but I could not have written from memory, do I understand what I have built?
The Aristotelian perspective would argue: this works, time to ship it. The Kuhlmann perspective would say: this is a borrowed result without concrete internalisation leading to clarity.
Perhaps this is not an either-or. The distinction is not between using AI or not using AI but between using it as a scaffold versus using it as a substitute. A scaffold holds you up while you build the structure yourself, and is removed once it stands. A substitute replaces you entirely, leaving nothing behind when it is removed.
In other words, AI is used to accelerate parts of learning where brute-force exposure is the bottleneck. This includes tasks like reading documentation, exploring solution spaces and generating examples for testing. It still matters to reconstruct the reasoning personally afterwards. The labour still carries value, but the path to that labour no longer needs to be slow. AI can compress the journey without eliminating it, if only we insist on doing the reconstruction step.
The danger is not the tool but the temptation to skip that step because the output already looks complete.
In the realm of AI specifically, this becomes recursive, like Ouroboros. We use AI to learn about AI, use language models to understand how language models work. The snake can consume itself, each cycle hollowing out understanding until nothing remains but fluency without foundation. Or it can renew itself, each cycle compressing what would have taken months into days, as long as the reconstruction still happens by hand.
I can get the perfect answer and prototype rapidly, but until I have sat with the implementation and failed a couple of times, I risk holding borrowed knowledge that I cannot extend, debug, or teach. But there is a pragmatic tension here too. In a world that rewards speed, the person who delivers with borrowed understanding gets promoted, while the person who insists on deep clarity but moves slowly gets overlooked. The middle path is to do both in sequence, ship first, then circle back and earn what you shipped. Delivery buys you credibility and time; reconstruction turns borrowed knowledge into something you own and can extend. Neither the fraud nor the martyr wins long-term. The fraud eventually ships something they cannot defend. The martyr eventually has nothing to show. But the person who delivers and then reconstructs compounds in both directions, building reputation and understanding simultaneously. You are not choosing between speed and depth. You are choosing the order.
Tenen's book helped me articulate something I had been feeling strongly: that the real risk of AI in learning might not be ignorance but false fluency, the sensation of understanding without the underlying structure to support it. The Platonic model warns us that this kind of knowledge is fragile. It passes the Aristotelian test, but is it sustainable?
Reflecting on Hidden Ingenuity
Every tool I use is inherited labour made invisible, centuries of insight compressed into something I invoke in a line of code. The honest position is neither cynicism nor blind embrace. It is to use these compressions with the awareness that they are not replacements for understanding, and to keep building your own.