I have noticed an interesting pattern in business and government: we only acknowledge a problem once it comes into focus. And when it does, we treat it as new and novel, as though information gained value simply by being recent. But when looking to the future, the past has far more value than we care to admit. The "modern" problems of AI aren't actually new; they are simply the latest chapter in a very old book where we can finally apply what we have learnt. We've been chewing over the "how" and the "why" of machine intelligence for centuries, building a foundation of first principles that many people are simply not aware of. Consider the timeline:
- 16th Century: The Golem of Prague enters our folklore. It is the first real debate on machine intelligence: a powerful, man-made entity created to serve, but one that lacks a soul or moral compass. It was a warning that a tool without strict parameters eventually becomes a threat to its creator. You can hear the same fear, centuries later, in Mary Shelley's Frankenstein.
- 1840: Ada Lovelace publishes what many consider the first computer program. She didn't build a tool; she mapped out the logical steps (loops, if-else statements, memory management) for a machine that didn't even exist yet, working out the logic of the problem long before the hardware caught up.
- 1942: Isaac Asimov sets out the first real manual for AI ethics, his Three Laws of Robotics, later collected in "I, Robot". A scifi author, before the computer even existed, was already working out how to keep an "autonomous" mind aligned with human safety.
Why does this matter now? Because you do yourself a disservice by ignoring the thinking others have already done, work laid down long before the AI mania of the past few years could cloud it. These pioneers weren't focused on the "machine"; they were focused on the logical problem space, and on how to reason about intelligence without being swept up in the model, breakthrough or framework of the week.
To bring the conversation into the here and now, two old ideas capture the trap we're in. The first is Douglas Adams' supercomputer **Deep Thought**, built to calculate the "Answer to Life, the Universe, and Everything." After 7.5 million years it returns "42," and when its creators object, it tells them the fault is theirs: they never actually knew what the question was. The second is the **Infinite Monkey Theorem**: enough monkeys at enough typewriters will eventually produce the complete works of Shakespeare, buried somewhere in an ocean of gibberish.
The two don't really conflict. They're the two halves of one problem. Deep Thought has every resource and no question worth asking; it fails at the front, on **definition**. The monkeys generate every possible answer and can't tell which page is Shakespeare; they fail at the back, on **judgment**. One can't frame the problem. The other can't recognise a good answer when it stumbles onto one.
That's exactly where we are with AI, because today's machine is both at once: Deep Thought when you can frame the question, the monkeys when you can't. It'll answer anything you can define and generate something plausible for everything you can't, and it won't tell you which is which. The disillusionment people feel isn't really with the machine. It's that we've skipped both jobs that were always ours: defining the problem space, and knowing what "good" looks like when it comes back. We ask vague questions of a probabilistic engine, we can't tell "42" from the answer or Shakespeare from the garbage sitting beside it, and so we blame the box.
You can watch this play out in how most people actually use AI. It gets pointed at surface-level chores: "Fix this spreadsheet," "Summarize this email," "Research this topic." Useful work, but barely scratching what the tool can do. We treat it as a magic box, when it is closer to an **amplifier**: it takes whatever thinking you bring to it and scales it up, for better and for worse.
So if you aren't sharpening the mind behind the machine (the same logical mind Ada Lovelace was using in 1840), you'll be replaced. Not by AI, but by a human who has learned to use it as a lever. In a world where AI executes tasks faster than you ever could, where does your value lie? Many suggest moving "up" to become an "orchestrator," but they rarely explain the how.
The "how" is the shift from being an Artisan of your craft to being the Operator of a machine. The Industrial Revolution made the same move, from manual craft to machine operation. The machine performed the labor, but the operator still had to know what "good" looked like. An operator who didn't understand the parameters of a high-quality textile simply produced high-speed trash.
That is the future we are quietly building toward. By "top-loading" companies with experienced staff while neglecting the junior pipeline, we are destroying the training ground where people learn what "good" actually looks like. If you have never spent time in the "low-level execution" trenches, you lack the context to judge the machine's output. You can't be a manager if you don't understand the work you are managing.
And you can't hand that understanding to the machine, because it hasn't got any. Current AI is really the world's most sophisticated **autocorrect**. It predicts the best next word, but it doesn't understand your intent, your company's culture, or the wider context of your industry. It is a Golem—it will do exactly what you tell it to, even when what you told it is a mistake.
That leaves you with judgment. Execution is becoming cheap. Judgment is not. It shows up in the questions you learn to ask:
- "If I execute this task this way, what does it mean for the wider system?"
- "When the AI hands something back, can I tell whether it's actually any good?"
If you can't engage with the problem space at that level, you aren't "managing" AI at all. You're just gambling that it knows what it's doing, and we already know it doesn't. It amplifies both your strengths and your failures, so if you don't know where your blind spots are, you can safely assume it's amplifying those too.
To the Prompter go the spoils. AI isn't coming for your job, but the person who can frame the problem space and knows what "good" looks like when they see it is. The race is on.
· · ·
Share