contextual judgment – A translator who worked for years between Irish and English at the European Union says AI-driven machine translation has already cut his income by about 70%, while 43% of translators report similar drops. The episode is a warning that the knowledge economy is
For Timothy McKeon, the shift wasn’t theoretical—it showed up in his pay. After years translating to and from Irish for the European Union. he watched machine translation improve and the market start paying less for the kind of work he had long relied on. He says his income fell by roughly 70% as his EU work dried up.
“The more it learns, the more obsolete you become,” McKeon told CNN.
His experience isn’t isolated. The text of his warning is echoed by another figure: 43% of translators have seen their incomes drop thanks to the increasing presence of AI alternatives in the marketplace.
This is the early cut of a bigger change. For decades. much of the value in white-collar work was built on a simple bargain: if you could do the knowing—memorize the tax code. marshal case law. pull market data. speak a language fluently—you could be paid to retrieve it and turn it into results. Now. a frontier large language model can read more tax code. more case law. and more market reports than any individual ever could. and return it on demand “fluently and instantly.”.
Even the idea that AI might mostly be harmless because it sometimes “hallucinates”—filling answers with errors—has weakened. Hallucinations are becoming increasingly rare, and in many contexts they can be mitigated through effective prompting. Access to reliable LLMs isn’t quite free or frictionless. but when compared to human labor. the cost is becoming negligible.
As a result, the center of gravity is shifting. In an increasing number of fields, a chatbot can deliver work that’s close to, or in some cases better than, that of an average professional. The broad base of competent-but-unremarkable cognitive work is being priced downward toward zero.
It’s tempting to assume the damage stops with “average” talent—that deep expertise is safer. This notion is only half right. Translators are an example of a migration toward higher stakes: the volume jobs have gone to the machine. while literary translators and high-stakes legal and diplomatic interpreters—the people whose errors carry real consequences—still find their phones ringing. Specialists look safe “for now. ” but the line between what AI can take and what it cannot isn’t where most people assume it is.
The comforting refuge of depth has limits because recorded knowledge is still knowledge a machine can drill into. The obscure corner of tax law may be rare to a human, but to an LLM it’s just another corner—available so long as it exists in a recorded form.
That puts the question in sharper focus: the value of knowing won’t hold just because it’s rare. It will hold because of what kind of knowing it is.
Two kinds of knowing emerge as more durable than simple stored facts. The first is contextual judgment. A seasoned consultant’s value isn’t only the industry detail in her head; it’s knowing which detail matters for a specific client or board. how a background fact should guide reading a problematic balance sheet. or how to interpret the half-articulated fear a CEO mentions in passing. Contextual judgment depends on cues that aren’t in the record because the moment itself hasn’t happened before in quite the same way. The decisive cue is something fleeting—read from the room—and current models are far less reliable at that kind of inference than at recorded-knowledge reasoning. It may eventually become more accessible, but it isn’t the main threat professionals are facing today.
The second is procedural knowledge. captured by the difference between “knowing that” and “knowing how.” You can know every proposition in every physics textbook and still be unable to keep your balance on a bicycle. You can absorb everything written about music theory and still not be able to play the violin. In business. the parallel is familiar: reading every book on negotiation doesn’t automatically translate into the ability to hold your nerve. time a
concession. and keep your footing when the other side pushes. That kind of knowing lives in doing—acquired through practice and experience—and at higher levels it’s bound up with trust. authority. and the ability to read and relate to other humans. It exists between people. which means it can’t be handed off as a stock of facts and it can’t be offloaded without turning into a bottleneck of the very kind professionals are trying to remove.
Neither contextual judgment nor procedural knowledge can be downloaded. Both can be built, deliberately—and that’s where career development is being pulled now.
Three moves are proposed for getting on the right side of the shift. First, own outcomes, not outputs. An AI model produces outputs: a draft, an analysis, an answer. The guidance is to audit what you’re actually paid for—your core value proposition—and strip away what a good model can do in minutes. What remains should be outcomes only you can deliver: the messy problem carried from initial diagnosis through to a result you can stand behind. or the insight into what the client really needs that goes beyond what they said.
Second, build judgment in the room, not on the page. Situation-specific judgment can only be picked up firsthand by being present for consequential decisions and watching how they turn out. The people who advance fastest won’t be those who store the most information. but those who improve their contextually grounded judgment.
Third, delegate the routine; protect the practice. Since procedural know-how lives in doing, the work you hand entirely to AI is work you stop getting better at. The recommendation is to push genuinely rote tasks to the model but keep doing high-skill work yourself—negotiation. argument. and the hard thinking—because even if a model can turn out a passable version faster. convenience purchased now can steal capability later.
McKeon’s verdict—“The more it learns. the more obsolete you become”—lands hardest on the types of knowledge most professionals built their careers around for decades. But the argument here is that other forms of knowing are less vulnerable. Some may even be impervious to AI, at least in the forms available today.
That “survivable” knowledge can’t be downloaded as a transferable package. It’s knowledge embodied rather than possessed—earned in the doing, carried in the person, and owned in a way a stock of facts never was.
AI and jobs translators machine translation European Union large language models hallucinations contextual judgment procedural knowledge career development