THE PRICE OF AN INSTANT ANSWER

The Price of an Instant Answer: How AI Changes the Culture of Learning

A culture passes down its knowledge in ways that rarely look like formal instruction. A younger worker watches how a veteran handles a difficult conversation. A designer discovers that the first attractive sketch does not answer the client’s real question. A writer learns to recognize a sentence that sounds persuasive but evades the truth. In each case, the visible product matters less than the judgment formed while making it.

Artificial intelligence changes that process. A prompt can produce an elegant paragraph, a plausible design proposal, a research summary, or a customer response almost immediately. The time saved is real. So is the temptation to remove the slower stages in which novices learn to distinguish good work from convincing imitation. Our cultural problem is no longer simply whether machines can make useful things. It is whether we will preserve the human paths by which people learn why those things are useful.

The disagreement over AI often falls into familiar camps. One side sees a machine that can relieve people of tedious labor. The other sees a device that makes thought mechanical. Both describe important possibilities. Neither tells us what happens to the apprentice when the finished product arrives before the apprentice has developed the capacity to assess it.

THE WORK HIDDEN INSIDE THE WORK

Consider a familiar creative task: drafting the description for an exhibition. The final text may be only a few paragraphs. Yet composing it traditionally requires choosing which details matter, identifying assumptions about the intended audience, deciding what the exhibition can legitimately claim, and seeing where its argument becomes too neat. Each revision can be a small encounter with ambiguity.

A model can now supply a fluent description in seconds. The curator may correctly use it to reduce administrative friction. But suppose a junior colleague receives the same fluent output and is instructed only to check its spelling and tone. The institution has preserved the delivery of exhibition copy while narrowing the opportunity to learn curatorial judgment.

This problem extends well beyond art. In service work, an AI assistant can recommend a response before a trainee has learned why a particular complaint is unusual. In education, a student can receive a polished outline before wrestling with what the question asks. In management, an assistant can produce a performance summary before the supervisor has considered which facts would change the interpretation.

The philosopher Michael Polanyi gave us language for the invisible part of competence. In The Tacit Dimension, he argued that knowledge involves more than what people can state explicitly. A practiced musician hears a small error that a beginner cannot name. An editor senses a weak argument before finding the sentence that exposes it. The book describes a form of knowing carried in trained attention and participation, not merely in rules that can be copied.^[1]

AI creates a new question about this tacit dimension. If the novice rarely performs the work that developed the expert’s eye, where will the next generation of experts come from?

EFFICIENCY IS EVIDENCE, NOT A COMPLETE VERDICT

There are substantial reasons to welcome AI assistance. In a peer-reviewed experiment involving 453 college-educated professionals, Shakked Noy and Whitney Zhang found that access to ChatGPT reduced average time on selected writing tasks by 40 percent and improved rated quality by 18 percent. The study measured those assigned tasks, not a lifetime of learning or the cultural effects of removing early practice.^[2]

Another study by Erik Brynjolfsson, Danielle Li, and Lindsey Raymond followed the introduction of an AI assistant to customer support agents. The researchers reported a particularly large productivity gain among less experienced and less skilled agents and found suggestive evidence that the assistant helped transmit effective practices. That result complicates any simple story of inevitable deskilling. Under some conditions, AI can help novices learn faster.^[3]

These findings concern different tasks and settings. Together, they suggest a more useful question than whether AI makes us smarter or lazier: what kind of practice does the tool provide, and which capacities can workers develop while using it? A system that explains a recommendation, invites correction, and exposes a difficult exception can become part of an apprenticeship. A system that produces acceptable answers and rewards employees for clicking approve can hollow the apprenticeship out.

The distinction matters for institutions whose work depends on cultural memory. Libraries, universities, newsrooms, cultural organizations, and businesses need experienced people capable of assessing novelty, recovering context, and refusing a smooth but incorrect answer. Their expertise may look expensive when measured as minutes per document. It becomes invaluable when a routine case turns out to have a hidden exception.

WHAT THE OLD TOOLS TEACH US

People have worried before that a new technology would erase a vital human capacity. Calculators displaced many everyday calculations. Search engines made locating information easier. Photography transformed the meaning of visual representation. In each transition, some practices became less common, while new forms of expertise grew around the tool.

The historical analogy is encouraging only up to a point. A calculator returns an answer to a defined operation. A search engine offers sources that a reader must select and interpret. Generative AI can supply an explanation, a judgment, and a plausible rationale together. That makes it possible to delegate parts of the decision process that older tools left visibly in human hands.

We may need less practice at producing the first draft of routine work. We will need more practice at framing the right question, recognizing missing perspectives, interpreting evidence, and determining when the familiar answer is wrong. Those capabilities do not appear automatically when a tool improves. They must be cultivated deliberately, with time and permission to make mistakes.

In my consulting work on AI adoption, the most productive conversations often begin with a practical question: which decisions are genuinely routine, and which exceptions require a person who understands the underlying process? The answer varies by occupation and organization. The important step is to make that distinction explicit before rewarding everyone for increasing the volume of machine-assisted output.

AN APPRENTICESHIP FOR THE MACHINE AGE

A practical response begins by separating the task of producing work from the task of learning how work should be judged. The two can overlap, but they should not be treated as interchangeable.

A museum preparing an exhibition label might ask a junior curator to identify the source of three claims, explain what would make a proposed description misleading, and defend one change to the AI draft. A newsroom could require early-career reporters to establish what is known and unknown before asking a model for structure. A service team could ask new staff members to compare the model’s recommended response with a deliberately unusual case, then explain why they would accept or reject the recommendation.

These exercises take time, and they can seem inefficient compared with immediate publication or closure of a case. That is the point. Training has always involved an investment that looks unproductive when measured only against today’s output. Institutions gain something in return: people who can handle the unexpected tomorrow.

Leaders should preserve room for work done without AI as well. A novice who occasionally writes a difficult paragraph from scratch, troubleshoots a case without automatic suggestions, or reviews competing accounts before seeing a machine summary develops a different relationship to evidence. This is not a ritual of technological purity. It is a way to maintain the underlying skills required to verify a fast answer.

The evaluation of AI adoption must then include learning outcomes. Did the worker become better at spotting omissions? Could a junior colleague explain the decision when the system was unavailable? Did an employee identify an exception the assistant missed? Are people moving toward greater independence or becoming more dependent on an interface they cannot question?

A culture that measures only how much labor its tools eliminate may gradually lose the ability to create the knowledge those tools summarize. A culture that measures how people grow while using them can turn automation into an opportunity to develop new kinds of expertise.

THE HUMAN HANDOFF

The phrase “human in the loop” often promises that a person will approve the machine’s work. That promise is weak if the person has neither sufficient context nor a learned basis for disagreement. Clicking yes is an action. Exercising judgment is a capacity.

The central challenge of the AI era is therefore cultural. We must design environments in which the novice can still become an expert and the expert still knows when to challenge the machine. Tools can accelerate the production of answers. Communities decide whether there will be people capable of asking better questions, recognizing false certainty, and taking responsibility for the results.

The next generation does not need to repeat every step of the previous generation’s labor. It does need the chance to acquire the understanding that made that labor meaningful. The cost of an instant answer should never be a future in which no one knows how to disagree with it.