For most of human history, the person who knew things had something the person who did not know things needed. The knowing was scarce. The text existed in one place. The expertise resided in one person.
Access to the text and to the person required the permission, the proximity, or the institutional membership that made the access possible.
The university held the books. The guild held the techniques. The profession held the certified knowledge. The church held the interpretive authority. Each of these was, among other things, a management structure for information scarcity—a way of organising access to what was known and controlling who could claim to know it.
The scarcity was real before it was managed. The management became the institution. The institution became the authority. The authority became the credential. The credential became the thing people were paying for when they paid for education, even after the original scarcity that justified the institution had begun to erode.
The erosion is now structural. The text that once resided in one library is available on a device in a pocket. The expertise that once resided in one practitioner is available as a query to a system trained on the accumulated output of thousands of practitioners. The calculation that once required a specialised mathematician is available in seconds to anyone who can type the question. The information bottleneck that education has historically managed is not narrowing. It is collapsing.
What collapses with it is not knowledge. What collapses is the institutional premium that was attached to mere possession of knowledge in a context where possession was scarce. The university that charged for access to texts and lectures was providing something genuinely valuable in a context where texts and lectures were otherwise inaccessible. The school that certified reading and calculation was providing something genuinely valuable in a context where many adults could not read or calculate. As access to both has widened—first through mass literacy, then through public libraries, then through the internet, now through AI systems that can produce fluent explanation of almost any topic on demand—the specific institutional premium that was attached to access has declined.
The institutions have not declined with it. They have persisted, partly because the credential they issue retains social and economic value even as the scarcity that justified the credential has eroded, and partly because the functions they serve beyond information transmission—the childcare infrastructure, the social sorting mechanism, the cultural production of shared experience—remain valuable and have no obvious replacement. The credential is partly a proxy for the scarcity that no longer exists and partly a proxy for the socialisation and sorting that the institution performs regardless of the information it transmits.
This is where the disruption is more interesting than the simple observation that AI can answer questions. The more disruptive possibility is that AI may force education systems to reveal that much of what they sold as education was, in significant part, the management of an information bottleneck. If the bottleneck weakens, the institution is required to justify itself by what it does that the bottleneck management was masking.
The progression from information to knowledge to judgement to wisdom is worth tracing precisely because it maps onto the gradient of what the bottleneck collapse actually affects. Information—what is known—is what AI now provides with considerable fluency and increasing accuracy. The question that could be answered by an encyclopaedia, by a textbook, by a search engine, can now be answered by a conversational system that synthesises the relevant information into a coherent response. The information bottleneck at this level is largely gone.
Knowledge—what is understood, the organisation of information into a structure that allows inference and application—is more variable. AI systems can demonstrate knowledge in the sense of producing outputs that reflect the organisation of information, can explain the relationship between concepts, can apply known principles to described situations. Whether this constitutes understanding in any meaningful sense is a question about the nature of understanding rather than about the capability of the system. What it does is make the appearance of knowledge available to anyone who can query the system, which changes the value of demonstrating knowledge through institutional certification.
Judgement—what matters here, in this specific situation with these specific conditions and these specific stakes—is where the bottleneck collapse encounters its first significant limit. Judgement is not the retrieval of information or the application of known principles to described situations. It is the capacity to assess which principles are relevant, which features of the situation are significant, what the likely consequences of different responses are, and what weight to give competing considerations in conditions of genuine uncertainty.
AI systems can scaffold judgement—can present relevant considerations, can model likely outcomes, can surface analogous cases—but the assessment of what matters here requires the kind of situated knowledge that comes from having been in situations, having experienced consequences, having developed the particular sensitivity to the features of situations that repeated engagement with them produces.
Wisdom—what should be done now, given the consequences—is further still from what the information bottleneck managed. Wisdom is judgement operating across time and with full acknowledgment of its own limitations. The wise practitioner is not the one who knows most but the one who knows what they do not know, who can assess the reliability of their own judgement, who has accumulated enough experience of being wrong to have calibrated their confidence appropriately. This cannot be transmitted as information. It can only be developed through experience and shaped through relationship with people who have more of it.
Taste, in this framework, is something closer to judgement than to mere preference. The distinction matters because taste is often dismissed as subjective and therefore as unsuited to educational transmission. But taste, in the sense of the developed capacity to assess the quality of work in a domain—to know when a sentence is doing what it needs to do, when a structure serves the content or constrains it, when a design is solving the right problem or the wrong one elegantly—is not arbitrary preference. It is evaluative pattern recognition shaped by extensive encounter with the domain’s best and worst work, refined through the kind of attention that distinguishes what is working from what is merely present.
The person who has read extensively and carefully in a domain has developed a sensitivity to its qualities that the person who has read little has not. The sensitivity is not innate. It was produced by the reading. It is also not fully articulable—the person with taste can often recognise the quality before they can explain it, can sense that something is wrong before they can name the specific failure. The recognition precedes the analysis. This is what the mentor transmits when they teach someone to write—not the rules but the sensitivity, the capacity to feel the difference between the sentence that is almost right and the sentence that is right, which the rules can approach but cannot capture.
This transmission does not occur between a person and an AI system, not because AI lacks access to the work or the capacity to generate explanations, but because the transmission is relational. The student’s sensitivity is developed partly through encounter with the work and partly through encounter with someone whose sensitivity they can test their own against. The mentor’s response to the student’s work is not information about the work. It is a calibration event—the student’s assessment is placed against the mentor’s, and the gap between them is the thing being reduced.
The discernment that this process develops is different from the information retrieval that the educational system has historically certified and different from the pattern generation that AI systems can now scale. Discernment is the capacity to assess—to determine what is relevant, what is reliable, what is valuable, what the situation actually requires. In a high-information environment, where the volume of available information exceeds any individual’s capacity to process it and where AI systems can generate plausible-seeming outputs on any topic regardless of their accuracy, discernment may be the most consequential educational outcome available.
The student who can assess the reliability of a source, who can identify the assumptions embedded in an argument, who can distinguish between a well-supported claim and a confidently stated one, who can recognise when a framework is being applied to a case it was not designed for—this student has a capacity that is more valuable in an AI-saturated information environment than the capacity to retrieve or reproduce information that the system can provide on demand.
Developing discernment is not the same as teaching critical thinking as a generic skill. It is the development, through specific practice in specific domains, of the sensitivity to quality and reliability that distinguishes the experienced practitioner from the novice. It requires encounter with enough cases to calibrate the judgement, relationship with enough practitioners to test the calibration, and the experience of being wrong often enough to know where the calibration is unreliable.
The educational institution that justifies itself as formation of judgement, development of discernment, relational apprenticeship, and adaptive human capability is describing something genuinely different from the institution that justified itself as managing the information bottleneck. Whether the institution that exists can become the institution the description requires, or whether the description requires a different kind of institution, is a question the current arrangement has not yet been forced to answer directly. The collapse of the bottleneck is forcing the question. The answer is not yet available. The question is worth asking precisely.
The Vorpel Foundation