The Ascending Skill Barrier

In the AI Story

The Trivandrum training provides the empirical pattern. Segal observed that the more capable the person, the more robust the output they extracted from the AI. Entry-level engineers produced entry-level output amplified in volume. Senior engineers produced architecturally sound systems reflecting decades of accumulated judgment about what works, what breaks, and what matters. This is not what a simple democratization narrative would predict. If AI merely removed the skill barrier, outputs should be roughly equivalent regardless of creator background. The correlation between pre-existing judgment and AI-assisted output quality reveals the nature of the ascending barrier.

The educational implications of the thesis are the subject Shirky has engaged most directly in his work as NYU's Vice Provost. His proposed medieval turn — toward in-class assessment, oral examination, real-time demonstration of knowledge — is an ascending response to the unreliability of take-home essays in the AI era. The lower-level assessment has been rendered unreliable; the higher-level assessment ascends to a level the technology cannot reach because it requires the student to be present, embodied, and responsive in ways no tool can simulate.

The surgical analogy Segal develops illuminates the physical reality of the ascending pattern. When laparoscopic surgery replaced open surgery for many procedures, surgeons lost the tactile feedback of hands in the body cavity — the ability to feel the difference between healthy and diseased tissue. The loss was real. But laparoscopic surgery made possible operations that open surgery could never attempt. The surgeon operating laparoscopically is not doing easier work; she is doing different work, harder at a higher level, demanding forms of cognitive integration that open surgery did not require.

The consequences for the distribution of value in the second cognitive surplus are significant. Applications where the ascending barrier is low — personal utilities with clear requirements and limited scope — will be produced in enormous quantities by enormous numbers of people. These are the experimental substrate, the lolcats of the second surplus. Applications where the ascending barrier is high — tools that serve broad communities, platforms requiring sophisticated architectural judgment, systems that handle sensitive data or make consequential decisions — will be produced by people who possess the evaluative skill and domain expertise to direct AI tools toward genuinely valuable outcomes.

Origin

The thesis is developed in conversation with Segal's ascending friction thesis in The Orange Pill and with Shirky's own engagement with the educational consequences of AI. The underlying pattern — that abstractions in computing ascend rather than eliminate difficulty — has been observed by many commentators over several decades, but the specific application to the AI transition requires attention to the particular character of the skill being relocated.

Key Ideas

Relocation, not elimination. AI tools do not remove the skill barrier; they move it upward from implementation to judgment.

Asymmetric learnability. Lower-level skills are learnable through structured curricula; higher-level judgment is harder to teach, measure, and transmit.

The judgment-domain correlation. AI-assisted output quality correlates strongly with pre-existing judgment, revealing that the ascending barrier tracks domain-specific evaluative capacity.

The distribution of value. Applications requiring high judgment will be produced by fewer people with deeper expertise; applications requiring low judgment will be produced by many people experimentally.

The educational implication. Curricula must shift from transmitting lower-level skills that AI replicates to developing higher-level judgment that AI cannot provide.

Debates & Critiques

The principal debate concerns whether the ascending pattern is genuinely structural or whether it merely describes a transient period during which higher-level judgment has not yet been automated. The response developed here is that the higher-level judgment involves tacit knowledge, domain expertise, and evaluative capacity that are harder to formalize than the lower-level implementation skills, and that while some aspects of higher-level judgment may eventually be automated, the pattern of ascent — from what is automated to what is not — will persist across subsequent waves of automation. A second debate concerns whether the ascending barrier is inherently less democratic than the lower-level barrier it replaces; the evidence from Trivandrum suggests that people with deep domain knowledge can develop the higher-level judgment even without traditional credentials, but that the development requires conditions not all populations have access to.