이미지: The Decoder
Summary
- The paper argues that when companies replace entry-level jobs with AI, the efficiency gains stay with the individual company, but the cost of eroding expertise is shared across the entire industry
- It points to a "validation tether" problem: catching errors in AI output requires deep domain knowledge, but the entry-level experience that builds that knowledge is disappearing
- The effects of the decline in entry-level hiring that began in 2023 are expected to surface only between 2030 and 2045, once today's new hires should have become experienced professionals
- 논문 저자
- 놀런 러벳(Nolan Lovett), NATO 특수작전대학
- 게재지
- Human Resource Development Review
- 핵심 개념
- 인지 공유지의 비극(tragedy of the cognitive commons)
- 개념 유래
- 1968년 개럿 하딘의 '공유지의 비극' 이론
- 침식 경로 1
- AI가 신입 업무를 직접 대체해 직군 자체가 사라짐
- 침식 경로 2
- AI 보조로 신입이 숙련자 수준 성과에 조기 도달, 실제 학습 과정은 생략됨
- 예상 가시화 시점
- 신입 채용 감소(2023년 시작)의 영향이 2030~2045년 사이 나타날 것
One more cow on the pasture, one more junior hire at the company
In 1968, ecologist Garrett Hardin described a situation in which herders keep adding one more cow each to a shared pasture as the "tragedy of the commons." It benefits each individual herder, but if everyone does it, the pasture itself collapses. In a paper published in the journal Human Resource Development Review, Nolan Lovett of the NATO Special Operations School applies this exact structure to corporate AI adoption. When a company replaces entry-level roles with AI, that company captures 100% of the efficiency gains, while the cost—the erosion of expertise—is shared across the entire industry that draws from the same talent pool.
When entry-level jobs disappear, so do experts
Lovett splits the path of expertise erosion into two channels. The first is straightforward: AI systems simply take over the tasks once assigned to junior employees, eliminating the positions themselves. The second is less visible. Even where entry-level roles survive, if AI-assisted juniors immediately reach a productivity level that once took years of experience to achieve, the trial-and-error and repeated practice needed to build deep domain knowledge get skipped entirely.
This leads to what Lovett calls the "validation tether" problem. Catching domain-specific errors in AI's plausible-looking output requires more than surface-level review—it ultimately requires someone with deep expertise in that field. But the paper points out that the very process that produces such deeply knowledgeable people is disappearing due to AI adoption. People who habitually trust AI answers lose the reflexive habit of questioning them, and along with it, the training environment in which junior employees once learned to question authority and verify facts also vanishes.
The effects won't show up until the 2030s
The experienced professionals active in today's market were trained anywhere from five to twenty years ago. Lovett argues that the impact of the reduction in entry-level hiring that accelerated starting in 2023 won't fully surface until sometime between 2030 and 2045. He calls this the "Human Reserve Paradox": organizations feel secure that they currently have the skilled workforce they need, while the pipeline that would produce the next generation of that skilled workforce has already been cut off. By the time the problem becomes visible, it may be too late to fix, according to the paper's outlook.
Editor's take
What makes this paper interesting isn't new data—it's that it puts a name to something already happening. The decision to cut entry-level hiring is perfectly rational from each individual company's perspective: labor costs go down, short-term productivity goes up. The problem is that no company making that decision factors in whether there will be any senior professionals left in the industry ten years from now. What makes the tragedy of the commons so dangerous is that individual rationality and collective catastrophe hold true at the same time—and AI-driven expertise erosion follows exactly that same structure.
Here's how this plays out on the ground: everyone is impressed when a junior developer uses an AI coding tool to quickly produce senior-level output. But whether that same junior can, years later, explain why the code was written that way or where it might fail is an entirely different question. The judgment that used to be built by manually tracing bugs and learning from repeated failure is now being replaced by the habit of simply approving whatever AI produces. The polish of the output has gone up, but paradoxically, the number of people who actually know how to verify it is going down.
The practical lesson for domestic companies right now isn't "don't cut entry-level hiring"—it's that whatever hiring is cut needs to be matched with a deliberate plan for where verification capability will come from instead. A realistic alternative is a dual-track approach: keep at least some of the tasks now handled by AI as a hands-on training track where people still do the work from start to finish, and give the people who've been through that track the authority to review AI output. Skimping on this cost now means that in five or ten years, only organizations with a broken validation chain will be left.
Warnings like this paper will likely keep coming over the next several years, while the trend of companies cutting entry-level hiring will probably continue unabated—problems whose effects aren't immediate tend to get pushed down the policy priority list. The real signal will likely come after 2030, in the moment when specific industries start hearing simultaneous complaints along the lines of, "This person has five years of experience—why can they only do this much?"
