AI Efficiency Could Cost Us the Next Generation of Experts

In an IEEE Spectrum essay, systems engineer Richard Mitchell argues that AI-driven efficiency gains may degrade the pipeline that produces experienced engineers. Mitchell, who led controls design for a full digital-control system for a U.S. nuclear plant that was never built, writes that his team deliberately left manual steps inside sequences the system could execute on its own so operators would stay current on the plant's behavior. He says automation only returns control when it is confused or in trouble.

Mitchell cites a Harvard University working paper covering about 65 million workers at more than 280,000 U.S. firms, which found junior employment fell roughly 9 percent within six quarters after companies adopted generative AI relative to nonadopters, while senior employment continued to grow. He also cites a Stanford analysis of ADP payroll records finding that the youngest workers in the most AI-exposed occupations lost ground after late 2022 while more experienced colleagues held theirs, with losses concentrated where AI automates work rather than augments it. He notes the causal story is contested, writing that New York Fed researchers attribute much of the rise in young-graduate unemployment to remote work rather than AI.

Drawing on aviation, Mitchell points to the automation paradox and to Air France flight 447, which fell into the Atlantic in 2009 after iced-over airspeed sensors fed the autopilot bad data and it disconnected. He describes the subsequent failure as a competence failure. He notes the FAA issued Safety Alert for Operators 17007, "Manual Flight Operations Proficiency," in 2017, and that some airlines amended procedures to encourage hand-flying during initial climb and descent in benign conditions, trading some fuel efficiency to keep flying skills alive.

Mitchell proposes deliberate "manual gates": chosen points in a workflow where a human takes control specifically to preserve a skill. He gives an example in which an engineer must reproduce a defect, trace it to root cause, and write an automated test with an AI assistant switched off before the model proposes fixes. He acknowledges manual gates are less efficient in the near term and writes that the math works best for founders with control, private companies, long-horizon institutions, or regulators.