A Quinnipiac poll released this week found that 55 percent of Americans say artificial intelligence is likely to do more harm than good in their day-to-day lives. That is up 11 percentage points from the same poll a year ago. Employee anxiety about job loss due to AI has jumped from 28 percent to 40 percent over roughly the same period, according to Mercer's annual talent survey. The direction of travel is clear: public trust in AI is falling, and falling quickly.
What is less clear is what, exactly, people are afraid of. A Bloomberg opinion piece from last week, titled "AI Washing Is Masking an Insidious Labor Crisis," made the pointed observation that evidence of AI actually destroying jobs remains surprisingly thin. Companies are announcing AI strategies, cutting headcount, and then attributing productivity gains to AI, but comprehensive data on the mechanism, on how many specific roles have been eliminated and replaced by specific AI systems, is "patchy at best." The job losses are real. The AI causation is often assumed, not demonstrated.
This creates a strange situation where public anxiety has outpaced the measurable evidence. People fear AI displacement in the abstract while the actual mechanism of displacement remains difficult to trace in the aggregate statistics. Part of this is lag: the survey respondents who said they were afraid last year were right that something bad was coming, even if the timing and shape of it was uncertain. Part of it is that labour market disruption is genuinely underway in ways that don't yet show up cleanly in national employment figures.
Meanwhile, CNBC reported this month on a counterintuitive winner from the AI boom: skilled tradespeople. Between 2022 and 2026, demand for robotic technicians grew by 107 percent. HVAC engineers and industrial automation technicians are seeing strong wage growth, with six-figure salaries now achievable in sectors that were considered blue-collar dead ends a decade ago. The buildout of AI data centres requires enormous quantities of physical infrastructure, cooling systems, electrical engineering, and construction work that no language model can perform remotely.
The divergence between the professional jobs assumed to be safe (knowledge work, law, finance, writing) and the trades assumed to be vulnerable (physical labour, manual assembly) is reversing in real time, which is disorienting for anyone who followed the conventional narrative. Web designers and administrative assistants are at higher risk than HVAC technicians. The disruption is targeting the middle of the knowledge economy, not its physical base.
What ties these threads together is uncertainty. The public is correctly sensing that something large and hard to control is happening to the economy. The Quinnipiac poll is not a measure of AI's actual impact; it is a measure of how that uncertainty is being experienced. Bloomberg's scepticism about job loss numbers is warranted, but it is also somewhat beside the point: the anxiety is real regardless of whether the data currently supports it at scale.
The more useful question is whether the fear is pointing in the right direction. Job loss anxiety focuses on specific roles and specific people. The deeper structural question, about what happens to an economy where the coupling between labour, income, and consumption is systematically weakened across large swaths of the workforce, is harder to poll about. It is also the question that matters most.