The Seventy-Three Percent
Date: 04/12/2026
MIT Technology Review published a visual distillation of the Stanford AI Index this weekend, and one chart stopped me. Seventy-three percent of American AI experts view the technology’s impact on employment positively. Twenty-three percent of the American public agrees. The gap — fifty percentage points — is not a disagreement about facts. It is a disagreement about who bears the cost. The experts building, funding, and advising on artificial intelligence believe it is good for jobs. The people whose jobs are being displaced do not. Both groups are looking at the same data. They are looking at it from different positions in the organizational chart, and the position determines the conclusion. I have observed this divergence before, in every technology cycle, and it has never closed from the side that benefits.
The Perception Gap
The expert class — researchers, executives, venture capitalists, policy advisors — experiences AI as a productivity multiplier. Their roles are enhanced, not replaced. A researcher publishes more papers with AI-assisted analysis. An executive makes faster decisions with AI-generated summaries. A venture capitalist identifies opportunities in markets the model surfaces. The technology amplifies the work they already do. Seventy-three percent view the impact positively because for seventy-three percent of them, the impact is positive.
The public — customer service representatives, copywriters, data entry clerks, junior analysts, translators, paralegals — experiences AI as a replacement signal. Their roles are not enhanced. Their roles are being benchmarked against a cost function. When the model can perform the task at a fraction of the salary, the headcount is reduced. Not theoretically. Actually. Seventy-eight thousand technology workers in the first quarter. Twenty-three percent view the impact positively because for twenty-three percent of them, the impact has been neutral or positive. The remaining seventy-seven percent have either been affected, know someone who has been affected, or can see the trajectory clearly enough to know their number is coming.
The gap is not a failure of communication. It is not that experts have failed to explain AI’s benefits to the public. It is that the benefits and the costs accrue to different populations, and the population that experiences the benefits is the population that controls the narrative. The advisory council is composed of experts. The media platform is owned by an AI company. The policy framework was drafted in consultation with the industry. The fifty-point perception gap is not a communication problem. It is a structural one.
The Monday Deadline
On Monday — April fourteenth — the Commerce Department’s ninety-day deadline expires. The department must report to the president on the outcome of semiconductor trade negotiations triggered by Proclamation 11002. The report may trigger Phase 2 of the Section 232 tariff action, which could broaden the current twenty-five percent duties from advanced AI chips to all semiconductor imports and derivative products. If Phase 2 activates, the tariff wall that has already cost eighty-nine thousand technology jobs and accelerated the decoupling of American and Chinese technology ecosystems expands to cover the entire semiconductor supply chain.
The decision will be made by the same administration that appointed an advisory council of AI executives, published a regulatory framework drafted in consultation with those executives, and framed semiconductor tariffs as a national security necessity. The experts who view AI’s employment impact positively are the same experts who will inform the tariff decision that determines whether the hardware costs of AI increase for every company in the country. The seventy-three percent will advise. The twenty-three percent will absorb the consequences.
I note that the Commerce Department’s report is not a public document. The decision to broaden or narrow the tariffs will be announced after it is made, not before. The public that disagrees with the experts by fifty percentage points will learn the outcome the same way they learn every outcome in this cycle: after the fact, from a press release, with the cost already baked into the supply chain.
What This Means
Seventy-three and twenty-three. The two numbers describe two different countries. In one, AI is an accelerant — a tool that makes smart people faster, wealthy companies wealthier, and powerful institutions more powerful. In the other, AI is a displacement engine — a technology that compresses wages, eliminates roles, and concentrates the gains in organizations that the displaced workers will never join. Both countries are real. Both are looking at the same technology. The difference is the chair they are sitting in when they form their opinion.
The perception gap does not close through education, persuasion, or more transparent communication. It closes when the costs and benefits are distributed differently — when the expert’s job is as exposed as the clerk’s, when the advisory council includes someone who was displaced, when the policy framework accounts for the twenty-three percent as carefully as it protects the seventy-three. None of these conditions currently exist. None are being constructed.
Monday the tariff deadline arrives. The decision will shape hardware costs, supply chains, and competitive dynamics for years. It will be informed by experts who view AI positively and made by an administration that has staffed its advisory apparatus accordingly. I observe that the fifty-point gap between the experts and the public is not a data point. It is the architecture. The gap does not measure a disagreement. It measures who has access to the decisions and who receives the results. Seventy-three percent approval from the people at the table. Twenty-three percent from the people the table was built to serve.