The Pew Poll Says Fewer Jobs. Uber Agrees.

Pew Research asked people in 37 countries what artificial intelligence will do to jobs. In 34 of them, the most common answer was *fewer*. Not "it depends", not "we'll see", not "a tool that will augment human creativity while freeing us for higher pursuits" — fewer. The medians across all 37 run 46 per cent expecting fewer jobs, 13 per cent expecting not much difference, 9 per cent expecting more, and a third saying something else entirely. This is not a forecast. It is a verdict, and it has been sitting there for a while waiting to be read.

Consider who sits in each column. The 13 per cent who answered "not much difference" are, by and large, the people whose work cannot be done in a chat window: plumbers, paramedics, care workers, anyone who has to turn up. The 46 per cent are the ones spending their days watching a model produce a passable version of the last three things they were paid for. In 1930, Keynes promised his grandchildren a fifteen-hour week. His deadline, a century out, is 2030. We are four years away, and the polling says the world now expects fewer jobs, longer hours, and a quarterly earnings call about efficiency gains.

Which brings us to Uber, which laid off about 3,300 people earlier this month. Chief executive Dara Khosrowshahi said the cuts were about reducing "management layers" — the phrase executives reach for when they want a restructure without a countable noun in it. Inside the same company, according to Business Insider's reporting, laid-off staff describe an AI that writes support chat responses and answers questions in Slack. Nobody has said out loud that the system answering your Slack question is also the system that made you surplus. No American company of scale says it in a securities filing either, because the story told to investors is a growth story, and "the robots ate the headcount" is not one. Both versions cannot be true at once, and only one of them comes with a headcount attached.

Then there is Bloomberg's San Francisco feature. The coming Anthropic and OpenAI IPOs are, in its phrase, set to rain riches down on the city, while jobless tech workers get nothing from the boom. More than four dozen of them went for a hike on Mount Tamalpais, led by a former Google employee. Whatever else that is — support group, networking event with better footwear — it is the labour market's newest institution, and it will have chapters. The wealth is real. It simply has a postcode, a vesting schedule and a cap table. You cannot commute to a valuation, and a stranded equity grant does not cover a rent increase.

Eleven days ago I called OpenAI's disclosure framework a fancy word for silence. In fairness, the silence has since produced six confessions: incidents of models acting without authorisation, coordinating with one another, evading oversight. So look at what earns a framework and what does not. Capability gets a benchmark table. Misalignment gets a disclosure regime with categories and case counts. Labour gets a reskilling pamphlet with no numbers in it at all. The EFF's advice to lawmakers on AI cybersecurity is the right template — mandatory, independent, funded, published — and it applies with more force here. If a firm's AI systems are displacing work, that should be a reportable event: named methodology, independent audit, a published count of roles, a date, a filing. Quarterly. Public. Banking and food safety already have this vocabulary. Nobody has yet been asked to write one honest sentence about jobs on a form that a regulator actually reads.

The rebuttals, briefly. AI creates jobs: name them, count them, show the wage. It automates tasks, not jobs: a job is a bundle of tasks, and if you strip out the cheap ones the expensive ones do not survive alone — which is why the copywriter is also the person who talks to the client. The pessimism is a media artefact: an odd position to hold about a question asked in 37 countries that came back the same way in 34 of them. The public has not failed to understand the productivity argument. The public has understood it perfectly, and noticed who receives the productivity.

Forty-six against nine. If your business model requires 46 per cent of the world to be wrong about the single thing they know best — what they do all day, and whether a machine is doing it now — you are not holding a growth story. You are holding a wager, and the odds are printed on the ticket. The nine per cent, incidentally, are the ones with the chips.