What the expectation measures

Pew Research Center surveyed 42,151 people in 36 countries from 8 February to 13 May 2026. In 34 of 37 country results, people expect AI to produce fewer rather than more jobs over the next 20 years. In the United States, the only country where Pew asked this question before, that share rose by 7 points in two years. In many places about a fifth of adults say they do not know the effect. The survey records an expectation. It supplies no headcount table.[1]

In wealthy examples such as Australia, South Korea and the United States, around seven-in-ten adults or more expect job losses. Pew ties the sharper worry to places with higher GDP per capita. The survey does not show which task vanished, whose contract was not renewed, or who was credited with the wage.[1]

Where posting language writes the score

In the same 24 hours a preprint accepted at AIES 2026 audited Llama 3.2, Mistral, Gemma 3, Qwen 3, Phi 3 and DeepSeek-R1 in hiring simulations. The authors report that agentic posting language lowered recommendation scores for female candidates at r_rb = 0.309 (Bonferroni p = 7×10⁻⁵), and 0.448 with model fixed effects. Coded-exclusion language produced effects of 0.646–0.758 against non-White candidates. A label-ablation run isolates the explicit demographic persona label as the main driver.[2]

The two figures do not measure the same object. Pew scans an expectation on the street; the preprint scans the shift between a job post’s wording and a model’s score. The productivity claim does not appear here as a deleted headcount. What appears is a score that decides who reaches a shortlist, left to posting language and the model. The authors turn that into a pre-deployment audit for EU AI Act Annex III high-risk duties and the U.S. EEOC four-fifths threshold.[1], [2]

Who still carries the burden of proof?

UBS had already made AI proficiency a condition of 2027 graduate and intern entry; the employer had not published what the spending returned. Pew’s new survey does not fill that gap. What it fills is the sense that uncertainty grew on both sides of an application: the public expects job loss, while the measured mechanism sits in a hiring score. The next-quarter signal is whether a bank or a platform publishes both a posting audit and a headcount in the same window.[1], [3]