On 28 October Stanford's Institute for Human-Centered AI ran a one-day conference at the Hoover Institution called Confronting Our AI Future: Hope, Fear, and the Choices Ahead, with Fei-Fei Li, James Landay, Jeff Dean and Audrey Tang on the bill. Around it, HAI ran an informal survey of more than 600 attendees and followers and shared the results on LinkedIn the following day. We have been through the 12-page results PDF.
There is one limitation one must keep in mind. The respondents had chosen to attend an AI institute's event, so their views cannot be taken as representative of the wider industry. Still, they offer a useful glimpse into how people who work closely with AI see its adoption. They are much like the enthusiasts in your own firm who are likely to push for new tools. More importantly, the questions HAI put to them are rarely discussed with project teams or QS departments before a new tool arrives in the workplace.
What the room said
Asked for one word for how AI makes them feel, 41 people said excited, 30 anxious, 25 curious, 20 concerned and 18 hopeful. The word cloud behind the top five carried tired, overwhelmed, uneasy, terrified, angry and empowered.
Almost everyone in the room already uses the stuff. Of the 558 who answered, 391 use tools like ChatGPT or Claude daily, 95 weekly, 31 a few times a month and 41 hardly ever. That is 70% daily use.
The trust question came next. Asked how much they trust that AI is being developed, deployed and regulated in a way that keeps people safe, 228 of 554 said not at all. That is 41%. Another 160 said a little, 137 somewhat and 29 a great deal. These are people who use the technology every day and, for the most part, do not trust the system around it.
Belief in benefit held up anyway. Asked whether AI will meaningfully improve people's lives over the next decade, 202 of 553 said a great deal and 174 somewhat, against 123 a little and 54 not at all. For comparison, the Ipsos AI Monitor in the 2026 AI Index has 59% of adults across 30 countries saying AI products have more benefits than drawbacks, up from 55% in 2024.
Hopes in Stanford's free text clustered on medical research, on AI becoming "humanity's trusted partner" and on overcoming disparities if it is deployed ethically. Luke Soon, who wrote up the survey on his Genesis Human Experience blog on 30 October, lists the top concerns as privacy and security, hallucinations and human extinction.
Heavy use and low trust are the same person
The usual reading of resistance to AI in a delivery team is that the sceptics are the people who have not tried it. Stanford's room suggests the opposite can be true. The heaviest users can hold the lowest trust in the system around the tool, because they have seen what it gets wrong.
Two of HAI's own leaders said as much in the AI Hopes and Fears series the institute ran on LinkedIn from 24 to 29 October. Mykel Kochenderfer, an HAI associate director who works on validation and safety, named his fear as "premature acceptance of AI without proper validation", as quoted by Soon.
Russ Altman, a physician-scientist and also an associate director, said "it's time to regulate frontier companies". The QS who runs a tender comparison through an AI tool every day is also the person who has watched it misread a rate. Their trust is informed, and informed trust is what a governance process should be built on.
The five questions to ask before the rollout
Strip out HAI's branding, and the survey is a usable intake form. Put these to the team that will actually use the tool, anonymously, before the rollout and again three months after:
In one word, how does AI make you feel at work right now?
How often do you already use AI tools, whether or not the firm has sanctioned them?
How much do you trust that the tools we are deploying are being built, deployed and governed in a way that keeps you and the client safe?
How much do you believe this tool will meaningfully improve your work over the next year?
Which of these are you: a pragmatist, a cautious optimist, a builder, a sceptic or an explorer?
That last question is HAI's own, and the labels are worth borrowing. Stanford's room split 28.3% pragmatist ("AI is a tool that we have to decide how to use"), 25.3% cautious optimist ("hopeful, but guardrails first"), 22.7% builder, 18.2% sceptic ("currently not convinced the benefits outweigh the costs") and 5.4% explorer.
We would not guess which way a construction team splits. That is the point of asking.
Where this meets competence debt
On 27 October we carried Digital Storm's piece on 'competence debt'. At four Polish endoscopy centres, 19 experienced doctors saw their unaided adenoma detection rate fall from 28.4% to 22.4% within three months of an AI rollout. The argument is that the human in the loop becomes a rubber stamp unless the firm deliberately keeps unaided skill alive.
The Stanford questions are the front end of that problem. A team that does not trust the tool's governance will check its output, which is tiring but safe. A team that trusts it a great deal may have stopped checking. The sentiment survey and the unaided skill baseline are two halves of the same control.
Takeaway
Survey the team before the tool arrives rather than after the first complaint. Five anonymous questions take ten minutes. Record the baseline and repeat it at three months alongside a test of unaided performance, so you can see whether trust has moved and whether skill has slipped. Where your own people land is the one number in this article we cannot give you.
Project Flux follows the governance side of AI in construction every week, from Stanford's trust gap to the competence debt evidence. If you would rather have the questions before the rollout than after it, the newsletter is free at projectflux.ai.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
