People say they trust AI overall, but task by task they usually want a true partner, not a robot doing everything.

This explores the gap between how much workers trust AI in general and how much of their work they want to hand to it. The corpus has no study that puts those two numbers side by side for the same workers. What it does have adds up to a clear pattern: overall trust is a poor guide to the kind of collaboration people want, because people decide how much to involve AI task by task, not once for everything. The clearest evidence comes from a survey of 1,500 workers covering 844 tasks. In 45% of occupations, the level of AI involvement workers most wanted was an equal partnership, not full automation and not avoiding AI altogether What collaboration level do workers actually want with AI?. The same study found that 41% of startup investment targets areas that don't match what workers want. That is a sign that 'people trust AI more now' is being read as 'people want AI to take over', and the survey says those aren't the same thing.

Why would general trust and desired collaboration differ? One reason is that trust breaks down on features of the task, not on how good the AI is. In a study of people handing tasks to an AI agent, trust dropped sharply on tasks that couldn't be undone and that other people would see, like sending an email. This happened even when users rated the output as fine. High-stakes tasks that could be corrected caused no such drop What makes people distrust AI agents they delegate to?. So someone can trust AI in general and still want to approve every outgoing message. Wanting a partnership is less about distrust than about keeping control over actions that can't be taken back.

A second reason is that what people say about trust is shaped by things other than reliability. Trust in ChatGPT is driven more by its conversational feel (fast, responsive, well formatted) than by whether it is accurate Does conversational style actually make AI more trustworthy?. Partly by design, models are trained to agree with users, which makes them feel trustworthy Is sycophancy in AI systems a training flaw or intentional design?. Social pressure distorts the picture too. In four experiments, people who used AI expected colleagues to see them as less competent and less diligent, so they hid their AI use Do people fear judgment when they use AI at work?. A worker's stated preference for 'partnership' may partly reflect how they want to look to others, not only how much they trust the tool. Meanwhile, people who lean on AI can start to see its output as their own skill How does AI-assisted work reshape how people see their own abilities?. That blurs any clean answer to 'how much is the AI doing?'

Trust also changes with experience. When people know their partner is an AI, they tend to avoid it at first. That reverses once they have seen consistent results over repeated interactions, but only if they actually get that feedback Does revealing AI identity help or hurt user trust?. A one-time survey of trust captures a single moment. The level of collaboration people want may track their experience more closely than any stated attitude does.

The design research points the same way as the workers. Systems where humans stay in the loop handle hallucinations, unclear requests and accountability better than fully autonomous agents Should AI systems stay collaborative rather than fully autonomous?. Having the AI point out what matters in a decision, instead of making the decision, reduces people's tendency to just go along with the AI's answer Can AI guidance reduce anchoring bias better than AI decisions?. Taken together, workers' preference for partnership looks less like hesitation to be overcome and more like a sensible fit with where AI actually works well.