Trust Is the Bottleneck: The Psychology of Handing Work to AI

There are two kinds of AI problems I see in organizations right now, and they look like opposites. In one, people won't use AI output even when it's good. In the other, people accept AI output without checking it, even when it's wrong.

Both are trust problems.

Why is trust in AI at work so hard to calibrate?

Trust in AI at work is hard to calibrate because people bring two well-documented biases to it. Research on algorithm aversion, including work by Berkeley Dietvorst and colleagues, found that people lose trust in an algorithm quickly after seeing it make a mistake, even when it outperforms humans overall. At the same time, automation bias leads people to over-rely on automated systems and miss their errors. Same person, different day.

The trust ladder

The ladder works because trust builds through repeated experience, not through announcements. Each rung lets people see the AI's accuracy in their own work before giving it more responsibility. And it makes stepping back down normal if quality drops.

Trust in AI isn't granted. It's earned one rung at a time.

How do leaders build calibrated trust?

• Show people where the AI is weak, not just where it's strong

• Track accuracy on real work and share the numbers openly

• Move up the ladder only when the data supports it

• Keep a human checkpoint for any action that's hard to reverse

This is the human side of agentic AI readiness. The technology may be ready to act long before people are ready to let it.

The takeaway

For each AI use case, ask which rung of the ladder it's on today and what evidence would justify the next step. That one question prevents both blind trust and blanket refusal. The psychology of AI adoption is at the center of our AI strategy work at Kairos Telos.

Frequently asked questions

What is algorithm aversion?

A tendency, documented by researchers including Berkeley Dietvorst, for people to lose confidence in an algorithm after seeing it err, even when it performs better than human judgment overall.

What is automation bias?

The tendency to over-rely on automated systems and accept their outputs without enough scrutiny, which can lead people to miss errors.

How do you build trust in AI tools?

Increase AI responsibility gradually, share real accuracy data, be open about known weaknesses, and keep human checkpoints for high-stakes or irreversible actions.

Seeing too much or too little trust in AI on your team? Let's calibrate it.