Trust calibration has been the dominant framing for AI‑assisted decision making, aiming to narrow the gap between users' perceived and the actual trustworthiness of AI systems so that reliance is appropriate.
We argue that while this approach works for well‑defined, scoped problems, it overly restricts what decision making with AI support can entail. To overcome this limitation we outline three dimensions of AI‑assisted decision making:
- Trusting AI is only one of several strategies for managing uncertainty; users may also employ supervision, verification, or hybrid approaches.
- When AI tackles “large‑world” decision problems—situations where goals and states are partially known or unknowable—traditional trust calibration is often misapplied.
- In such settings, human‑AI alignment must address higher‑order decision processes, including uncertainty identification, goal negotiation, and co‑evolution of the decision process.
Reflecting on these observations, we advocate for process‑oriented human‑AI coupled systems that continuously surface and negotiate uncertainty throughout the task, enabling genuine bidirectional alignment.
Review