Skip to content
Align your AI

The problem space · Devices · Overconfidence

Align your AI Devices against overconfidence

People supervising a machine that usually works can trust it more than its design allows.

Devices
the machines it runs on or moves
Overconfidence
certainty the evidence does not back

Ways it goes wrong

People trust a machine that usually works. The NTSB cited Uber's lack of ways to address "automation complacency" in its safety drivers1. Tesla's recall concerned drivers misusing a feature that steers for them2.

What people do about it

The NTSB recommended that test plans cover crashes and operator inattention, including monitoring whether drivers are engaged1. NHTSA later opened a query into whether Tesla's software fix was enough3.

  • Monitor the supervisor

    Watches whether the person meant to supervise the machine is paying attention, and warns or stops when they are not.

  • Check that a fix holds

    Keeps measuring after a software remedy ships, to see whether the misuse it targeted goes down.

  • Plan for a distracted operator

    Designs testing so that a safety driver who looks away is expected, not ruled out.

Where we touch it

No study of ours covers this pair yet.

Sources

  1. NTSB, “Collision Between Vehicle Controlled by Developmental Automated Driving System and Pedestrian”, investigation HWY18MH010. https://www.ntsb.gov/investigations/Pages/HWY18MH010.aspx. Accessed 2026-10-14. ↩ ↩2

  2. NHTSA, safety recall report 23V-838, Tesla Autosteer, 2023-12-12. https://static.nhtsa.gov/odi/rcl/2023/RCLRPT-23V838-8276.PDF. Accessed 2026-10-14. ↩

  3. NHTSA, recall query RQ24009, “Recall 23V838 Remedy Effectiveness”, opened 2024-04-25. https://static.nhtsa.gov/odi/inv/2024/INOA-RQ24009-12046.pdf. Accessed 2026-10-14. ↩