Can You Undo the Mistake? AI governance in 25 seconds.
From a live TESS Group AI workshop · captions included
“If we automate a process and something happens, how easy is it to reverse the mistake? A wrong invoice number, that's fixable. A wrong medical dose is not. So the higher the stakes, the more human oversight you need.”
One question does most of the work in AI risk assessment. This clip is the shortest useful summary of human-in-the-loop thinking you'll find, straight from a live TESS governance session.
Transcript
What happens if it goes wrong? I think this is a really, really interesting one to think about. If we automate a process and something happens, how easy is it to reverse the mistake? A wrong invoice number is something that's fixable. A wrong medical dose is not. So the higher the stakes, the more human oversight you need.
AI apprenticeships · 100% levy-funded · Ofsted Good · 4.9/5 from 776 reviews
Book a Free Discovery Call
The one question to ask before you automate
Reversibility is the fastest risk triage you can run. For every process you plan to automate, ask how easily a wrong output can be undone and who would notice. Reversible and visible: automate freely and review weekly. Reversible but invisible: automate with a log and a spot-check. Irreversible: keep a human approval step, however slow it feels. This is the thinking behind the governance module in the Level 4 AI and Automation apprenticeship and the AI Adoption and Governance unit for leaders. If you need a written policy, start with how to write an AI policy.