Impact Levels: How Much Oversight Does an AI System Need?

The more an AI system affects people’s rights, money, health or safety, the more human oversight it needs. Governments sort systems into impact levels to decide how much. This simplified simulator is a small risk register: answer a few plain questions about an AI system, tick the safeguard that reduces each risk, and see the Impact Level (I to IV) and the oversight it carries. It illustrates the idea behind the Government of Canada’s Algorithmic Impact Assessment (AIA).

Pick an example system, or answer the questions yourself. Each risk has its own safeguard beside it: tick it to see how many points it removes and whether it changes the Impact Level. The result stays in view as you go. The leader’s rule: tier first, then control.

What this teaches: Answer plain questions about an AI system, then add the safeguard that reduces each risk and see which Impact Level and oversight it lands on. It teaches risk-based tiering, how each control targets a specific risk, and why safeguards reduce risk but never remove it.

The impact-level simulator puts the “oversight scales with impact” idea from DAVHILL’s AI Fundamentals for Professionals (opens in a new tab) course in your hands.

Sorting AI systems by risk before choosing the controls is part of the closing module of DAVHILL’s Intro to Data Science (opens in a new tab) course.