Harms rarely appear at the end.
They enter early, in how a problem is framed, whose data is used, which model is chosen, and they compound from there.
The lifecycle approach is how you catch them where they start.
The AI Lifecycle
A way of seeing where human rights enter the building of an AI system, at every phase, not just the final audit.
Rights enter at every phase
An AI system moves through a recognisable arc
setting the objective, defining requirements, discovering data, selecting a model, testing, and deploying with monitoring.
At each phase, choices are made that shape who the system serves and who it fails. Treating human rights as a box ticked before launch misses where the real decisions happen. The lifecycle approach turns each phase into a point where bias can be found, questioned and addressed, before it travels downstream.
The approach our work is built on
This lifecycle approach is the foundation the rest of the toolbox grows from. It shaped the AI & Equality course taught with the Sorbonne Centre for AI, and it underpins the HRIA Workbook, which turns the approach into a document teams complete stage by stage.
Adapted with CENIA and the University of Chile, the same approach became the first free course of its kind in Latin America.
Technical and non-technical, on the same map
The approach does not stop at framing. It reaches into the technical work, fairness concepts and metrics, methods for detecting and reducing bias, so the people building a system and the people affected by it can reason about it together.
The hardest questions in AI are not only technical, and not only ethical. This is built for both at once.
Where to go from here
To apply the approach to your own system, use the HRIA Workbook. To learn it from the ground up, take the free course.
Support your Human Rights Impact Assessment with our HRIA Workbook
This integrated Human Rights Impact Assessment (HRIA) tool combines two essential approaches for responsible AI development. It can be used as an ongoing reflection tool for developing teams, and for formal assessment documentation, guiding you through a comprehensive assessment throughout the AI lifecycle.
