At ethics by design: At which lifecycle stage should the governing policy require this ethical evaluation to

Building fairness and harm considerations into requirements before development is the ethics-by-design stage of the lifecycle.

The question

Preparing for AIGP? Take the free 5-min readiness quiz →

1. At deployment and monitoring, so that discriminatory effects are detected through live metrics once the system is serving real users.

   Plausible for catching drift, but this is post-launch detection, not the pre-development evaluation the company wants.
2. At ethics by design, so fairness and potential-harm considerations are built into the system's requirements before development begins. ✓

   Correct. Ethics by design embeds fairness considerations into requirements before building starts.
3. At incident management, so the organization can respond and remediate after a discriminatory outcome has already harmed some individuals.

   Almost right as a safety net, but incident management is reactive and occurs after harm, not before code is written.
4. At training and testing, so bias is measured against representative datasets after the model has already been designed and constructed.

   Plausible, since bias testing matters, but it happens after design and build, later than the requested pre-development stage.

How to remember it

Building fairness and harm considerations into requirements before development is the ethics-by-design stage of the lifecycle.

More Understanding the Foundations of AI Governance questions

• Deciding what information is appropriate to enter and how: From a governance perspective, which concern →
• The data privacy policy: Which existing policy most directly needs review before proceeding? →
• Conduct a vendor due-diligence assessment covering: Which action best reflects sound third-party AI risk →
• All 337 Understanding the Foundations of AI Governance questions →