Build It With the People It Affects

The cheapest fairness intervention in AI isn't a toolkit or a consultant. It's a conversation — with the people the system will affect, held before you build it.

Too often, affected communities meet the AI system for the first time as its subjects: the applicants it screened, the customers it flagged, the students it sorted. By then, the design choices are baked in and feedback is expensive. Invite them earlier — when you're defining the problem and the success criteria — and they'll spot issues your team can't see from the inside.

In plain English: the people closest to the problem usually see the bias first.

This isn't charity; it's diligence. Frontline staff know where the current process already fails certain groups. Community members know which data about them is wrong or incomplete. That knowledge is free, and no amount of post-launch tuning replaces it.

One practical format: a single listening session before design is finalized, with the explicit question "how could this system get this wrong for people like you?" Write down what you hear. Act on what you can. Document the rest.

Nothing about us without us — it applies to AI too.

I explain frameworks; I don't give legal counsel.