Opinion
We keep asking AI to be neutral. It never was
Every system reflects the choices of the people who built it. Pretending otherwise is the most dangerous choice of all.
Technology columnist
5 min read
Every model is trained on choices: which data, which rules, which answers count as correct.
Every few months, a new controversy erupts because an AI system gave an answer someone found biased. And every time, the company promises to make it more neutral.
It is a promise that cannot be kept.
Choices all the way down
What data to use. What to filter. Which answers to reward. Each decision reflects values. There is no view from nowhere, and no model trained on one.
Neutrality is not a setting. It is a story companies tell to avoid explaining their choices.
What we should ask for instead
Not neutrality, but transparency: clear documentation of how systems are built and tested, independent audits, and the right to know when a decision about us was made by software. That is harder to promise. It is also possible.
Topics
Ines Okafor
Technology columnist
Ines writes about technology and power, and who gets to decide what the machines are for.
