Bias You Can Notice vs Bias You Can’t

While generating exam questions with generative AI, I noticed a subtle pattern: the correct answer almost never appeared in position (a). The content was fine. The bias was procedural — and invisible until I knew where to look.

This essay explores the difference between bias we can notice and bias we can’t, and why the most dangerous biases aren’t ideological or malicious. They’re structural, normalized, and easy to miss — especially in systems that move quickly, confidently, and without looking back.