Every AI system you deploy is a set of frozen decisions — about what to optimize for, what to ignore, and what counts as success. The medieval Islamic philosopher Al-Biruni noticed something structurally similar when he studied foreign knowledge systems in the eleventh century: each tradition had embedded assumptions so foundational that practitioners couldn't see them as assumptions at all. They just looked like reality. He called this the difference between explicit doctrine and tacit premise — and he argued that the tacit premises were the ones that actually governed behavior. Your AI tools work the same way. The loss function, the training data, the evaluation metric — these are tacit premises baked in by someone else, often optimized for a context that isn't yours. The danger isn't that the system fails. It's that it succeeds so smoothly that you stop noticing what it's quietly deprioritizing. This Sunday is a reasonable moment to treat your AI stack not as a set of answers but as a set of inherited arguments — and to ask which of those arguments you've actually examined.

Pick one AI tool central to your work. What outcome was it actually optimized for — and is that the same outcome you're trying to achieve?

Drawing from Islamic Comparative Philosophy / Epistemology — Abu Rayhan Al-Biruni