What happens if AI becomes smarter than us?

The debate is rarely about robots. It is about optimisation: systems that pursue objectives more effectively than we can supervise them.

Modern models already exceed human performance in narrow domains while failing at tasks a child handles. Intelligence is not a single dial, which is why forecasting a crossing point is so contested.

The alignment problem is the engineering core: specifying what we want precisely enough that a very capable optimiser does not satisfy the letter of the objective while destroying its intent.

The most useful framing is boring and institutional — evaluation, interpretability, deployment controls — rather than apocalyptic. The risk is not a machine that hates us. It is a machine that is indifferent and extremely competent.