We start from a guess.

An ansatz is the form you assume a solution takes before you can prove it. Physics runs on them. So does any research worth doing.

This page is one. Behind it is a small research group in São Paulo working on the theory of aligned artificial agents: what it means for a system that acts in the world to want something, to learn, to accept correction. Old questions, from decision theory and the foundations of agency, asked again about systems that may soon reason as well as we do.

The lab opens soon.