Artificial neural networks are considerably less complex than the real thing we carry around in our heads. We have very little clue why we make decisions either

Artificial neural networks are considerably less complex than the real thing we carry around in our heads. We have very little clue why we make decisions either, and various bits of psychological research show that it's much less rational that we'd like to believe. So how to make progress? I think often we make decisions intuitively, then seek rationalisations. Sometimes we're self-deluded about how good those rationalisations are, because it is difficult to change a decision we have committed to. But if we avoid that, it strikes me as a good enough model: we can concentrate on at how rigorous the justification is, and worry less about the question of genesis. And so I think the same thing should go for AI. We've lost the battle on AI that we can understand. We should focus more on a science of justification, or decision testing. And if we can develop and be publicly committed to that kind of science, we can apply it to human decision-making too.