Why is it that things we find very easy to do in our minds (process language, identify people, etc.) are things we can’t get machines to do, but things we find very cumbersome (arithmetic, data manipulation, etc.) we can make machines to do extremely well?

This dilemma, that stuff we can do easily doesn’t imply easy automatability, leads to all kinds of optimism on how human-like computers and other machines are. So what’s the root of it? I think it’s because we don’t know how to understand language, identify faces, etc. We do it, but we don’t know how! For all of recorded history, we’ve been refining mathematics, improving manufacturing processes, so we know in fine detail how to do arithmetic, how to weave cloth, etc. and we teach it to the next generation, write it down in books — whereas we don’t teach our children how to understand language, how to identify their parents and friends, these are mysteries that our brains do for us without being taught, processes which we have no visibility into the internals of. We can make a step by step guide to how to multiply numbers, and make a machine to do it. But we don’t have step-by-step guides to language. The steps are inside our brains and we get only the output. We don’t understand how we understand language, and so find it impossible to make a machine that does it.

Update: For the ~0 of you who’ll read this, I’ve just learned that this is called Moravec’s paradox.