design ai interfaces where i don't have to type. we are missing out ai's inferential potential when we have to constantly spell out what we want . if i point at a face from a photo, ai should have several guesses about my intent: - the experession - facial features - the age/identity