AI is clearly useful, but usefulness doesn't automatically justify unlimited deployment, opaque training practices or turning every public service and workplace into an experiment