I think the difference is that it's not really able to produce _new_ thought that isn't based entirely off existing data. It _has_ to have an existing dataset already otherwise it'll just produce random noise.