On The Human Capacity For Physical And Analogical Inference