linkblog: just read 'Now that machines can learn, can they unlearn? | Ars Technica'