El mal alumne. Pedagogia crítica per a intel·ligències artificials studies the mechanisms of artificial sight and learning by means of neural nets. Nowadays, machines can see and learn, but what do ‘seeing and learning’ mean? Is it possible to formulate a critical pedagogy concerning their learning? How can we subvert the usual uses – commercial and monitoring – of these tools? Our research project aims to experiment in this field by using three basic strategies: producing alternative learning; evaluating the unexpected results produced by these tools (very often regarded as errors); and making these tools talk not about images but about themselves, revealing characteristics of their learning.