Can a Machine Judge a Book By Its Cover?

It’s not a question most of us have considered, but hey, that’s what the scientists are for, right?

Brian Kenji Iwana and Seiichi Uchida are neuro-scientists at Kyushu University in Japan, who set out to answer this exact question. Well, a version of it, at least. 

The goal was to test an AI’s ability to match a book cover with 20 given genres. To do so, the pair developped a highly sophisticated neural network that would perform this task. So we still don’t know what a robot’s favorite Harry Potter book is, but hey, we’ll get there. 

The results were surprising. The machine correctly matched the cover to the genre 40% of the time in its first three guesses. 20% of the time, it nailed the first one.

Upon further analysis, it was observed that tech and travel books were the most easily discernable, while the algorithm frequently confused children’s books with comics and graphic novels as well as medical books and science books. 

The data is not fully conclusive, and it hasn’t yet been compared to a human performance. But it’s an exciting direction for AI to go!

Featured image courtesy of MIT review