Although we are taught not "to judge a book by its cover," it appears AI can evaluate book covers just as good as we do.

This is an incredible leap for artificial intelligence, although developments do point toward an eventual arrival in aesthetics. We already know that AI can "judge" people by the way they look, and it appears they can do the same for books.

According to tTechnology Review, Brian Iwana and Seiichi Uchida from Kyushu University in Japan "trained" a neural network to "study" book covers. It was successfully able to determine the category of said books, a whopping 137,788 unique book covers from Amazon.com.

The neural network had 20 possible genres to fit the books on. When Amazon lists a book in more than one category, they used the first one as its primary category.

The pair then used 80 percent of the set to "train" the neural network to recognize the book's genre via its cover image. It has four "layers" with 512 neurons each. All of them will try to see the correlation between the design and the genre.

The pair used another 10 percent to validate their model and the last 10percent to check how the AI analyzes books it has never seen.

The results are astounding. Apparently, the AI neural network was able to list the correct genre in its top three choices for over 40 percent of the tests. It was also able to find the exact genre 20 percent of the time.

Apparently, other categories also appear to be more recognizable than the other. For instance, travel books and technology books are easier to spot because designers use similar images and design. Cookbooks are also easy to spot if they use food in the cover but are harder to see if they use different designs.

The network also had trouble recognizing biographies as the AI often chooses "history" as a category. Interestingly, Amazon also had "history" as the said books' second category.

The AI appears to confuse children's books, comic books and graphic novels with each other. The same appears to be the same for medical books and science books. However, the aforementioned book genres do have similar designs.

However, Iwana and Uchida did acknowledge that they were not able to compare the performance of their network to an actual person "recognizing" the said books. That could make for another interesting experiment.

Until then, no one exactly knows if machines are better than this than humans. However, it may be time to acknowledge that robots are slowly getting more capable of outperforming us.