Can Computers Pass A Rorschach Inkblot Test?

Can a computer tell this looks like a deamon*?

Source: Wikimedia

Here's an important thing to know about artificial intelligence: it is not very good with pictures. It can't really tell what a picture is of, nor can it determine context. Show a computer a picture of a glass of milk, and it wouldn't be surprising for it to guess it's a bird.

Humans, on the other hand, are very good at pictures. This is why Rorschach Tests—those mirrored inkblots—are interesting for us. They aren't really pictures of anything, but we humans are very good at seeing things in them.

So what happens when you give a meaningless inkblot to artificial intelligence algorithms that are trying (but not always doing very well) to see what's in an image? Some pretty amusing and interesting results!

Fernanda Viégas and Martin Wattenberg did just that. They tested four artificial intelligence systems with a series of inkblots and found some interesting results. One algorithm was pretty good at recognizing it was being tested with ink blots. (Imagine the passive-aggressive, irritated robot voice saying, "Sigh. That is a picture of a Rorschach Test inkblot") Another was highly interpretative, labeling many of the ink blots with the word "isolated." Poor algorithm.

Poor, sad Robot 3.

Source: Wattenberg & Viegas

Check out the full results over on their blog.

_{*Yes, I spelled it that way on purpose. A little computing joke. https://en.wikipedia.org/wiki/Daemon_%28computing%29}