The problem with this picture is that everything in it is familiar.
If you take a quick glance, you’ll be a 100% sure that you’ve seen all these objects before but a closer look can change everything. Everything you thought was familiar will suddenly look strange and alien and you’ll have no idea what you’re looking at.
This image was shared by @melip0ne on April 22nd and the caption challenged viewers to identify any one of the objects in it. Over the past week, it has confused most of the people who’ve come across it and responses have ranged from mundane to utterly bizarre.
So why exactly is this weird picture making people feel this way?
According to Dr. Frank McAndrew who teaches psychology at Knox College, Illinois, the picture is full of ambiguity which makes viewers feel very uncertain and consequently it also seems creepy. When you’re not sure what an object is, you are also not sure whether it is dangerous and so it only to be expected that you’ll feel ill at ease.
But again, it is quite evident that this picture isn’t like to hurt anyone so why are we still not okay with it?
McAndrew also explained that we feel mildly scared because the confusion leads to a competition in the brain as it tries to put these objects into categories that we can comprehend. Even after a lot of effort, we still can’t understand it and according to Dr. Steven Schlozman, who teaches psychiatry at Harvard, this only makes us feel even more discomfited.
Schlozman says that we feel creeped out because our brains keep trying to zoom in on a recognizable pattern but the pattern keeps getting disrupted as it shifts into another one that we think we can recognize. We understand a bit of it but not enough to figure out what we’re looking at.
What Do Robots Dream About?
Janelle Shane, an electrical engineer, pointed out that the image must have been created by artificial intelligence or AI. Her work is to create neural networks or AI that can gather knowledge the way the human brain does. She says that the image was most likely created by BigGAN, a Google algorithm that can create pictures with a lot of details by itself.
The Generative Adversarial Network or GAN is given over a thousand pictures to use as examples as well as different categories of pictures to work with. It’s even more interesting because it can even be trained to create a picture that falls into more than one category. We can’t identify the things in these pictures because there’s nothing like them in reality. They are combinations of different things juxtaposed on each other by a machine.
On her blog, AI weirdness, Shane herself had earlier written a post on BigGAN. If you play around with its capabilities and the examples you give it, the results can be delightful. For example, giving BigGAN pictures of flowers and dogs can give you some really cute dogflowers.
Using AIs trained to recognize images, Shane was able to figure out that the weird things in the pictures possibly came from other pictures of bakeries, toy stores, department stores, etc. She explains everything in this twitter thread.
While AI sometimes isn’t great at coming up with accurate depictions, some of them are excellent at what they do. StyleGAN can create almost terrifyingly real pictures of people but when researchers tried to do it with cats, the results were scary and best forgotten. The problem is AI isn’t human but it is created by humans so the way it interprets images is close to the way we do it but it is not the same. These differences are enough to make human viewers intensely uncomfortable. The AI understands certain aspects of photography but it doesn’t realize what the objects in the images are used for.
Shane also threw in a bone for the people who actually like looking at these scary pictures and would like to generate some of their own. Check out Ganbreeder and have fun.