AI makes Halloween scarier than ever


In the true spirit of Halloween 2020, horror doesn’t just come from the minds of human beings – artificial intelligence (AI) is getting in on the frights as well. Matt Reed, who works at creative agency Red Pepper, used a generative adversarial network – a type of AI – to create a series of profoundly unsettling Halloween masks. In this case, the AI is ‘trained’ on a large set of data (5000 Halloween masks in this case), after which it is set free to produce its own freakish masks using what it learnt of the similarities between the training images.

“What’s so scary or unsettling about it is that it’s not so detailed that it shows you everything. It leaves just enough open for your imagination to connect the dots,” said Matt, in an interview with New Scientist. And they truly are horrifying, evoking a sense of dread and confusion that is difficult to capture in human-made objects.

Interestingly, human-like images don’t need to be as featureless and blurry as these masks to evoke a sense of fear; pictures of faces with only a slight irregularity in features can give rise to profound unsettlement and even fear. The reason behind this lies, perhaps, in The Uncanny Valley, a theory developed by Masahiro Mori stating that “viewers are generally less accepting of a character as the human likeness of that character increases”. Basically, the closer a character looks to a human, the easier it is for people to notice that something is ‘off’, leading to rejection. 

Robots are designed to look less human-like to avoid negative reactions

ITU Pictures

Many people have remarked online that the human characters in the 2004 animated film The Polar Express are pure nightmare-fuel, inducing feelings of revulsion and horror perhaps similar to those caused by Matt Reed’s masks. Baby Yoda, on the other hand, looks much less like a human, and is therefore insanely cute – perhaps more so than Disney anticipated, given that Baby Yoda toys have consistently sold out as soon as they’re released.

The Uncanny Valley isn’t just restricted to the realms of movies and Halloween; many researchers in the field of robotics actively design their robots to look less human-like to avoid these negative reactions. For example, compare your reaction to these two robots: Sophia (right), a robot designed to look extremely similar to a human (and now an official citizen of Saudi Arabia!), and Pepper (left), a Japanese ‘companion’ robot able to read human emotions. I know which one I wouldn’t like to see in the corner of my room at night (although admittedly a large white robot suddenly appearing out of nowhere would probably be fairly frightening as well…).

Pictures of faces with only a slight irregularity in features can give rise to profound unsettlement and even fear

@melip0ne via Twitter

AI is also being applied to make other images immensely unsettling as well. Twitter user @melip0ne posted this image with the caption “name one thing in this photo”, a challenge many found incredibly difficult, frustrating, and profoundly disturbing. This image, once again, appears to have been created using a generative adversarial network (the same as the masks), having been ‘primed’ on thousands of example photos. 

The reason we find the result unsettling is because the AI has created objects from composites of the material it was primed on, resulting in a confusing mess that still seems ‘realistic’. Overall, it seems like rather than taking over the world, AI is much better suited at creating unsettling images. I’m excited to see where the technology takes us in the future, although hopefully it will be a little less scary!

Image: Matt Reed via Twitter

Leave a Reply

Your email address will not be published.


This site uses Akismet to reduce spam. Learn how your comment data is processed.