Just in time for Halloween, MIT has launched a horror website, Nightmare Machine, to answer the question: can machines learn how to freak humans out?
Judging by some of the dark landscapes and ghoulish faces its machine created, the answer would be yes.
Recent advances in AI have made some people image a scary future ruled by merciless, intelligent robots. Telling horror stories about AI is nothing new.
As the researchers note, storytellers have been terrifying audiences with similar human-created monsters for centuries. However, with the help of deep neural networks, it won't be just humans creating these stories.
MIT's researchers are also hoping for help to improve their AI nightmare factory. To help out, go to this link and say which of the 10 images you find scary.
The researchers from MIT's MediaLab say they summoned "deep learning algorithms and evil spirits" to generate the scary images.
The researchers point to two algorithms they used. For scary faces, they borrowed an algorithm developed by researchers at Indica Research in Boston and Facebook's AI Research, who've used them to generate entirely fictional but realistic-looking images of people's faces, rooms and buildings.
Another algorithm they point to was developed to mimic the processes humans follow to create art. It's been used previously to create impressionist art.
MIT's Nightmare Machine creators point out that AI and horror fiction have a long shared history, most notably, Frankenstein, created by Mary Shelley in 1816 while locked in a mansion with Lord Byron and John William Polidori, both key figures in the creation of vampirism as literary subject.
Byron's daughter Ada Lovelace would go on to write the world's first machine-learning algorithm in 1840.
Today, companies like Google and Microsoft are asking whether AI can make art. Microsoft collaborated with Dutch researchers and museums to create Rembrandt-like 3D-printed portrait of an entirely fictional figure based on the artist's paintings between 1632 and 1642.
Google's AI has been producing art in a quest to understand why some artificial neural networks work and others don't. In some cases, its networks would end-up 'imagining' things in an image based on data they were trained with.
Tech giants weren't the first to make machines that create art. Painter and programmer Harold Cohen developed Aaron in the 1960s. Aaron became a "world-class colorist" without guidance from a human, Cohen told the BBC.