Halloween, observed on October 31st marks the day before the Western Christian feast of AI Saints and initiates the season of Allhallowtide, which lasts three days and concludes with All Souls' Day. In recent times, this Western festival has been gaining popularity in India as well. 

 Halloween 2022 is the first since the widespread ability of the public to generate AI imagery from simple prospects. This time of the year, AI researchers and artists are exploring AI algorithms' creativity to commemorate the festival. It is evident from the results that AI's creativity is escalating. From developing ideas for scary masks and costumes to generating bizarre haunted houses, AI is taking the Halloween season to the next level. 

The Nightmare Machine 

 MIT Media lab scientists' creation of the Nightmare Machine proves that AI can evoke terror similar to what one might feel watching a horror movie. Nightmare Machine is a website displaying images of people and places that an ML algorithm has completely zombified. The project suggests a truly frightening future of Halloween. The AI taps into individual fears and exploits them. 

 Using 200,000 images of celebrities, the MIT team generated a data set of fake faces. They borrowed a framework that can learn the artistic style of a given image and applied it to the generated faces to maximize the fear factor. 

 The researchers then launched a site asking people to rate the faces, and each response was used to train the algorithm on what terrifies people. Some regard the faces as very scary, and others don't. According to Pinar Yanardag, a postdoc at the MIT Media Lab, the result reveals extra information about how humans perceive horror. 

Horror writing AI 

 Pinar Yanardag created the world's first collaborative AI horror writer. The bot, 'Shelley,' is named after Frankenstein writer Mary Shelley. Shelley is a Deep Learning system trained on more than 140,000 horror stories. 

 Shelley lives on Twitter. Every hour, it tweets out the beginning of a new horror story and the hashtag #yourturn to invite human collaborators. Anyone is welcome to reply to the tweet with the next part of the story. Shelley will take the story forward by replying again with the following part. 

Virtual haunted house 

AI research scientist Janelle Shane, who runs an ML humor blog, generated a virtual haunted house using Open AI's CLIP and VQGAN neural networks. CLIP learns visual concepts of natural language supervision, using images and descriptions to rate how well a given image matches a phrase. The algorithm uses zero-shot learning, a training methodology that decreases reliance on labelled data. 

 Shane focused on the phrase "Haunted Victorian House", starting with a photo of a regular Victorian house and then letting the AI use its feedback to modify the image with the details associated with the word "Haunted". 

Apart from the haunted house, Janelle Shane has also been working on the algorithms with out-of-the-box costume ideas. 

Image credit: Janelle Shane, AI Weirdness

The future 

 AI could have some intriguing implications for the business of horror. According to MIT's Yanardag's opinion, we could generate personalized horror images in the future. The researchers consider the project a fun way to get into the Halloween spirit. But it lays a foundation for understanding the obstacles to human and machine cooperation.