Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsMIT’s Halloween AI demonstrations explored two different kinds of machine-made horror: the 2016 Nightmare Machine generated unsettling images for people to rate, while Shelley, introduced in 2017, collaborated with people on horror stories. Both relied on human reactions and contributions; neither showed that AI understood fear as people do or proved that AI was independently dangerous.
Two MIT projects, not one “nightmare-fuel AI” system
The title phrase can suggest a single frightening machine, but MIT’s Halloween demonstrations were separate projects with different outputs and forms of participation.
| Project | Introduced | What it produced | How people participated |
|---|---|---|---|
| Nightmare Machine | 2016 | Scary images of faces and places | Visitors rated the images; the votes were used to steer the algorithm toward scarier results. |
| Shelley | 2017 | Collaborative horror stories | People replied to story openings with continuations, and Shelley added to the story. |
MIT framed the work as an exploration of how people respond to machine-made horror. In 2016, Iyad Rahwan, then an associate professor of media arts and sciences in the MIT Media Lab, said: “Halloween is a time when people celebrate the things that terrify them. So it seems like a perfect occasion for an MIT project that explores society’s fear of AI.”
How the Nightmare Machine made scary images
In its October 31, 2016 account, MIT News described a deep-learning image-generation approach. One example involved learning features of a haunted house and applying them to a photograph of the Media Lab; a related approach generated frightening faces. Visitors could view and rate the results on the project website. MIT News’ 2016 account reported more than 300,000 individual votes at the time.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
#1 Best Overall
A contemporaneous report in MIT’s student newspaper, The Tech, described “Haunted Places” and “Haunted Faces” categories and said that visitor votes helped train the algorithm toward scarier images. Researcher Manuel Cebrian told The Tech that the site had received more than 800,000 individual evaluations and over one million visitors in one week. The Tech’s November 3, 2016 report gives those figures as an interview-era account. These figures are reported separately by the two publications; they should not be treated as measurements of the same metric.
How Shelley turned horror into a collaboration
MIT introduced Shelley on October 27, 2017, as a system trained on more than 140,000 horror stories from Reddit’s r/nosleep. It posted story openings on Twitter with the hashtag #yourturn. People could reply with continuations, after which Shelley generated another segment; completed stories were collected on its project website at the time.
Rank #2
- Used Book in Good Condition
Shelley’s project lead, Pinar Yanardhag, described it this way: “Shelley is a combination of a multi-layer recurrent neural network and an online learning algorithm that learns from crowd’s feedback over time.” The human replies were part of the storytelling process, not merely ratings of a finished output. MIT News’ introduction to Shelley also cautioned that the source community included adult content and that researchers had limited control over the system, adding “so parents beware.” It was not presented as suitable for children.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What these demonstrations do—and do not—show
The projects illustrate ways to involve people in machine-generated horror: judging images in one case, and adding story continuations in the other. Their results depended on human feedback and interaction. The accounts show what the systems produced and how people used them; they do not establish that either system experienced or understood fear as a human does, or that AI is inherently dangerous.
The cited coverage is historical. It does not establish whether either demonstration remains accessible today, so their project sites should not be assumed to be working services.
Quick Recap
Rank #4
- Piano/Vocal
- Pages: 82
- Instrumentation: Guitar
- Instrumentation: Piano/Keyboard
- Instrumentation: Vocal
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
