Researchers are breaking down the barrier between people and machines by teaching computers to recognize fear.
On the 4th floor of the pristine Media Lab Complex at MIT lives a Nightmare Machine.
These computers earned that nickname for a reason: they have been learning how to terrify people. A series of algorithms generates disturbing and grotesque images, like movie monsters, dead people, and other things that go bump in the night.
“We wanted to playfully explore how artificial intelligence (AI) can become a demon that learns how to scare you,” said Pinar Yanardag Delul, one of the creators of the gore-loving computer program.
Delul may facetiously call her creation a demon, but deep down, the study is an attempt to bridge a wide gap between AI and users. Though research in emotional computing is still nascent, there are far-reaching applications. From better horror games to even making real-world driving safer, the possibilities appear endless for AI that can react to human fear.
“Our research group’s main goal is to understand the barriers between human and machine cooperation,” Delul explained.
The project arrived at a salient time, as AI appears to be an essential ingredient for the future.
Recently, president of online taxi service Lyft John Zimmer predicted that all their transportation services would be provided by self-driving cars by the year 2021. Meanwhile, intelligent personal assistants like Siri are getting smarter by learning to adapt and predict user needs. As more and more devices start to operate independently, the need for emotionally-intelligent AI will only increase.
“AI allows machines to make sense, learn from, and interact with the external world, without human beings having to specifically tell the machine what to do,” explained Nidhi Chappell, director of machine learning at Intel.
The kind of AI described above is “machine learning,” which enables computers to “think” by collecting data, and then creating its rules based on that data. Taking this one step further, “deep learning” uses a set of rules called a “neural network,” which are programs modeled after the human brain. Nowadays, “neural networks” are improving automatic translators and, thanks to the Nightmare Machine, even drawing scary images.
Teaching computers about human emotions was a tricky (but important) challenge. The MIT researchers downloaded some 200 thousand images of celebrity faces onto the computer. Then, its “neural network” went to work on analyzing those images.
After some time, it was able to learn the basic features of the human face. Soon enough, the Nightmare Machine graduated to generating new faces itself by simply revising and mashing up old ones.
The results were impressive, but nothing groundbreaking. The researchers wanted a computer that could create faces that actually scared people, which meant figuring out how to translate human psychology into a program.
After some trial and error, they discovered that the computers first needed a crash course in creativity. A second neural network taught them to mimic the artistic styles of painters with a knack for disfigurement and gloom, like Vincent Van Gogh and Edvard Munch.
Finally, for good measure, they added “zombie style” to its arsenal, resulting in the computer churning out thousands of ghoulish creations.
“Even though there is a lot of room for improvement, some of the results already look remarkably creepy,” Delul said.
The research sheds light on how computers may perceive and misperceive human psychology, such as fear, according to Delul. Understanding this part of the human brain can benefit future AI tech in countless ways.
For instance, it can make scary things even scarier. Delul explained how the Nightmare Machine could generate customized images based on individual fears if the computer was synced up to personal data.
Many video games have sought tech that incorporates player feedback, like in the Intel-sponsored horror title Nevermind. Using the player’s own biofeedback, the game transforms environments depending on player stress levels.
Another upcoming stealth horror game Hello Neighbor brings the concept to an advanced AI that learns to adapt according to the players strengths and weaknesses.
Ultimately, the Nightmare Machine’s tech predicts a world where virtual monsters are tailored to respond to the player’s fears and emotional reactions.
Beyond entertainment, interacting with systems that understand a user’s fears even carries the potential to advance mental health therapy.
“Playing these kind of games can help people learn about their fears and potentially help manage them,” said Gabi Zijderveld, of the emotional AI company Affectiva.
Perhaps the most practical use for computers that understand fear is new tech that can help deescalate real world situations when it recognizes people are scared. Zijderveld said that her company is partnering with automakers to embed emotional AI in self-driving vehicles for this reason.
Should a self-driving car detect signs of fear in its passengers, the car needs to be able to understand how to react. This is largely determined by context. For instance, if the passenger’s fear is caused by being followed by a strange automobile, the car’s AI needs to know to speed up. But if the fear is caused by heavy traffic, the car needs to slow down.
Overall, teaching machines about nightmares doesn’t appear to turn them into demons. On the contrary, it could actually make the world a much less frightening place to live in.