The ‘Hallucination Machine’ Alters Consciousness in the Name of Science
To study the nature of consciousness, researchers have replaced LSD with a mind-bending virtual reality experience powered by Google's Deep Dream.
There are dog heads everywhere — materializing from the sides of buildings, etched eerily into the overcast sky. A passerby turns to look at you and her body transforms into a hybrid pheasant-poodle. The sidewalk below you is peppered with eyeballs.
No, you’re not on drugs, but your brain might have a hard time telling the difference. This is an experience within the Hallucination Machine, a VR headset that immerses users in a 360-degree psychedelic version of reality that is generated with the help of Google’s Deep Dream generator.
Researchers at the Sackler Center for Consciousness Science at the University of Sussex in the UK developed the Hallucination Machine to see what altered states of consciousness can teach us about normal consciousness. Early results published in the online journal Scientific Reports point to striking similarities between the experience of virtual hallucinations and the real thing.
It’s been nearly 400 years since René Descartes first wrestled with the enigma of human consciousness, and the best minds in neuroscience and cognitive psychology are still trying to unravel the mystery of how our brains conjure up “the self” and process a constant flow of incoming sensory perceptions as “reality.”
One way to understand how a complex system works is to break it. If you want to know how the different parts of an engine work together, take out the carburetor and see what happens. The same is true of consciousness.
To make sense of the complex relationship between external reality and internal perception, you can start by messing it up. By giving test subjects small doses of psychedelic drugs like LSD or psilocybin, researchers can study how changes in brain chemistry can disrupt the normal processing pathways of consciousness.
But giving people acid in the lab is tricky. Not only because they keep wanting to hear that same Pink Floyd song over and over, but because it’s hard for scientists to distinguish between the drug’s physical effects (sensory overload, basically) and deeper psychological shifts in consciousness.
That’s where the Hallucination Machine comes in.
“We don’t really know how drugs affect your brain,” Keisuke Suzuki, a postdoctoral fellow at the Sackler Center and lead author of the first proof-of-concept study of the VR headset’s effectiveness on human subjects, told Seeker.
“It might be that you just experience modified sensory input, not necessarily a change in consciousness,” he said. “Our machine can separate these different aspects that are normally triggered all at once when taking psychedelic drugs.”
To create the hallucinogenic virtual world, Suzuki and his colleagues first shot a panoramic 4K video of a bustling student plaza on the Sussex campus. They then gave each frame of the three-minute video the Deep Dream treatment. Deep Dream is an algorithm that uses artificial neural networks to transform normal images of flowers and faces into wildly psychedelic landscapes. After a week of processing, the result was a 360-degree street scene straight out of Alice in Wonderland.
In one experiment, participants watched two versions of the panoramic video in their VR headsets — the normal street scene and the altered Deep Dream version. Then they were asked to indicate whether they experienced various psychedelic sensations, such as seeing patterns and colors, floating, dissolution of self, emotional arousal, and distortion of time.
As expected, the plain version of the videos elicited few abnormal sensations, while the hallucinogenic version triggered a bunch.
The most striking result was when Suzuki compared the responses from the Hallucination Machine to the experience of people who had actually been dosed with psilocybin. They matched up almost exactly — the same distorted sense of size and space, the same visualization of patterns and colors.
The only thing that didn’t correlate was the classic distortion of time that people experience on psychedelics, when just a minute of a really intense trip can feel like an hour. In a separate experiment, participants were challenged to reproduce the exact time length of different musical tones while immersed in the normal and psychedelic VR worlds. They performed equally well in both.
Suzuki said the negative result might have to do with the fact that they were using pre-recorded video from another location, where the participants’ only interactivity with the virtual world was rotating their heads around.
“Everyone immediately knows this isn’t real,” said Suzuki. “What I want to do next is patch through a real-time video feed from a camera on their heads. What happens if you start seeing a bunch of dots on your hands, and your real-time actions are turned into hallucinations? That might elicit much stronger reactions.”
What’s fascinating about the Hallucination Machine is not only the similarity between the effects of the virtual hallucinations and real psychedelic drugs, but also the similarities between our brains and the Deep Dream neural network that conjured up the altered images.
Google stumbled on Deep Dream as an unexpected byproduct of the race to create an accurate image search engine. Neural networks are designed to mimic how a brain learns by training to slowly recognize the shape of a bicycle or an apple among a database of a million images.
Image search neural networks are built of multiple layers of artificial neurons, with the lowest layer responsible for lines, curves, and rough shapes, and higher layers responsible for identifying component figures like leaves and hands, and the top layer responsible for tying all the clues together and deciding if it’s a tree or a boy. This is called a bottom-up process.
What Deep Dream does is move back down the neural network in the other direction. If you upload an image into the Deep Dream generator, the network starts searching for recognizable shapes in the clouds, trees, buildings, and faces. Then, instead of simply categorizing and naming shapes, it reverses directions and creates its own versions of those shapes, embedding them in the image. (The training database that Google engineers used had a lot of dogs in it, hence the ubiquitous dog heads in Deep Dream hallucinations.)
The brain uses a similar hierarchical system in visual perception, with the earliest parts of the visual cortex responsible for processing low-level features and the inferior temporal cortex handling the highest-level categories. That’s not to say that the brain always works from the bottom up.
“I believe there are people who have actual visual hallucinations who could have similar process in their brain, a top-down visual process in their visual cortex,” said Suzuki, including people who take psychedelic drugs. “That’s why they see an imaginary image on top of the real world.”