Dream Interpretation Machine Predicts Sleep Images from Brain Scans [VIDEO]
Dream interpretation is possible- in a groundbreaking new brain imaging study, researchers developed a computer model that predicted a person's visual imagery during sleep with at least 60 percent accuracy.
That brings us only slightly closer to "Inception"- this dream interpretation model is relatively basic. Researcher Yukiyasu Kamitani of the Advanced Telecommunications Research Computational Laboratories in Kyoto told Wired that the dream decoder can only identify whether dreamers are seeing particular objects based on their previous brain scans, without any context like color, narrative, or emotion.
Still, the research, presented last year at the Society for Neuroscience conference and published today in the journal Science, is a major step towards understanding why and how we dream, and paves the way for more vivid and coherent dream interpretation using brain imaging.
"We know almost nothing about the function of dreaming," study co-author Masako Tamaki, a neuroscientist at Brown University, told Livescience. "Using this method, we might be able to know more about the function of dreaming."
The research team measured the brain activity of three participants while they were asleep in a functional magnetic resonance imaging (fMRI) brain scanning machine and connected to electroencephalography (EEG) electrodes over the course of ten days.
The researchers woke them every few minutes when the EEG indicated that they were in a light stage of dreaming, to ask what they saw. By the end of that part of the experiment, each participant had been woken up 200 times to provide descriptions of their dream visualizations.
"Many of them were just about daily life in the office," Kamitani told NPR, and some were fantasies like dining with a Japanese movie star. Others were absurd or surreal:
"From the sky, I saw something like a bronze statue, a big bronze statue," one drowsy man told researchers when he was woken up from a dream. "The bronze statue existed on a small hill. Below the hill, there were houses, streets, and trees in an ordinary way."
The participants' visual descriptions were fed into a computer machine learning model that tracked dream keywords to fMRI brain imaging patterns during the dreams. After that part, the participants underwent more fMRI scans while they were awake and looking at images of things that corresponded to the keywords from their dreams. That gave the computer dream interpretation model further context, enabling it to compare what brain regions were active while seeing particular images while sleeping and while awake.
This is your brain on dreams from Science News on Vimeo.
In the last part of the experiment, participants slept and dreamed in an fMRI machine one last time. The machine dream interpretation model analyzed their brain activity during that session and, amazingly, predicted what objects they were visualizing 60 percent of the time.
That's much higher accuracy than chance, and indicates a reliable tool for examining dream imagery.
"It took just a huge amount of non-glamorous work to do this, and they deserve big props for that," neuroscientists Jack Gallant of the University of California, Berkeley, told Wired.
It'll be a long time before we live in a Jetsons world in which personalized dream interpretation machines can record and play back our dreams by morning, but Kamitani's team hopes this breakthrough can lead to more research examining memory formation during dreams, or, with further refinement, a technique for more vivid representations of dream imagery.
"In this field of dream decoding no one has managed to successfully do this before," Gallant added to NPR. "So this is not the final step down this road, it's the first step."