Can dreams be recorded?
No one can yet record a dream the way a camera records a scene. Researchers have shown, however, that some information about what a person is picturing can be read from brain activity. In a 2013 study using MRI scanners, Japanese researchers predicted broad categories of imagery that sleepers had been experiencing as they fell asleep. That is a real result, but it is far from recording a full dream.
What the 2013 Japanese study did
The study was led by a team that included Tomoyasu Horikawa and Yukiyasu Kamitani, at the ATR Computational Neuroscience Laboratories in Kyoto. The method had three stages:
- Training while awake. Participants lay in an MRI scanner and looked at images on a screen, so the researchers could record the brain patterns that went with each visual category.
- Recording at sleep onset. Participants were scanned as they drifted into light sleep. Each time they were woken, they were asked what they had been seeing. This was repeated many times for each person, with more than 200 reports from each of the three participants.
- Matching. The researchers compared the brain patterns recorded during sleep with the patterns from the waking training, and tried to predict what category of imagery each sleep report contained.
Participants’ reports were often surreal. They included things like statues, keys, and tools, and the researchers grouped these into broader categories, such as “structures” for buildings and rooms.
What “60 percent accuracy” means
The system predicted the broad category of a sleep report with about 60 percent accuracy. That is above chance, which is what made the result notable. But it is a long way from reading a dream. The system did not reproduce specific objects, scenes, people, or stories. It worked with a small set of categories, and the accuracy figure applies only to those categories.
What the study does not show
Several limits matter:
- It is not a video. No images from the dream were displayed. Some online articles say the machine can “show” dreams. That is not what the study did.
- It covers sleep onset, not full dreams. The reports came from the transition into sleep, which is not the same as the long, vivid dreams of REM sleep. For more on REM, see REM sleep and dreaming.
- The sample is small. Three participants is too few to draw broad conclusions about people in general.
- The models are personal. The system had to be trained on each participant’s brain patterns. An expert quoted in the news coverage of the study noted that brain patterns differ between individuals, so a single general “dream reader” would be hard to build.
Other ways brain activity is recorded during dreams
Other studies have measured dreaming with EEG, which records electrical activity on the scalp. A 2017 study led by Francesca Siclari and colleagues used high-density EEG to link dream experience to activity in the back of the cortex. The method could tell whether a participant had a dream, including during non-REM sleep. It did not reveal what the dream contained.
These studies show that dream experience can be detected in brain activity, but they do not turn that activity into a readable record. For more on the brain side of dreaming, see the neuroscience of dreaming.
Why full recording is hard
Recording a full dream would need several things that current methods do not provide:
- Reliable decoding of rich content. Current systems work with broad categories. Full dream content involves many objects, people, places, emotions, and actions, all linked together.
- Ground truth. The only way to check a decoded dream is the dreamer’s report, and dream reports are often incomplete. Many dreams are forgotten shortly after waking. See why we forget most of our dreams.
- Individual calibration. Each person’s brain would need its own training data, which is time-consuming and requires the person to be awake and cooperative during training.
Claims about dream recorders
Some consumer devices and online articles claim to record or play back dreams. These claims should be judged against the research above. A device that tracks movement or sleep stages can be useful for sleep research, but it does not show the content of a dream. A headline that says scientists can “read” or “record” dreams usually refers to broad categories from a small study, not to full dream recording.
Bottom line
The 2013 Japanese MRI studies showed that some broad categories of imagery can be predicted from brain activity at sleep onset, with about 60 percent accuracy in a small sample. That is a real scientific result, but it is not a dream recording. Full recording of dream content remains out of reach. For background on the field, see oneirology, the scientific study of dreams.
For a related claim about experiences that people describe during sleep, see out-of-body experiences: real or fake?.