Science & TechExplained

Can a Machine Record Your Dreams and Play Them Back Like a Movie? What the Kyoto Study Really Found

A viral story says scientists can replay your dreams on screen. The research behind it is real, but it decoded object categories, not film.

M
Muhammad Moawaz Alam • 6 min read • Checked against the original research
Dream decoding research explained: what the Kyoto brain scan study found
Scientists in Kyoto decoded object categories from brain scans of sleeping volunteers | Image: JobsBlogs

⚡ Quick Answer

Not yet. A 2013 study at ATR in Kyoto, Japan, used fMRI brain scans and machine learning to predict which kinds of objects three people had just seen in early-sleep dreams. It did not record dreams or produce a movie. The famous 60% figure is how often the decoder picked the right answer in a two-way test, where chance is 50%.

  • The study is from 2013, so this is not a new invention. It keeps resurfacing online.
  • Participants were woken in early, light sleep, not REM sleep as many articles say.
  • The latest brain-to-image AI, reported in October 2026, works on images people view while awake, not on dreams.

A story that keeps going viral says scientists in Japan have built a machine that can record your dreams and play them back for you to watch like a movie. The research behind it is real, and it is remarkable. The headline, however, goes much further than the science.

Researchers at ATR in Kyoto showed that a computer could predict, better than chance, which categories of objects, such as people or places, three volunteers were seeing in dream-like imagery as they fell asleep. They did not capture footage of a dream. Online reactions have been mixed, with some people excited to revisit their dreams and others uneasy about what it means for privacy.

Last checked on Nov 8, 2026 against the lab's research page, Smithsonian Magazine, Snopes, a 2017 follow-up paper and an October 2026 report on newer brain-decoding work.

What Did the Kyoto Researchers Actually Do?

The paper, "Neural Decoding of Visual Imagery During Sleep," was published in Science in 2013 by Horikawa, Tamaki, Miyawaki and Kamitani of ATR's neuroinformatics group. The experiment ran in five steps:

  1. Sleeping in a scanner. Three participants took turns sleeping inside an fMRI scanner, which tracks changes in blood flow in the brain, while EEG monitored their sleep stage. Sessions lasted about three hours and ran over roughly 10 days.
  2. Waking and asking. When EEG showed they had drifted into early sleep, researchers woke them and asked what they had just seen. This was repeated about 200 times per person.
  3. Sorting the reports. The team grouped what people described into about 20 object categories, such as people, places and everyday objects.
  4. Training on awake brains. While awake in the scanner, participants viewed web photos of those categories, so the system could learn each person's brain pattern for each category.
  5. Testing on sleep. The trained decoder was then tested on brain activity recorded just before each awakening, to see whether it could predict what the person had reported.
Why not REM sleep? Many write-ups say the volunteers were in REM sleep. According to the lab and Smithsonian Magazine, they were woken during the sleep-onset period, light sleep when brief dream-like images occur. That timing let researchers collect hundreds of reports in a short time. It is not the same as the long, vivid dreams of REM sleep.

The Viral Claim vs. What the Study Found

Viral claimWhat the research shows
"Records your dreams"The decoder predicted which of about 20 object categories a person had reported. It did not capture the dream itself.
"Plays them back like a movie"There was no film. Snopes rates the claim a "Mixture" and says the output was not a narrative video of a dream.
"Early REM sleep"Participants were woken during sleep onset, in light sleep, not REM.
"Participants confirmed a 60% match with the video"The 60% figure is the decoder's accuracy when choosing between two categories, not a viewer rating of a video.
"A new invention"The study dates from 2013 and has been recycled in viral posts many times since.
"Three participants"Correct. That small sample is one of the study's main limits.

What Does the 60% Accuracy Figure Mean?

For each test, the researchers picked two categories: one the dreamer had reported and one they had not. When the decoder identified one of them, it chose the correct one about 60% of the time. Random guessing in a two-way choice would be right 50% of the time, and the researchers said the 60% result could not be explained by chance.

That is a real scientific result, but a modest one. The decoder was better at telling broad categories apart, such as a person versus a scene, than at telling specific images within a category apart. It also worked only with the preset categories, and only in three people.

✅ What the study showed

  • Brain activity in early sleep carries information about the kinds of objects people report seeing
  • A decoder trained on awake viewing could predict reported categories better than chance
  • Each person's brain pattern for an object was similar awake and asleep

❌ What it did not show

  • Recording or replaying a dream as video
  • Decoding the long, vivid dreams of REM sleep
  • Reconstructing specific images from a sleeping brain
  • Reading anyone's dreams without a scanner and that person's own training data
  • Results from more than three people

What Has Happened Since 2013?

WhenDevelopment
2013The Kyoto team publishes the dream-decoding study in Science.
2017The same lab publishes a follow-up in Frontiers in Computational Neuroscience with two participants. It links dream reports to features from a deep neural network and identifies dreamed objects above chance, but it does not reconstruct images.
Oct 2023A popular article repeats the movie-playback framing and the 60% figure as a participant rating.
Jan 2024Snopes reports that the study's lead author said there had been little progress on dreams specifically. The team's image-reconstruction method had improved, and they were testing whether it could reflect dream content.
Oct 2026Israel's Weizmann Institute reports Brain-IT, an AI that reconstructs images people are viewing while awake from fMRI scans. Its lead researcher says decoding dreams or imagined scenes is a goal, not yet achieved.

Is Dream Decoding a Privacy Risk?

The idea of a machine reading dreams unsettles many people, and the concern is fair to raise. For now, the practical limits are large. The published methods need a cooperative person lying still inside an fMRI scanner, and the decoder was trained on that person's own brain activity while awake. Snopes also notes that brain patterns for the same object differ between individuals.

Researchers say they are exploring simpler tools such as EEG, which uses scalp electrodes. If brain-decoding technology becomes cheaper and easier to use, questions about consent and mental privacy will matter more, so the debate is likely to continue.

Frequently Asked Questions

Can scientists record dreams and play them back like a movie?
No. The Kyoto study used brain scans to predict which object categories, such as people or places, three participants had just seen in early-sleep dreams. It produced no film of a dream.
Who ran the Kyoto dream-decoding study?
Researchers Horikawa, Tamaki, Miyawaki and Kamitani at ATR in Kyoto, Japan. The paper, "Neural Decoding of Visual Imagery During Sleep," appeared in Science in 2013.
What does the 60% accuracy figure mean?
The decoder was asked to choose between two object categories, one the dreamer reported and one they did not. It picked the right one about 60% of the time, where random guessing gives 50%. It does not mean a video matched 60% of a dream.
Was the study done during REM sleep?
No. Participants were woken during the sleep-onset period, light sleep when brief dream-like images occur. Many popular write-ups wrongly say REM.
How many people took part in the study?
Three. Each slept in an fMRI scanner in sessions of about three hours over roughly 10 days and was woken around 200 times to report what they had seen.
Is this a new invention?
No. The study was published in 2013. The story keeps resurfacing on social media and in viral articles, often with exaggerated claims about video playback.
Can AI read my dreams today?
Not in any practical way. The closest recent work, the Weizmann Institute's Brain-IT system reported in October 2026, reconstructs images people view while awake from fMRI scans. Its lead researcher says decoding dreams is a goal the team has not achieved.
Is dream decoding a privacy risk?
Current methods need a cooperative person lying still in an fMRI scanner and a decoder trained on that person's own brain activity, which limits misuse today. Researchers are exploring simpler tools such as EEG, so the privacy debate is likely to continue.
📌 Disclaimer: This article is for general information about published research and is not medical or scientific advice. Findings and technology may change, so check the original sources below for the latest details.

Sources

  1. Neural Decoding of Visual Imagery During Sleep, ATR Kamitani Group (research page)
  2. Scientists Figure Out What You See While You're Dreaming, Smithsonian Magazine
  3. Fact check: Japanese scientists and an MRI machine that records dreams, Snopes
  4. Hierarchical Neural Representation of Dreamed Objects Revealed by Brain Decoding with Deep Neural Network Features, Frontiers in Computational Neuroscience (2017)
  5. This AI Reconstructs What You're Looking at From a Brain Scan, Singularity Hub (Oct 2026)
  6. Example of the viral framing: "Researchers Created a Machine That Can Record Your Dreams…", YourTango (Oct 2023)

For ongoing verified news updates and global current affairs, keep visiting www.jobsblogs.online.

Tags:#DreamDecoding#BrainScan#fMRI#Neuroscience#ScienceExplained#JobsBlogs
M

About JobsBlos

JobsBlogs on world news, official statements and global current affairs.