"Long before it's in the papers"
January 27, 2015


Dream-reading machine in the works?

April 4, 2013
Special to World Science  

Sci­en­tists have ap­plied com­put­er pro­cess­ing to brain scans to find out with some ac­cu­ra­cy what im­ages pop up in sleep­ing peo­ple’s heads, ac­cord­ing to a new re­port.

The re­search­ers, at the ATR Com­puta­t­ional Neu­ro­sci­ence Lab­o­r­a­to­ries in Kyo­to, Ja­pan, ar­ranged for vol­un­teers to re­port their sleep im­age­ry dur­ing brief awak­en­ings. The sci­en­tists then com­piled large lists of im­age types and their ac­com­pa­nying brain ac­ti­vity, as read by a brain scan­ner.

The in­ves­ti­ga­tors next checked wheth­er the same brain ac­ti­vity oc­curred when peo­ple were view­ing im­ages from the same cat­e­gories while awake. It turned out this was largely true, they re­ported.

Next, the sci­en­tists “trained” a com­put­er us­ing ma­chine-learn­ing tech­nol­o­gy to match brain ac­ti­vity to im­age cat­e­gories, us­ing the da­ta from the awake vol­un­teers. The com­put­er learn­ing was custom-tailored to each par­ti­ci­pant.

Fi­nal­ly, the in­ves­ti­ga­tors trained the ma­chine on sleep­ing sub­jects again, and found out that they could pre­dict what the peo­ple had seen with 60 per­cent ac­cu­ra­cy.

The find­ings pro­vide “a means to un­cov­er sub­jec­tive con­tents of dream­ing us­ing ob­jec­tive neu­ral mea­sure­ment,” the sci­en­tists wrote, re­porting their find­ings in the April 5 is­sue of the jour­nal Sci­ence.

The brain scanning technology used was func­tional Mag­netic Re­son­ance Im­ag­ing, which mea­s­ures and maps blood flow in the brain.

Strictly speak­ing, the re­search­ers didn’t study true dream­ing. They in­stead stud­ied dream-like vis­u­al im­ages that oc­cur dur­ing the ear­li­est stage of sleep, be­cause this was eas­i­er to in­ves­t­i­gate for prac­ti­cal rea­sons: “it al­lowed us to col­lect many ob­serva­t­ions by re­peat­ing awak­en­ings,” they wrote. This type of dream­like im­age­ry is called hyp­na­gog­ic hal­lu­cina­t­ion.

Still, “this is probably the first real demon­stra­t­ion of the brain ba­sis of dream con­tent,” said Rob­ert Stick­gold, a neu­ro­sci­ent­ist and dream ex­pert at Har­vard Med­i­cal School who was not in­volved with the study but made the com­ments in an ac­com­pa­nying ar­ti­cle in the jour­nal. He al­so called the find­ings “in­credibly ro­bust.”

More­o­ver, “it’s nice to hear that what peo­ple re­port see­ing when they’re asleep is at least some­what ac­cu­rate,” Stick­gold added in the ar­ti­cle. “Up un­til this mo­ment, there were no grounds on which to say we don’t just make up our dreams when we wake up.”

* * *

Send us a comment on this story, or send it to a friend


Sign up for

On Home Page         


  • St­ar found to have lit­tle plan­ets over twice as old as our own

  • “Kind­ness curricu­lum” may bo­ost suc­cess in pre­schoolers


  • Smart­er mice with a “hum­anized” gene?

  • Was black­mail essen­tial for marr­iage to evolve?

  • Plu­to has even cold­er “twin” of sim­ilar size, studies find

  • Could simple an­ger have taught people to coop­erate?


  • F­rog said to de­scribe its home through song

  • Even r­ats will lend a help­ing paw: study

  • D­rug may undo aging-assoc­iated brain changes in ani­mals

Scientists have applied computer processing to brain scans to find out with some accuracy what images pop up in sleeping people’s heads, according to a new report. The researchers, at the ATR Computational Neuroscience Laboratories in Kyoto, Japan, arranged for volunteers to report their sleep imagery during brief awakenings. The scientists then compiled large lists of images and the accompanying brain activity, as read by a brain scanner The investigators next checked whether the same brain activity occurred when people were viewing images from the same categories while awake. It turned out this was largely true, they reported. Next, the scientists “trained” a computer using machine-learning technology to match brain activity to image categories, using the data from the awake volunteers. The computer learning was custom-tailored to each participant. Finally, the investigators trained the machine on sleeping subjects again, and found out that they could predict what the people had seen with 60% accuracy. The findings provide “a means to uncover subjective contents of dreaming using objective neural measurement,” the scientists wrote, reporting their findings in the April 5 issue of the journal Science. Strictly speaking, the researchers did not actually study dreams. They instead studied dream-like visual images that occur during the earliest stage of sleep, because this was easier to investigate for practical reasons: “it allowed us to collect many observations by repeating awakenings,” they wrote. This type of dreamlike imagery is called hypnagogic hallucination. Still, “This is probably the first real demonstration of the brain basis of dream content,” said Robert Stickgold, a neuroscientist and dream expert at Harvard Medical School who was not involved with the study but made the comments in an accompanying article in the journal. He also called the findings “incredibly robust.” Moreover, “it’s nice to hear that what people report seeing when they’re asleep is at least somewhat accurate,” Stickgold added in the article. “Up until this moment, there were no grounds on which to say we don’t just make up our dreams when we wake up.”