The Dream Catcher experiment: Blinded analyses disconfirm markers of dreaming consciousness in EEG spectral power

Summary

A test called the Dream Catcher test was conducted for the first time in a simplified form to see if brain activity alone can reveal whether someone is dreaming. Data Team collected brain measurements (polysomnograms) during NREM sleep from 9 participants, producing 54 one-minute recordings—27 from dreamful sleep and 27 from dreamless sleep. A blinded Analysis Team tried to classify each recording as dreamful or dreamless using an unsupervised machine learning classifier based on EEG spectral power and electrode location. Over five iterations with gradually reduced blindness, the team never performed significantly better than chance. The results suggest that EEG spectral power does not carry signatures of phenomenal consciousness, and the study also failed to replicate key findings from earlier reports on dreaming consciousness.

Study at a glance

Abstract

The Dream Catcher test defines the criteria for a genuine discovery of the neural constituents of phenomenal consciousness. Passing the test implies that some patterns of purely brain-based data directly correspond to the subjective features of phenomenal experience, which would help to bridge the explanatory gap between consciousness and brain. Here, we conducted the Dream Catcher test for the first time in a graded and simplified form, capturing its core idea. The experiment involved a Data Team, who measured participants’ brain activity during sleep and collected dream reports, and a blinded Analysis Team, who was challenged to predict better than chance, based solely on brain measurements, whether or not a participant had a dream experience. Using a serial-awakening paradigm, the Data Team prepared 54 one-minute polysomnograms of NREM sleep—27 of dreamful sleep (3 from each of the 9 participants) and 27 of dreamless sleep—redacting from them all associated participant and dream information. The Analysis Team attempted to classify each recording as either dreamless or dreamful using an unsupervised machine learning classifier, based on hypothesis-driven, extracted features of EEG spectral power and electrode location. The procedure was repeated over five iterations with a gradual removal of blindness. At no level of blindness did the Analysis Team perform significantly better than chance, suggesting that EEG spectral power does not carry any signatures of phenomenal consciousness. Furthermore, we demonstrate an outright failure to replicate key findings of recently reported correlates of dreaming consciousness.