In laboratories at Oxford and Cambridge, researchers are recording the electrical activity of the brains of volunteers who are simply lying down with their eyes closed, not performing any tasks. Instead of traditional experiments with stimuli and responses, scientists are applying machine learning algorithms to these 'free-form' signals. The resulting patterns allow not just the observation of brain activity, but the prediction of what a person is thinking at any given moment – a meeting that lies ahead, or a conversation that has already occurred.
The review article, published in Nature Reviews Neuroscience on July 15, 2026, is authored by Andrea Luppi and his colleagues from Oxford, Cambridge, Montreal, and Dublin. Luppi, a Wellcome Early Career Fellow at St John's College Cambridge, is the first and corresponding author of the work.
The article argues that spontaneous brain activity is ceasing to be a hindrance to understanding consciousness. The authors demonstrate how data-driven approaches extract meaningful information from signals that researchers previously dismissed as noise. The work integrates data from EEG, fMRI, and comparative analysis of brain activity in humans, mice, and macaques. However, it does not present new experiments; it is a systematic review summarizing already published research. The authors report no conflicts of interest; funding was provided through standard university grants.
Traditional theories of consciousness, such as the Global Workspace Theory, were built on studies involving external stimuli, where a person's attention is directed towards the external world. Spontaneous activity challenges this conventional scenario: is a separate mechanism truly required for an 'internal' mode of thinking? Predictive Processing, an alternative theory, posits that the brain constantly generates predictions about the external world.
The data from Luppi and his colleagues suggest that these predictions remain structured and cognitively significant even when a person receives no external input – the brain continues to model, analyze, and forecast. However, the review itself does not provide a definitive answer to the fundamental question: is such structured spontaneous activity sufficient for the emergence of phenomenal consciousness (subjective experience), or does it merely reflect functional information processing without accompanying awareness?
Methodological limitations are evident: most studies were conducted on small sample sizes of participants, and machine learning algorithms were trained on data collected during states of complete rest, where volunteers' control over their thoughts and attention is minimal. The authors honestly acknowledge that translating these scientific findings into clinical practice – from managing anesthesia during surgery to diagnosing disorders of consciousness – requires further rigorous testing and validation on larger patient cohorts.
To better understand the approach, imagine a train moving along tracks without a fixed schedule. Previously, engineers only studied stops at stations – moments when the train is stationary or begins to move. Now, they analyze how the carriage moves between stations: its speed, oscillations, and system interactions. Similarly, data-driven methods reveal not just pauses in thinking, but how spontaneous brain activity 'guides' subsequent reactions and prepares the brain for new tasks.
If this approach becomes established in neuroscience, it will change not only the experimental practices of researchers but also our fundamental understanding of what constitutes the 'norm' for a healthy brain. Studies already indicate a reduction in the uniqueness and complexity of functional connectivity under anesthesia and in premature newborns – patterns that could serve as biomarkers of consciousness states.
