Replay is a memory process we can measure, not a dream we can watch
Replay means that the brain, while asleep, runs again through patterns of activity it produced while awake. In rats, hippocampal cells that fired in a particular order during exploration fire again in a similar order during rapid eye movement (REM) sleep, the stage in which human dreaming is most often reported [2]. In humans, researchers can now detect these reactivations during sleep using machine-learning classifiers applied to EEG recordings, and can even trigger them by replaying sounds that were paired with specific memories during wake—a method called targeted memory reactivation [3].
The key point for the dream question is what these methods actually measure. They detect that a memory-related pattern has reappeared; they do not measure whether the sleeper is having an experience. A classifier that flags a reactivation in slow-wave sleep tells us the brain is reprocessing something, not that a dream is happening [3]. That gap is why the evidence cannot settle whether replay is the machinery behind dreaming or a separate, non-conscious process running alongside it.
Replay does occur in REM sleep, the stage most linked to dreaming
If replay were only a non-conscious, deep-sleep phenomenon, the dream question would be easier. But replay has been found in REM sleep too. In rats, temporally structured sequences of awake hippocampal activity are reactivated during REM sleep [2]. More recent work in mice confirmed that representations of awake exploration reappear during REM episodes, with varying degrees of temporal structure, and that temporally organized sequences were replayed more often during REM sleep after a novel experience [5].
This overlap in timing is suggestive but not decisive. REM sleep is when people most often report vivid dreams, and it is also a stage where replay occurs [2][5]. That co-occurrence is consistent with replay contributing to dream content, but none of these studies recorded dream reports alongside the neural recordings, so the link is a correlation in timing rather than a demonstrated correspondence.
The evidence points to replay serving memory, whether or not anyone is dreaming
Across species and methods, replay keeps showing up as a memory-consolidation mechanism. In mice, salient new experiences sharpen the temporal structure of sequences replayed in later REM sleep, which the authors interpret as a role for REM sleep in consolidating important experiences [5]. In humans, replay of task-related sequences has been detected during rest, and replay is thought to support learning and decision-making by reactivating internal representations rather than literal re-runs of experience [4].
Work in artificial neural networks makes the same point from a different direction: a sleep-like phase of spontaneous replay protected old memories from being overwritten during new learning, without any claim about consciousness [1]. Together these lines of evidence suggest replay is a functional process for organizing and preserving memories, and they leave open whether that process is accompanied by dream experience, causes it, or neither.
Why the dream question is still unanswered
The obstacle is methodological. Replay is fast, sequential, and distributed across the brain, and current non-invasive methods in humans offer either good timing or good spatial detail but not both [4]. Detecting replay in humans at all has required new analysis techniques, and doing so in REM sleep with targeted memory reactivation is only a recent first step [3].
Meanwhile, the studies that do find replay in REM sleep are in rodents, where dream experience cannot be reported [2][5]. The one human REM replay work described here is explicitly framed as a starting point for further investigation [3]. So the question of whether replay corresponds to conscious dreaming is not answered by the current evidence—it is a question the available methods are only now making it possible to ask.
Sources used in this answer
Sleep-like unsupervised replay reduces catastrophic forgetting in artificial neural networks
A sleep-like phase of unsupervised replay in artificial neural networks recovered previously learned tasks and reduced catastrophic forgetting, but the study did not address consciousness or dreaming.
Temporally structured replay of awake hippocampal ensemble activity during rapid eye movement sleep
Awake hippocampal activity patterns were reactivated in temporally structured sequences during REM sleep in rats, the sleep stage associated with human dreaming.
Detecting neural replay in sleep with EEG classifiers
Machine-learning classifiers applied to EEG detected memory reactivations after targeted memory reactivation in human slow-wave sleep, and a new pipeline successfully classified reactivation in human REM sleep.
Investigating neural replay of task representations in the human brain using fMRI
New fMRI analysis methods detected sequential replay of task-related stimulus representations during post-task rest in humans, supporting a role for replay in learning and decision-making.
Structured and unstructured reactivations during REM sleep are modulated by novel experiences
Hippocampal representations of awake exploration were recapitulated during REM sleep in mice, and temporally organized sequences were replayed more often after a novel experience.