Balancing Prediction and Surprise: A Role for Active Sleep at the Dawn of Consciousness?

Abstract

The brain is a prediction machine. Yet the world is never entirely predictable, for any animal. Unexpected events are surprising, and this typically evokes prediction error signatures in mammalian brains. In humans such mismatched expectations are often associated with an emotional response as well, and emotional dysregulation can lead to cognitive disorders such as depression or schizophrenia. Emotional responses are understood to be important for memory consolidation, suggesting that positive or negative 'valence' cues more generally constitute an ancient mechanism designed to potently refine and generalize internal models of the world and thereby minimize prediction errors. On the other hand, abolishing error detection and surprise entirely (as could happen by generalization or habituation) is probably maladaptive, as this might undermine the very mechanism that brains use to become better prediction machines. This paradoxical view of brain function as an ongoing balance between prediction and surprise suggests a compelling approach to study and understand the evolution of consciousness in animals. In particular, this view may provide insight into the function and evolution of 'active' sleep. Here, we propose that active sleep - when animals are behaviorally asleep but their brain seems awake - is widespread beyond mammals and birds, and may have evolved as a mechanism for optimizing predictive processing in motile creatures confronted with constantly changing environments. To explore our hypothesis, we progress from humans to invertebrates, investigating how a potential role for rapid eye movement (REM) sleep in emotional regulation in humans could be re-examined as a conserved sleep function that co-evolved alongside selective attention to maintain an adaptive balance between prediction and surprise. This view of active sleep has some interesting implications for the evolution of subjective awareness and consciousness in animals.

Keywords: REM sleep; consciousness; emotions; invertebrate; predictive coding.

FIGURE 1

Hypothesized evolution of active and quiet sleep, with rapid eye movement (REM) sleep and slow wave sleep (SWS) in mammals and birds representing specialized solutions to achieving distinct sleep functions. Example animals where different forms of sleep have been characterized are shown, arranged schematically by increasing brain complexity. Adapted from Kirszenblat and van Swinderen (2015).

FIGURE 2

Predictive coding and oddball paradigms. (A) A prediction of the animal hiding in a box (a cat) based on a set of ears turns out to be in error (it’s a fox) when further details about the animal are revealed. In this case the prediction error is the misattribution of the animal as a cat. (B) A simple schema of core tenets of predictive coding theory. Sensory input (rainbow arrow) interfaces with a low-level representation (R) unit, which generates a mismatch that is used to refine an error (E) signal within a feedback architecture. This error signal also receives predictions from higher-level representation units while simultaneously supplying these units with updates. By arranging these units in a hierarchical manner, each layer can be used to represent different levels in processing, all the way from simple visual features such as orientation up to abstract concepts and ideas. (C) A schema of a simple oddball paradigm and prediction error signal. In this case an image of a cat (the Standard, S) is presented repeatedly, occasionally replaced with an image of a fox (the Deviant, D). The standards (S) evoke a reproducible response from the brain (purple trace) while the deviant (D) (typically matched for low-level features) evokes a different response (yellow arrow), which is detectable by EEG and/or functional magnetic resonance imaging (fMRI).

FIGURE 3

A general overview of phase precession as represented by an example of a rat moving through 1-dimensional space. (A) A hippocampal theta (θ) waveform has overlaid onto it the activity of three place cells (spiking frequencies are represented as A, B, and C; bottom). The position of each neuron’s spiking on the phase of theta is determined by the temporal sequence of the rat’s movement. As the rat moves through the regions represented by A, B, and C, the phase of each of these neuron’s spiking shifts further from 360°, such that each successive space is represented by an ordered sequence within a single theta cycle. (B) By reactivating theta during REM sleep, a rat replays the temporal sequences that became phase locked to theta during waking. It can be seen that the activity of many more neurons than A, B, and C could be encoded and linked onto theta, representing the role of theta in encoding more than just place fields and thus creating a variety of predictive models.