Verified by 
Key points
- Information of supreme value involves disconfirmation of "priors" about the world and self.
- Information of supreme significance is preferentially processed via REM.
- The brain/mind, therefore, is oriented toward the pick-up of information that recruits REM.
While it is certainly true that rapid eye movement (REM) sleep functions, in part, to promote a healthy waking mind, there are compelling reasons to believe that the opposite is the case as well: the waking mind functions, in part, to promote a healthy REM mind. Part of our waking consciousness is given over to the service of REM sleep and dreams.
Perhaps the most important thing the brain does is seek out information that is of supreme value to the organism. The brain’s networks and information processing capacities are organized around this overriding imperative to identify, discover and process information of supreme value.
What qualifies as information of supreme value to the human organism? There are many cues the brain uses to find that gold: “surprise” is one such cue. Music and the visual arts use surprise to satisfy our everyday aesthetic appetites, but these are small potatoes when it comes to information of supreme value and real surprises.
We get information of supreme value when a surprising event disconfirms beliefs about the Self and world. It is especially valuable information to know that your “priors” or models are not adequate to the world. When that surprising event, in particular, detects a pattern in the noise when only noise was expected, the information undermines the assumption of statistically independent outcomes in the sensory feedback data. It suggests the presence of some alternative (to the expected) mechanism at work.
If the deviations of a model’s predictions lack expected randomness, it implies that a superior predictive model is available. Thus the mind is operating at peak intensity when seeking randomness–not so much meaning or patterns in noise. It is seeking randomness so that it will be surprised when it notices patterns in the noise as that is the gold standard in novel knowledge acquisition.
How can the mind disregard the evolutionary imperative to seek out patterns in noise and instead operate to detect randomness? For an individual mind to discover/process randomness where the pattern was expected requires recruitment of REM neurobiology. Surprises that lead to recruitment of REM are not merely improbable events; instead, they indicate a breakdown in the current cognitive model being used to model the world or quantify probability.
After a supremely surprising event occurs, the brain/mind attempts to recruit the physiological orienting response in order to begin to process it, but the magnitude of the surprise disorients the individual, and processing of the event is shunted over to the night REM system, which employs specialized systems to handle these kinds of events. For example, PGO (pontine-geniculo-occipital) waves index powerful orientating systems recruited to process the info. The default mode network, salience network, and emotional memory consolidation systems are recruited to further process the information during REM without interference from external visual information associated with waking consciousness.
REM, in short, specializes in processing surprise events that exhibit special properties such that the mismatch between expected and obtained data is large enough that it effectively calls into question the accuracy of the individual’s cognitive modeling strategy itself. Its “Bayesian priors,” or for our purposes here, we can assume informs the individual’s sense of self, becomes the object of the updating process.
The surprise event and an error signal are extreme enough to trigger PGO waves and REM as part of its orientation and integration reaction. In other words, normal cognitive processes are not strong enough to cognize the information delivered by the surprising event, and thus REM sleep is recruited to help. This REM-based integration process is accompanied by cognitive events we call dreams. It is in these dreams that additional cognitive materials for the integration of information of supreme value occur.
The simplest quantitative surprise model assumes that the intensity of surprise about an event is proportional to its improbability, i.e., where the unexpected event disconfirms a single belief (e.g., the person expects to see a red square, but a green square appears instead).
But info of supreme value disconfirms several beliefs about the modeler her/himself and the world as well, either simultaneously or sequentially. The more unexpected something is, the more information it likely has, and that information is the kind of information the brain needs to build adequate models of anticipated challenges.
In particular, surprise-related information points to how much work we have to do to update our models, beliefs, and expectations. The work involved in belief updating involves the generation of counterfactuals -endless "what if..." scenario spinning and evaluation. Generation of counterfactuals to the information of supreme value occurs as REM dreaming.
The degree of surprise I experience encodes the information on how well my current self is adapted to coming challenges and how much work I will need to do to transform myself to meet those upcoming challenges. Will I need to scrap my “priors” completely and build a whole new self from scratch? Or will simple editing procedures suffice?
Given that surprise encodes this kind of information—information absolutely crucial for the sense of self and model updating, it follows that my sense of self is necessarily anchored in estimations of the surprisingness of the things I encounter in the world. But surprisingness, in turn, may be measured by the ease with which or the number of the counterfactuals triggered by the surprising event.
These counterfactuals can occur anytime, but they most vividly and most often occur in dreams. The fluency with which counterfactual generation and comparison process occur after surprise yields information on how significant that event is for the truth value of prior beliefs. Suppose priors cannot account for the surprising event. In that case, that fact leads to an information gain and thus can be formally measured as a Kullback-Leibler divergence (DKL) between prior and posterior beliefs. In the brain, surprise as expectation violation correlates with fMRI blood-oxygen-level-dependent imaging (BOLD) responses in the salience network, including the anterior cingulate cortex and anterior insula. These same brain regions (among others) are modulated during REM.
I conclude that the brain/mind is actively seeking out information that can recruit REM physiology during the day. REM runs the show.
