An Information Theory of Losing Consciousness

Unconsciousness is more than just weird, it's unnerving. All of a sudden, we're just gone for many hours at a time, left only with a sometimes improbable-seeming return ticket to awareness. We dream briefly (and only briefly), but otherwise sleep is mostly non-space where the brain has decided to exclude you from the body's affairs. It's like autopilot, but you're locked out of the cockpit.

Consciousness itself remains a mystery in many respects, scientifically speaking, and it's consciousness that determines this present/not-present sleep boundary. Which seems tautologous, but I'm not so sure most of us think of sleep as the dissolution of consciousness so much as a switching on and off—or pausing—of it. It's easier to think of things like that, anyhow.

A study out this week in the Physical Review Letters takes a close look at the biophysics of anesthesia, in particular, and just how the brain manages to accomplish such a complete and sudden switching off of consciousness while later reversing the process just as swiftly. The general finding, which is based on computer simulations of neural activity with respect to information entropy, is that the brain does this via an abrupt, global change in its ability to transmit signals through networks of neurons.

So, yes, if unconsciousness by itself is unnerving, than anesthesia is terrifying, at least to me and quite a few others. Researchers (anesthesiologists) don't even know how it works exactly, just that it (usually) does. To get to the bottom of it, Yan Xu and colleagues at the University of Pittsburgh School of Medicine created a relatively simple model based on a tree-like network of nodes. An input signal arrives via the trunk and is then disseminated across the different branches, a process that's regulated by the nodes themselves, which determine which signals to pass along and which to bury.

Whether or not a node passes along a signal is a probabilistic determination. Each node is assigned a probability factor of p which can vary between 1 and 0, e.g. between 100 percent probability and no probability at all. Moment to moment a node is either stop or go (open or closed), with the likelihood of either case governed by p. When p is equal to 0, we can say that the node is "turned off." The general model is known as percolation, a theoretical framework used to describe the likelihood of a signal making it through all of those p-valued nodes to the other side of the network.

Xu and his group found that in using such a simulation based on 7381 nodes they were able to mimic the brain activity of anesthetized patients. They arrived at their interpretation using standard information theory and statistical white noise-based inputs. The network's input trunk was fed some randomness with some known degree of entropy and this entropy was then compared to that of the output layer. The entropy of the output layer dropped dramatically at p = .3, signaling that little information was being transmitted through the network. Turns out that this is what we should expect.

"[The model] generates stereotypical EEG features under general anesthesia while reproducing dose-response characteristics for loss of consciousness," Xu and colleagues write. "Linking the loss and gain of information access to anesthesia induction and emergence, the model provides a fundamental theory of information emerging from a stochastic process and suggests that cognitive features are enabled as a phase transition." So, you are either on or off, depending upon that phase.

"Although the brain is many orders of magnitude more complex," the group concludes, "it is tempting to speculate that the transition between conscious and unconscious states is also regulated by a single connectivity parameter, especially considering the clinical observation that a sharp transition between conscious and unconscious states occurs within an extremely narrow anesthetic concentration range, with little variation among human subjects or even among different species of vastly different brain scales and capacities."

I'm not sure that this model of unconsciousness, that of the depowered or disassembled robot, puts me much at ease about the whole ceasing to exist for eight hours a night thing. To mix metaphors, you're not locked out of the cockpit after all; it's just that the cockpit no longer exists.