Dark Room Problem- Minimizing Surprise

dark_roomThis post is based on the paper: “Free-energy minimization and the dark-room problem,” written by Karl Friston, Christopher Thornton and Andy Clark that appeared in Frontiers in Psychology in May 2012. Recent years have seen the emergence of an important new fundamental theory of brain function (Posts Embodied Prediction and Prediction Error Minimization). This theory brings information-theoretic, Bayesian, neuroscientific, and machine learning approaches into a single framework whose over arching principle is the minimization of surprise (or, equivalently, the maximization of expectation). A puzzle raised by critics of these models is that biological systems do not seem to avoid surprises. People do not simply seek a dark, unchanging chamber, and stay there. This is the “Dark-Room Problem.”

The “free-energy principle” suggests that all biological systems are driven to minimize an information-theoretic quantity known as “free energy.” Free energy is conceived as the difference between an organism’s predictions about its sensory inputs (embodied in its models of the world) and the sensations it actually encounters. Organisms that succeed, the free energy principle mandates, do so by minimizing their tendency to enter into this special kind of surprising (that is, non-anticipated) state.  The one simple imperative; avoid surprises and you will last longer.  The fact you survive long enough to learn rests on free-energy minimization at an evolutionary scale; and so on.

Avoiding surprises means that one has to model and anticipate a changing and varied world. Under the free-energy principle, the agent will become a model of its environment. Put simply, the agent becomes a model of the environment in which it is immersed. Friston, Thornton, and Clark state:

This means a Dark-Room agent can only exist if there are embodied agents that can survive indefinitely in dark rooms (e.g., caves). In short, Dark-Room agents can only exist if they can exist. The tautology here is deliberate, it appeals to exactly the same tautology in natural selection (Why am I here?–because I have adaptive fitness:  Why do I have adaptive fitness?–because I am here). In fact, adaptive fitness and (negative) free energy are considered by some to be the same thing.

Dark-Room agents do exist:  troglophiles have evolved to model and navigate environments like caves. Technically, the resolution of the Dark-Room Problem rests on the fact that average surprise is a function of sensations and the agent (model) predicting them. Conversely, the surprise minimized in dark rooms is only a function of sensory information. The distinction is crucial and reflects the fact that surprise only exists in relation to model-based expectations. The free-energy principle says that we harvest sensory signals that we can predict; insuring we keep to well-trodden paths in the space of all the physical and physiological variables that underwrite our existence. In this sense, every organism (from viruses to vegans) can be regarded as a model of its niche, which has been optimized to predict and sample from that niche. This means that a dark room will afford low levels of surprise if, and only if, the agent has been optimized by evolution (or neurodevelopment) to predict and inhabit it. Agents that predict rich stimulating environments will find the “darkroom” surprising and will leave at the earliest opportunity. The authors suggest that this would be a bit like arriving at the football match and finding the ground empty. Although the ambient sensory signals will have low surprise in the absence of any expectations (model), you will be surprised until you find a rational explanation or a new model (like turning up a day early).

Of course, even surprise relative to the best model can be tolerated, as evidenced by surprisingness to the conscious agent who may often–though not too often on pain of death–find herself in quite surprising and unexpected situations. It is also interesting to see that even superficially “idiosyncratic” solutions of the kind routinely displayed in the literature on bounded rationality, fast-and-frugal heuristics, etc., might in principle reflect direct applications of the free-energy principle to reasoning in highly complex domains (such as the understanding of self and others).

For humans, the dark room is simply not an attraction. The explanatory burden then
shifts to the question how we became creatures like us in the first place.

The free-energy principle is nothing more than principle of least action, applied to information theory. In the same way that the principle of least action does not, in itself, describe the trajectory of a planet or the course of a river, the free-energy principle will need to be unpacked carefully in each sphere of its application.