One way to curtail the free energy principle is to allow that the idea of a hypothesis-testing mechanism in the brain may be useful for some but not all purposes. Thus the idea could explain, say, visual illusions, but not action. Indeed, versions of the idea in this curtailed form have surfaced many times in the history of philosophy of mind, vision science, and psychology (see Hohwy 2013, Introduction). One view would be that evolution very likely has recruited something like hypothesis-testing, such that the brain can represent the world, but that this likely co-exists with many other types of mechanism that the brain makes use of, for good evolutionary reasons. From this perspective, the universal ambition of the free energy principle is preposterous because it goes against the evolutionary perspective of a tinkering, cobbled-together mechanism.
It is possible of course that a limited-use, Bayesian neural mechanism has evolved in this way. There is no strong evidence that there is in fact something like a circumscribed, modular mechanism. For example, Bayes optimal integration seems to work across modalities and types of sensory attributes (Trommershäuser et al. 2011). On the other hand, there is not yet strong empirical evidence for the ubiquitousness of free energy minimization, though there is emerging evidence of its usefulness for explaining a very surprising range of mental phenomena, from visual perception, illusion, movement, decision, and action.
Speaking more conceptually, the free energy principle is not a theory that lends itself particularly well to piecemeal, curtailed application. Recall that the principle concerns the very shape and structure of the brain, mirroring as it does the causal structure of the world. The very hierarchical morphology of the organ is shaped by free energy minimization. This means that other neural mechanisms, that are not involved in prediction error minimization, would have to have evolved in a way parasitic on the free energy principle rather than alongside it. In this sense, the free energy principle would, at the very least, lay the foundation for everything else. Against this, it could be said that perhaps parts of the brain are not, strictly speaking, part of hierarchical inference. Perhaps subcortical nuclei have evolved independently of free energy. This is therefore an argument for which empirical evidence would be important: are there areas of the brain that are not best described in terms of prediction error message passing?
Continuing the very general approach, the free energy principle has such generality that it tends to monopolize explanation. To demonstrate this, consider the theory of evolution, which is also an extremely ambitious theory in the sense that it aims to explain all parts of biology with just a few very basic tools. It is conceptually possible to curtail this theory: perhaps it explains only 70% of life, leaving some other mechanism to explain the rest, or perhaps it explains only non-human life, leaving some deity to fully explain us. This kind of curtailed view would of course ignore the mountain of evidence there is for evolution in absolutely all parts of life (a point we will revisit in a moment), but it would also miss something about the kind of theory that the theory of evolution is. It seems that, as an explanation, evolution is so powerful that it would be incredible that something else would be equally able to explain life.
Whereas it cannot be stipulated that the theory of evolution is true universally, it can be argued that if it is true, it is true everywhere. To see this, consider that if incontrovertible evidence was found that evolution does not explain, say, the eight eyes of most spiders, then for most people that would cast aspersions on the theory of evolution in all other areas—even where it is backed up with overwhelmingly strong evidence. This is not simply to say that some recalcitrant evidence lowers the posterior probability of the theory somewhat, but rather that it would begin to completely undermine the theory. It seems the theory of evolution posits such a fundamental mechanism that anything short of universal quantification would invalidate it.
Perhaps we can describe what goes on in terms of “explaining away” (Pearl 1988). Imagine, for example, that one night the electricity in your house cuts out. You consider two hypotheses: that a possum has torn down the power line to your house, or that the whole neighbourhood has blacked out due to the recent heat wave. Out in the street you see other people checking their fuse boxes and this evidence favours the second hypothesis. Importantly, this evidence considerably lowers the probability of the possum hypothesis even though the two hypotheses could be true together. There is debate about what explaining away really is, but agreement that it exists. Part of what grounds this notion is that our background knowledge of the frequency of events tells us that it would be rather an unusual coincidence if, just as the overall power goes out due to the heat wave, a possum caused the line to go down (unless possums are known to take to power lines during heat waves). In the case of the deity hypothesis and the evolutionary hypothesis, it seems that explaining away is particularly strong. It would be an utterly astrounding coincidence if something as fundamental as speciation and adaptation had two coinciding explanations.
After this excursion into philosophy of science, we can return to the free energy principle. Though it still has nothing like the amount of evidence in its favour that evolution has, it seems that if it is true then it too must apply everywhere, and if not then it must be false. There is no middle way. This again seems to relate to explaining away. It would be too much of a coincidence if two explanations both accounted for something as fundamental as the organism’s ability to sustain itself in its expected states. If the principle was directed at only fairly superficial aspects of mentality, such as the nature of visual illusions, then it would not strongly explain away other theories. But this misrepresents how deep the explanatory target actually is.
The issue was whether the explanatory ambition of the free energy principle can be curtailed, in order to make it seem less preposterous. If it is assumed that explaining away is particularly strong for fundamental rather than superficial explanations, then it appears that a principle as fundamental as the free energy principle cannot be curtailed. If it is believed, then it is believed with maximal scope. It is therefore misguided to think that one can take a divide and conquer approach to the free energy principle.
Of course, this can be taken to cement its preposterousness. If it is a hypothesis designed to be universal, then how can it be anything but preposterous? The immediate answer to this lies in comparing it again to the theory of evolution. This venerable theory must be preposterous in just the same way, but of course it isn’t—it is true. This means that the issue whether the free energy principle is preposterous cannot be decided just by pointing to its explanatory ambition, since this would also invalidate the theory of evolution. Not surprisingly, it must be resolved by considering the evidence in favour of the free energy principle. As mentioned, this does not yet compare to that of the theory of evolution, though it is noteworthy that evidence is coming in, and that it is coming in from research on a comfortably large suite of mental phenomena.
Consider next the question of what happens with existing, competing theories once something like the free energy principle or the theory of evolution begins to gain explanatory force. Existing theories may have considerable evidence in their favour (this may be a theory about a cognitive or perceptual domain, such as attention or illusion); and they may explain away the existing evidence relatively well and therefore have that evidence in their favour (this contrasts with the comparison with the deity hypothesis, which strictly speaking has no evidence in its favour). Nevertheless, once additional, relevant evidence becomes available, existing theories may begin to lose ground to a new theory, like the free energy principle, even if it as yet has less evidence in its favour. For example, once it is noted that the brain is characterized by plentiful backwards connections, it becomes clear that these must be relevant to phenomena like attention and illusion (for example, disrupting them disrupts attention and illusion). However, if existing theories cannot explain this new evidence, then a new theory can begin to usurp their explanatory job. This means the evidence in their favour begins to wane, even if the new theory is still only enjoying spotty support. Compare again to the electricity blackout example. There might be a very impressive theory of the whereabouts and heat wave-related behavior of possums that very snugly explains the blackout in the house and perhaps other things besides. But the moment we become aware that the whole neighborhood is without electricity, even a poor theory of the blackout that can also address this new evidence (“perhaps it is some central distributor thingamajig that has broken down”) becomes much more attractive than the existing possum theory. New evidence and new theories can very quickly wreak havoc on old, cherished theories. The free energy principle should therefore be expected to usurp the explanatory jobs of existing theories, and thereby challenge them, even if it is still a fairly fledgling theory. Of course, explanatory usurpation depends on acknowledging the occurrence of new evidence, such as the presence of backwards connections in the brain. Perhaps it is no surprise that the free energy principle is beginning to gain ground just as imaging brain science is maturing beyond the phase in which it was concerned mainly with collecting new evidence, and on to a new phase in which researchers consider the theoretical significance of the evidence in terms of both functional specificity and effective connectivity.