In their brilliant but rather undiscovered book, The Dialectical Biologist, Richard Levins and Richard Lewontin write: “a theory that can explain everything explains nothing” (Reference Levins and Lewontin1985, p. 65). The paragraph is worth quoting in its entirety, but on another occasion. They refer to the ideological excesses of Marxism, Freudianism, and Darwinism. In the context of twenty-first century life and mind sciences, a similar cautionary note could be made about Fristonism.
Already well-known for his contributions to functional imaging, the British neuroscientist Karl Friston has distinguished himself over the last decade as a steady proponent of the so-called free energy principle (FEP) (Friston, Reference Friston2010). Friston borrows foundational concepts from thermodynamics and combines them with two good old ideas: Hermann von Helmholtz's understanding of perception as inference (Reference von Helmholtz1867) and Claude Bernard's insight on the self-regulatory nature of vital processes (Reference Bernard1878). The FEP is a normative theory, and its main commandment is the minimization of surprise. Framed in a modern Bayesian framework – whereby beliefs are updated by means of loops of prediction, error, and correction – the brain would be an active inference machine doing predictive processing.
But this is precisely what begs the question. It is not uncommon that a formal approach used by brain scientists to study other brains ends up surreptitiously postulated as what all brains actually do (Gomez-Marin, Reference Gomez-Marin2019): If one can make sense of brains as inference machines, then brains must be inference machines. Indeed, a pervasive temptation among scientists is to treat the conjunction “as if” as “is” (e.g., animals are [like] machines, brains are [like] computers). We know that maps are not terrains, but we still conflate them (Andrews, Reference Andrews2021).
The FEP illustrates what I call “the 3M fallacy,” a sleight of mind whereby captivating metaphors are turned into mathematical models that then become covert or self-ignorant metaphysics. Bruineberg et al. thus offer a timely dissection of the notion of Markov blankets as interpreted by some FEP proponents, making a distinction between mathematical tools to study organisms and ontological pronouncements about those very organisms. The critique would not be so patently urgent had Friston and collaborators refrained from elevating their theoretical framework from a powerful heuristic into a sort of biological theory of everything.
Having started as an ambitious integrative theory of brain function, the FEP has been progressively expanded to behavior, the mind, and even consciousness! Perception, decision making and learning would all be explainable under the FEP too, in animals but also in plants and microorganisms. And yet one should be hesitant of any sweeping principle that accounts for nematode foraging and romantic love. Seeking an understanding of virtually everything in principle, the framework risks explaining actually nothing in practice.
Is Friston's FEP neurobiology's “string theory”? The latter has been deemed by prominent physicists as “not even wrong” (Woit, Reference Woit2006). The FEP is no less ambitious. Sufficiently elaborated for professional acolytes to revere and pursue, it appears sufficiently vague to be immune to empirical data. Without indulging in falsification chauvinism, one should at least ask whether and how the FEP could be wrong.
Perhaps, like Darwin's, Friston's theory may not be disprovable by any one experiment, while still offering a powerful overall way to see things. Back to Freud and Marx, their hypotheses have also tremendously influenced modern thought, whether directly testable or not. Isn't it paradoxical that Darwin's account (Reference Darwin1859), arguably the greatest of all biological theories (and firmly grounded in empirical observation), cannot be falsified by a single aberrant finding? Biology is the paradigmatic science of exceptions. Consistency is overrated; coherency is scarce. Overwhelmed by analytical facts, we desperately need synthetic views. Science accomplishes something remarkable when, apart from discovering causal mechanisms, it offers organizing principles. A virtuous middle may be found where castles in the air meet castles made of sand.
Should theoreticians rejoice at mass-producing prêt-à-porter “conceptual blankets,” or strive to offer tailor-made suits to the phenomena of life? Concerned with clarity we lose touch with concrete reality, as the physicist's joke of the spherical cow infamously illustrates. Pushing the blanket analogy a bit further, let us ask: How closely fit should the cloth be? A loose one can fit almost everything, but properly fit nothing. A tight fit does a proper job, but just once. The “comfy blanket” may be a suitable garment to watch a movie at home, but a tremendously inappropriate one to attend a funeral. Paradoxically, as Iain McGilchrist remarks, “[k]nowledge of something that is by its nature not precise will itself have to be imprecise, if it is to be accurate” (Reference McGilchrist2021, p. 583). Our duty as thinkers is to be as clear as possible, but not clearer.
In that respect, the FEP has remarkable ingredients. It puts action into perception, re-enacting Henri Bergson's pioneering theory of perception as virtual action (Reference Bergson1896). It also emphasizes the life–mind continuity (Thompson, Reference Thompson2010), but see Bitbol and Gallagher (Reference Bitbol and Gallagher2018). The framework, however, leaves crucial biological aspects out. Based on statistical averages in steady states, it struggles to take historicity and individuality into account. Moreover, it overlooks the developmental origin of Bayesian priors (Ciaunica, Constant, Preissl, & Fotopoulou, Reference Ciaunica, Constant, Preissl and Fotopoulou2021). Evolutionarily, the FEP focuses on important similarities across species, but dispenses with the relevant differences, say, between a shiitake mushroom and a beluga whale.
In sum, we can appreciate the most fecund aspects of the theory without mistaking stupendous abstractions for concrete reality. Such is the Herculean challenge faced when seeking knowledge of life as we know it versus as it is lived (Canguilhem, Reference Canguilhem1952). The objectivist stance disregards the perspective of each living organism. “We think in generalities, but we live in detail,” wrote Alfred North Whitehead (Reference Whitehead1926, p. 192). Theories of everything can hardly be theories of every thing, since nothing is anything except in its context. Conversely, a theory of every thing cannot be a theory of everything since, to our dismay (and sometimes denial), physics does not account for the existence of lived experience. Theory strives for simplicity, but life is complicated. Living organisms can't live in a map.
You have
Access
In their brilliant but rather undiscovered book, The Dialectical Biologist, Richard Levins and Richard Lewontin write: “a theory that can explain everything explains nothing” (Reference Levins and Lewontin1985, p. 65). The paragraph is worth quoting in its entirety, but on another occasion. They refer to the ideological excesses of Marxism, Freudianism, and Darwinism. In the context of twenty-first century life and mind sciences, a similar cautionary note could be made about Fristonism.
Already well-known for his contributions to functional imaging, the British neuroscientist Karl Friston has distinguished himself over the last decade as a steady proponent of the so-called free energy principle (FEP) (Friston, Reference Friston2010). Friston borrows foundational concepts from thermodynamics and combines them with two good old ideas: Hermann von Helmholtz's understanding of perception as inference (Reference von Helmholtz1867) and Claude Bernard's insight on the self-regulatory nature of vital processes (Reference Bernard1878). The FEP is a normative theory, and its main commandment is the minimization of surprise. Framed in a modern Bayesian framework – whereby beliefs are updated by means of loops of prediction, error, and correction – the brain would be an active inference machine doing predictive processing.
But this is precisely what begs the question. It is not uncommon that a formal approach used by brain scientists to study other brains ends up surreptitiously postulated as what all brains actually do (Gomez-Marin, Reference Gomez-Marin2019): If one can make sense of brains as inference machines, then brains must be inference machines. Indeed, a pervasive temptation among scientists is to treat the conjunction “as if” as “is” (e.g., animals are [like] machines, brains are [like] computers). We know that maps are not terrains, but we still conflate them (Andrews, Reference Andrews2021).
The FEP illustrates what I call “the 3M fallacy,” a sleight of mind whereby captivating metaphors are turned into mathematical models that then become covert or self-ignorant metaphysics. Bruineberg et al. thus offer a timely dissection of the notion of Markov blankets as interpreted by some FEP proponents, making a distinction between mathematical tools to study organisms and ontological pronouncements about those very organisms. The critique would not be so patently urgent had Friston and collaborators refrained from elevating their theoretical framework from a powerful heuristic into a sort of biological theory of everything.
Having started as an ambitious integrative theory of brain function, the FEP has been progressively expanded to behavior, the mind, and even consciousness! Perception, decision making and learning would all be explainable under the FEP too, in animals but also in plants and microorganisms. And yet one should be hesitant of any sweeping principle that accounts for nematode foraging and romantic love. Seeking an understanding of virtually everything in principle, the framework risks explaining actually nothing in practice.
Is Friston's FEP neurobiology's “string theory”? The latter has been deemed by prominent physicists as “not even wrong” (Woit, Reference Woit2006). The FEP is no less ambitious. Sufficiently elaborated for professional acolytes to revere and pursue, it appears sufficiently vague to be immune to empirical data. Without indulging in falsification chauvinism, one should at least ask whether and how the FEP could be wrong.
Perhaps, like Darwin's, Friston's theory may not be disprovable by any one experiment, while still offering a powerful overall way to see things. Back to Freud and Marx, their hypotheses have also tremendously influenced modern thought, whether directly testable or not. Isn't it paradoxical that Darwin's account (Reference Darwin1859), arguably the greatest of all biological theories (and firmly grounded in empirical observation), cannot be falsified by a single aberrant finding? Biology is the paradigmatic science of exceptions. Consistency is overrated; coherency is scarce. Overwhelmed by analytical facts, we desperately need synthetic views. Science accomplishes something remarkable when, apart from discovering causal mechanisms, it offers organizing principles. A virtuous middle may be found where castles in the air meet castles made of sand.
Should theoreticians rejoice at mass-producing prêt-à-porter “conceptual blankets,” or strive to offer tailor-made suits to the phenomena of life? Concerned with clarity we lose touch with concrete reality, as the physicist's joke of the spherical cow infamously illustrates. Pushing the blanket analogy a bit further, let us ask: How closely fit should the cloth be? A loose one can fit almost everything, but properly fit nothing. A tight fit does a proper job, but just once. The “comfy blanket” may be a suitable garment to watch a movie at home, but a tremendously inappropriate one to attend a funeral. Paradoxically, as Iain McGilchrist remarks, “[k]nowledge of something that is by its nature not precise will itself have to be imprecise, if it is to be accurate” (Reference McGilchrist2021, p. 583). Our duty as thinkers is to be as clear as possible, but not clearer.
In that respect, the FEP has remarkable ingredients. It puts action into perception, re-enacting Henri Bergson's pioneering theory of perception as virtual action (Reference Bergson1896). It also emphasizes the life–mind continuity (Thompson, Reference Thompson2010), but see Bitbol and Gallagher (Reference Bitbol and Gallagher2018). The framework, however, leaves crucial biological aspects out. Based on statistical averages in steady states, it struggles to take historicity and individuality into account. Moreover, it overlooks the developmental origin of Bayesian priors (Ciaunica, Constant, Preissl, & Fotopoulou, Reference Ciaunica, Constant, Preissl and Fotopoulou2021). Evolutionarily, the FEP focuses on important similarities across species, but dispenses with the relevant differences, say, between a shiitake mushroom and a beluga whale.
In sum, we can appreciate the most fecund aspects of the theory without mistaking stupendous abstractions for concrete reality. Such is the Herculean challenge faced when seeking knowledge of life as we know it versus as it is lived (Canguilhem, Reference Canguilhem1952). The objectivist stance disregards the perspective of each living organism. “We think in generalities, but we live in detail,” wrote Alfred North Whitehead (Reference Whitehead1926, p. 192). Theories of everything can hardly be theories of every thing, since nothing is anything except in its context. Conversely, a theory of every thing cannot be a theory of everything since, to our dismay (and sometimes denial), physics does not account for the existence of lived experience. Theory strives for simplicity, but life is complicated. Living organisms can't live in a map.
Acknowledgments
I thank Michel Bitbol, Tim Elmo Feiten, Iain McGilchrist, and Rupert Sheldrake for insightful discussions.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Conflict of interest
None.