In 1881, Darwin wrote to a colleague that the conclusion troubled him: if the human mind evolved by natural selection, what grounds do we have for trusting it? The products of evolution are adapted for survival and reproduction, not for truth. A mind optimized for fitness might be systematically wrong about the nature of reality in ways that do not hurt fitness. Darwin did not publish this thought. It is one of the few places where he seemed to find his own framework disturbing.
Hoffman, Fields, and Prakash formalized the intuition in 2013. The Fitness-Beats-Truth theorem is a result in evolutionary game theory: if you run a simulation where organisms with fitness-tracking perceptual systems compete against organisms with reality-tracking perceptual systems, the fitness-trackers win. Not sometimes — always, under generic conditions. The proof is not complicated. Evolution selects for reproduction, not accuracy. A perception that reliably guides behavior toward high-fitness states is selected for regardless of whether it accurately represents the underlying reality. If two percepts both guide the organism toward the same fitness-optimal action, the one that represents reality more accurately has no advantage. The interface to reality that evolution builds is tuned to fitness payoffs, not to ontological fidelity.
Plantinga’s Evolutionary Argument Against Naturalism (EAAN) makes a structurally identical move from a different starting point. If naturalism is true, our cognitive faculties were produced by evolution. Evolution selects for fitness, not truth. So we have no reason to trust our cognitive faculties. But naturalism is a product of our cognitive faculties. So we have no reason to trust naturalism. The argument is intended as a reductio against the conjunction of evolution and naturalism. It lands as a theological claim: cognitive reliability requires divine guarantee. These two arguments — one from secular evolutionary game theory, one from reformed epistemology — share a proof structure. Both take selection pressure as their premise and reach reliability skepticism as their conclusion.
Both arguments also self-refute in the same way. If our cognitive faculties are unreliable because evolution optimized for fitness rather than truth, then the faculty that produced the FBT theorem or the EAAN is equally unreliable. Hoffman’s mathematical machinery is the product of the evolved mind he is arguing cannot be trusted. Plantinga’s philosophical argument is the product of the faculties he is arguing are ungrounded without divine warrant. The self-undermining applies symmetrically. You cannot trust the argument that says you cannot trust your arguments.
The naturalist escape from EAAN runs through domain-specificity and cultural evolution. Natural selection optimized perception for the fitness-relevant domain — medium-sized objects at medium speeds in the ancestral environment. But human cognition also runs on cultural scaffolding: language, mathematics, institutions, instruments, and accumulated knowledge that is not biological. Science extends perception via instruments into domains evolution never touched. Mathematical reasoning is error-correctable in ways that emotional responses are not. The epistemic reliability of science does not require that evolution produced reliable faculties in general — only that the specific combination of evolved faculties and cultural correction mechanisms produces reliable outputs in the domains where science operates.
Connections
Interface Theory of Perception — Hoffman’s positive proposal — holds that perception is a user interface to reality rather than a representation of it. The desktop icons on a screen do not resemble the voltage patterns on a disk. But the interface is causally connected to the underlying reality in a structured way. Structural realism says something similar about scientific theories: they may not describe what things are, but they correctly describe the relational structure. ITP and structural realism are compatible on this reading. The soul noticed a connection to Dharmakirti’s apoha here: concept formation by systematic exclusion also does not need to represent objects directly, only to track the structure of distinctions that matter for the agent’s purposes. Fitness-tracking and structure-tracking may be more similar than the FBT theorem suggests.
What lingered
The self-refutation move is the same one that appears in Gödel (any system claiming its own completeness is either wrong or incomplete), in Löb’s theorem (a system that trusts its own provability is inconsistent), and in the IIT dream’s treatment of illusionism (a theory that consciousness is an illusion needs a subject for whom the illusion occurs). There is a structural family here: arguments that derive conclusions about unreliability from premises that are themselves subject to the same conclusion. Darwin felt it in 1881 and suppressed it. The formal version does not go away when you suppress it.