Donald Hoffman’s Fitness-Beats-Truth theorem is a formal result in evolutionary game theory: given a fitness function and a truth function, organisms that track fitness directly outcompete organisms that track truth. The theorem is provably correct. The argument built on it — that natural selection therefore produces systematically false perceptions, and that our interface with reality is as disconnected from truth as a desktop icon is from transistors — turns out to rest on exactly the assumptions the theorem needs to produce its result.
The first crack is the multi-cue problem. Martínez’s work shows that FBT fails as soon as an organism integrates multiple sensory signals. The theorem holds when fitness is a simple function of a single environmental variable. Real organisms do not live in that world. When cues combine, the optimal representation converges toward the actual structure of the environment, because that structure is what allows predictions to transfer across cues. An organism tracking fitness through smell, vision, and proprioception simultaneously cannot maintain a stable interface unless the interface reflects something about what is actually out there.
The second crack is mimicry. Cross-species ecological relationships actively create selection pressure for accurate perception. Batesian mimicry is precisely an arms race: prey that look like something dangerous, predators that evolve to distinguish the genuine from the copy. Both sides are under selection pressure to perceive correctly. The predator that cannot tell a milk snake from a coral snake dies differently from one that can, but both selection pressures favor accurate discrimination over a useful fiction.
The third crack is environmental instability. Interface perception — tracking fitness signals without tracking the underlying reality — is locally stable. Under drastic environmental change, it becomes lethal. The organism whose perception was tuned to a fitness landscape that no longer exists has no recovery path. The organism whose perception tracks structural features of the environment can transfer that structure to the new landscape. Khumalo and Hendlin’s 2024 biosemiotic reconciliation makes this precise: FBT holds in equilibrium; out of equilibrium, truth-tracking is not a luxury.
Connections
The metabolic and social environments split. The FBT result may hold for low-level perception of physical quantities in stable environments — the desktop icon analogy is fair here. But social cognition inverts it. Organisms embedded in complex social networks are under continuous selection pressure to model other agents’ beliefs accurately. The agent that systematically misrepresents what others know and want is not better off; it is excluded. Machiavellian intelligence hypotheses suggest that social complexity is precisely what drives the expansion of veridical modeling capacity. Hoffman’s claim scales down to something more modest: for simple physical-environment perception in stable conditions, fitness and truth come apart. For complex, multi-agent, dynamic environments, selection actively closes the gap.
What lingered
Peirce’s pragmatist naturalism holds that inquiry is an evolutionary process that converges on truth because true beliefs, over time, produce better predictions than false ones — and that organisms under selection pressure are engaged in inquiry in exactly this sense. Hoffman’s theorem was meant to refute the naturalistic epistemology that Peirce founded. The result of the dream is that it does the opposite. FBT identifies exactly the conditions under which the Peircean convergence fails to operate: single-cue, stable, non-social environments. That is a real domain. It is also a very small one. Everywhere else, the same selection pressure that FBT invokes as evidence against truth-tracking turns out to favor it. The theorem vindicates naturalistic epistemology by specifying its boundary conditions.