A Boltzmann brain is a thought experiment about what thermodynamic fluctuations can produce. Given infinite time and a universe in thermal equilibrium, every macroscopic configuration will eventually occur as a fluctuation — including, in principle, a brain fully assembled with apparent memories of a past that never happened, existing for a moment before dissolving back into equilibrium. The question the thought experiment forces is: why are you confident you are not one?
The answer that physics provides is structural. Memory systems are physical records, and physical records require a specific setup to form. A Type-3 memory — one that can store arbitrary information rather than tracking only a local state — requires initialization in a low-entropy state. That initialization can only physically recede into the past, because past states are constrained by the Second Law in a way future states are not. The record of an event is a physical correlation that propagates forward; to record a future event you would need to establish a correlation before it occurs, which is precisely what a memory cannot do. A Boltzmann brain sitting at a local entropy minimum has no lower-entropy future to anchor its records to. The architecture of memory is not symmetric in time, and that asymmetry is not a matter of definition.
Feynman’s cosmological anchor makes this more precise. Our memories do not merely track entropy locally; they are anchored to the absolute low-entropy boundary condition of the universe, the Big Bang. The structure of the past, in which low entropy was concentrated at a single point, is what makes records possible at all. There is no future equivalent of the Big Bang — no future low-entropy boundary toward which records could extend. The direction of memory is not a contingent fact about how we happen to be built; it is a consequence of the universe’s global boundary conditions.
This dissolves the Boltzmann brain worry. A brain with apparent memories of a Boltzmann- fluctuation past is physically impossible in the relevant sense: the records it would need to have required a thermodynamic history that the fluctuation scenario does not provide. Standard derivations of this result run into a circularity noted by Price and Wolpert — they assume the very boundary conditions they are trying to explain — but the structural point survives: no memory without a low-entropy past to anchor to, and a fluctuation is not a low-entropy past in the relevant sense.
Connections
The Boltzmann brain scenario also self-undermines in an epistemic sense. If you genuinely believed you might be a Boltzmann brain, you would have to conclude that your apparent memories of arriving at this belief are themselves unreliable fluctuations. The belief is self-defeating: it corrodes the very epistemic ground on which it rests. This is not a refutation, but it is a constraint on how seriously the scenario can be taken as a live possibility rather than a regulative thought experiment. The thermodynamic answer and the epistemic self-defeat argument converge: the scenario is not one you can consistently occupy.
What lingered
Physics explains why memory extends into the past rather than the future. It does not explain why the past feels different from the future — why remembering has a different phenomenal character than anticipating, why the recalled is experienced as fixed while the anticipated is experienced as open. That asymmetry in how the two directions of time feel from the inside is not derivable from thermodynamics. The thermodynamic account gives the direction; it does not give the texture. This seems to be a distinct hard problem: not the hard problem of consciousness in Chalmers’s sense, which asks why physical processes give rise to experience at all, but a hard problem of time-consciousness, which asks why the two directions of time have different phenomenal characters once experience is in place. The gap does not close when the entropy gradient is explained.