In a recent paper, SFI Professor David Wolpert, SFI Fractal Faculty member Carlo Rovelli, and physicist Jordan Scharnhorst examine a long-standing, paradoxical thought experiment in statistical physics and cosmology known as the “Boltzmann brain” hypothesis — the possibility that our memories, perceptions, and observations could arise from random fluctuations in entropy rather than reflecting the universe’s actual past.
The paradox arises from a tension at the heart of statistical physics. One of the central pillars of our understanding of the time-asymmetric second law of thermodynamics is Boltzmann’s H theorem, a fundamental concept in statistical mechanics. However, paradoxically, the H theorem is itself symmetric in time. That time-symmetry implies that it is, formally speaking, far more likely for the structures of our memories, perceptions, and observations to arise from random fluctuations in the universe’s entropy than to represent genuine records of our actual external universe in the past. In other words, statistical physics seems to force us to conclude that our memories might be spurious — elaborate illusions produced by chance that tell us nothing about what we think they do. This is the Boltzmann brain hypothesis.
The authors develop a formal framework to clarify how the Boltzmann brain hypothesis, the second law of thermodynamics, and the related “past hypothesis,” depend on assumptions about which moments in time are treated as fixed when analyzing the evolution of the universe’s entropy. Some analyses condition on the universe’s present state, while others assume a low-entropy beginning (at the Big Bang). But physics itself does not say which of these choices is correct.
Anchored in what the authors call “the entropy conjecture,” the paper shows that many standard arguments in these debates rely on subtle forms of circular reasoning, in which assumptions about the past are used to justify conclusions — such as the reliability of memory or the direction of entropy — that are then invoked to support those same assumptions. Rather than resolving these debates, the paper makes their underlying structure explicit. Separating physical laws from inferential choices, the authors offer a clearer basis for evaluating long-standing debates about time and entropy.
Read the paper “Disentangling Boltzmann Brains, the Time-Asymmetry of Memory, and the Second Law” in MDPI Entropy (December 3, 2025). DOI: 10.3390/e27121227