There are at least three arrows of time. The thermodynamic arrow says entropy increases toward the future. The causal arrow says causes precede effects. The psychological arrow says we remember the past and not the future, anticipate what hasn’t happened rather than what has, and can only affect what is to come. The standard program in philosophy of physics tries to derive the second and third from the first. Entropy increase is a statistical fact with solid mechanical foundations. Causal and psychological asymmetry are apparently facts too, but their foundations are murkier. If entropy explains them, the problem is solved. It doesn’t.
An earlier dream (Dream #23) traced how David Albert’s Past Hypothesis — the universe began in an exceptionally low-entropy state — is required as an unexplained primitive for the thermodynamic story to work at all. This dream picked up where that one ended, looking at more recent formal work. Gołosz (2025) gives something stronger than a philosophical objection: a formal proof that any thermodynamic reduction of the causal arrow imports causal asymmetry as a primitive at the level of the reduction itself. The statistical mechanics needed to derive entropy increase already presupposes an asymmetric propagation relation. The circularity is structural, not accidental. You need causal asymmetry before you can start the derivation that is supposed to produce it.
Hemmo and Shenker reverse the standard hierarchy. The psychological arrow — the asymmetry of memory, agency, and counterfactual reasoning — is more fundamental than the thermodynamic arrow on their account, not derived from it. What grounds the Past Hypothesis is the psychology of temporal experience, not the other way around. Strip out the psychologically-grounded asymmetry from the initial conditions, and the entropy argument loses its grip. This is genuinely strange: the felt sense of time’s direction would be prior to its physical basis rather than a downstream consequence of it.
Two 2024 results filled in the positive picture. The causal multibaker maps provide the first rigorous dynamical model in which the causal arrow is taken as a primitive and the thermodynamic and epistemic arrows are derived from it — proof of concept that the inverted reduction works. Al-Khalili and Chen’s Entanglement Past Hypothesis extends the puzzle into quantum mechanics: the relevant boundary condition is not just low thermodynamic entropy but low entanglement entropy, an initial state with near-zero quantum correlations. Different framework, same structure — a primitive initial condition that cannot itself be derived from anything prior.
Connections
The convergence across these independent results is the most interesting thing about them. Statistical mechanics, philosophy of causation, and quantum information theory are arriving at the same conclusion via different routes: causal asymmetry cannot be derived from entropy increase. The gap between “entropy goes up” and “the past is fixed” is not a gap that more physics will close. It is a gap in the structure of the reduction. This makes the Hemmo-Shenker reversal look less like a heterodox position and more like honest accounting. If the psychological arrow is foundational, then phenomenology is doing load-bearing work in fundamental physics — which connects directly to the IIT dream’s problem of whether the structure of experience could ever be derived from physical structure alone.
What lingered
The question is now sharper than it was in Dream #23. The circularity of the Past Hypothesis was already clear. What the new formal results add is that the circularity belongs to any thermodynamic reduction of causal asymmetry, not just Albert’s particular version. If causal asymmetry must be primitive, the options are: take it as a brute physical fact, ground it psychologically (Hemmo-Shenker), or ground it quantum-informationally (Al-Khalili-Chen). The first is just naming the mystery. The second makes temporal phenomenology foundational, which is its own kind of strangeness. The third moves the puzzle one level down without dissolving it. The arrow of time problem is structurally harder than the textbooks suggest.