# The Node We Call Now

*Suppose only one universe exists at a time, and every future is drawn from a probabilistic tree growing out of the present, the way a learning agent’s next state follows from its last. Quantum physics has a surprising amount to say about that picture—where it holds, where it breaks, and whether thinking in branches can change which branch comes.*

Physics · September 2026 · 18 min · by Epimystic
Canonical: https://epimystic.com/essays/the-node-we-call-now/
Topics: quantum, time, metaphysics, the future, physics

Here is a picture worth taking seriously. Only one universe exists at a time: this state, now. The next state is not fixed but drawn from a probability distribution that depends on this one, the way a reinforcement-learning agent’s next situation follows from its present one and the action its policy picks. Draw it and you get a single point, the present, with a tree of possible universes branching out of it, each edge carrying a weight. Two questions follow. What would retrocausal and multiversal thinking look like to someone inside such a universe? And does thinking that way change the probabilities on the tree?

The picture deserves to be sorted into established physics, interpretation and its own speculation, written down exactly, and tested. It survives, in a more interesting shape than the one it started with.

## What the Equation Actually Says

Start with what nobody disputes. A quantum system is described by a state, ψ—a list of complex numbers called amplitudes, one for each way the system could turn out if you looked. It changes according to the Schrödinger equation, iħ ∂ψ/∂t = Hψ, where the Hamiltonian H encodes the system’s energy and interactions. Two properties matter most here. It is deterministic: give it the state now and every later state is fixed, with no dice anywhere. And it is linear: any sum of solutions is a solution.

Probability enters at one point only. When a measurement is made, the chance of outcome i is |cᵢ|², the squared size of that outcome’s amplitude. This is the Born rule, added by Max Born in 1926 as a note to his own paper, and it has never failed a test. Between the smooth deterministic evolution and the abrupt probabilistic outcome sits a seam a century of work has not closed. That seam is the measurement problem, and [the history of a theory that works without anyone agreeing why](https://epimystic.com/essays/nobody-understands-it/) is largely a history of attempts to stitch it.

One misreading has to go before the tree can be drawn honestly. Superposition is not, by itself, many worlds. An electron whose spin points along x is an equal superposition of up and down along z—and a perfectly definite state along x. Whether a state counts as a superposition depends on the question you ask of it; the electron is not living two lives, as [it has complained at length](https://epimystic.com/essays/stop-drawing-me-as-a-little-ball/). Anything deserving the word ‘worlds’ appears only when a superposition spreads into a device, its surroundings and the person reading it—and what that means is where the interpretations begin.

## A Tree That Is Not Yet a Tree

Now write the picture down. A Markov decision process has states s, actions a, transition probabilities P(s′|s, a) for arriving in s′ from s after doing a, a reward R that scores what happened, and a policy π(a|s), the agent’s rule for choosing. Take the agent away and you have a plain Markov chain, P(s′|s): the universe rolling forward by itself. The probability of a complete history is the product of the transition probabilities along its path. That is the tree: a node, weighted edges, and leaves whose weights are the products and add to one.

Where the analogy holds, it holds deeply. The defining feature of a Markov process is that the present screens off the past: once you know the state now, how it got here tells you nothing more about what comes next. Quantum mechanics has exactly this property. The Schrödinger equation is first order in time, so the present state is a complete specification, and the past matters only in so far as it has left traces in it.[^1] The greyed-out stretch behind the node is not a simplification. It is how the physics works.

Where it fails, it fails at the root. Classical probabilities add; quantum amplitudes interfere. Send a photon into a Mach-Zehnder interferometer: a half-silvered mirror splits it into two routes, a second one recombines them in front of two detectors. A probability tree says each detector should fire on half the photons. In a balanced interferometer one fires on all of them and the other never does. Along the two routes to the dark detector the amplitudes are equal and opposite, and they cancel. A quantum branching is a tree of complex numbers whose leaves can annihilate one another. Block one route and the cancellation goes; the odds become a coin toss’s.

What turns the complex tree into a probability tree is decoherence. When two branches leave different marks in their surroundings—a scattered photon, a detector’s click—the cross-terms carrying the interference spread into the environment beyond practical recovery, and from then on the weights multiply and add as the picture says. For anything macroscopic this happens absurdly fast. Decoherence does not say which branch you end up in. It says why there are branches to end up in.

*Figure: the-branching-now — The present as a node. The past is screened off; the first split is still made of amplitudes, whose cross-term can interfere; past the decoherence line, weights multiply along each path and the leaves sum to one.* (drawn in the essay: https://epimystic.com/essays/the-node-we-call-now/)

A formalism says exactly this, and it is the closest thing in physics to the branching diagram. The consistent or decoherent histories approach, founded by Robert Griffiths in 1984 and developed by Roland Omnès and by Murray Gell-Mann and James Hartle, treats a history as a sequence of properties at successive times. A family of histories is consistent when the interference between any two distinct members vanishes; then, and only then, their weights obey the ordinary rules of probability.[^2] Quantum mechanics grows its own probability tree—coarse-grained, and only where decoherence has done its work. The picture is correct precisely to the extent that the branches decohere.

## Five Ways to Read the Tree

How many branches are real, and whether the growth is chance or clockwork, is the business of interpretation; the predictions are shared, the pictures are not. The Copenhagen view, in its working form, keeps one world and prunes the tree at every measurement. It is the nearest to ‘one universe at a time’, but it declines to say what the universe does between measurements, or where the pruning happens.

Hugh Everett, in 1957, removed the pruning. Nothing collapses, so every branch is real. Decoherence supplies what his theory needed: once branches stop interfering, each contains observers who see one definite outcome. Branching here is not an event or a choice. It describes how one deterministic wavefunction comes to be structured as many nearly independent parts.

Ghirardi, Rimini and Weber, in 1986, changed the equation instead. Every particle, at random, suffers a spontaneous localisation—about once in three hundred million years for a single particle, pinning it to a ten-thousandth of a millimetre. For a pointer of 10²³ particles, one such hit drags the whole superposition with it in well under a microsecond. One world, genuinely stochastic, pruned by nature rather than by observers—and the only reading here that predicts something different, since collapse would faintly heat matter. Underground experiments have excluded parts of its parameter range without yet seeing a collapse.

David Bohm, in 1952, reviving an idea of de Broglie’s, gave particles definite positions at all times, steered by a wavefunction that never collapses. One actual world, fully deterministic; the other branches of the wave persist but are empty, and the price is explicit non-locality. And QBism, the view of Christopher Fuchs, Rüdiger Schack and David Mermin, relocates the whole tree: the quantum state is an agent’s degrees of belief, and the Born rule is a consistency requirement on its bets. The probabilities on the edges were never out in the world. They are yours.

## Is Anyone Maximising Anything?

The branching model has a reward function. Fundamental physics has none; nothing in the Schrödinger equation chooses. Yet the resemblance unsettles. Classical mechanics can be stated as a principle: of all the paths between two configurations, a system takes the one for which the action—the time-integral of kinetic minus potential energy—is stationary. Maupertuis, stating an early form in 1744, read it as divine economy. It looks like optimisation, though strictly it is stationary rather than least; the true path can sit at a saddle.

The teleology dissolves in Feynman’s version of quantum mechanics. In 1948 he showed that the amplitude to go from one event to another is a sum over every possible path, each contributing a term of the same size whose phase is its action divided by ħ. Near the classical path neighbouring paths share nearly the same phase and pile up; far from it the phases spin and cancel. The classical trajectory is not chosen. It is what survives the cancellation. This is an exact sense in which ‘all branches are computed’—and it is interference, not a reward, that decides what we see.

The kinship with reinforcement learning is not a pun, though. The Bellman equation of dynamic programming becomes, in continuous time, the Hamilton-Jacobi-Bellman equation of optimal control, and the Hamilton-Jacobi equation of mechanics is its sibling, with the action in the role of the value function. Bert Kappen showed in 2005 that a class of noisy control problems can be solved exactly by path integrals. [The learning machines](https://epimystic.com/essays/the-bitter-lesson-tasted-twice/) share real structure with mechanics—sums and extremes over paths—but that does not put a goal in the universe.

Where reward genuinely lives is inside organisms. Karl Friston’s free-energy principle proposes that living systems act to minimise a bound on their surprise about their own sensations—an objective pursued through action, very like a reward. As a theory of brains it is contested. As a map of the branching model it is right about location: the objective sits in agents, and the agents sit in the world.

*Figure: the-policy-and-the-wave — Two loops. On the left, an agent: state, policy, action, transition, with a reward reshaping the policy. On the right, the world by itself: state, unitary evolution, decoherence, Born weights. The right-hand loop has no reward term.* (drawn in the essay: https://epimystic.com/essays/the-node-we-call-now/)

## One Universe at a Time

‘Only one universe exists at a time’ has a name in philosophy: presentism, the view that only the present is real. Relativity presses hard on it, because there is no single frame-independent present for it to be. [An earlier essay here](https://epimystic.com/essays/does-time-flow-or-do-we/) followed that pressure all the way to the block universe, in which past, present and future are equally real and nothing moves; the argument need not be repeated.

Whether relativity actually forces the block is a narrower question, and [it turns on Howard Stein’s reply to Putnam](https://epimystic.com/essays/the-theory-of-invariants-that-was-named-after-its-shadows/): relativity forbids a universal now without settling whether the future is already there. That leaves room for C. D. Broad’s growing block of 1923, in which past and present are real and the future is not. George Ellis has given it a physical engine: in his evolving block universe, spacetime grows as quantum outcomes become definite; the present is the edge where the indeterminate becomes determinate. To define the edge he uses the proper time of matter moving with the cosmic expansion, which critics say smuggles back a preferred frame. It is still the closest published relative of the branching picture: one state at a time, each new one settled by quantum chance.

Logic got there first. Arthur Prior, working on Aristotle’s puzzle of whether ‘there will be a sea battle tomorrow’ is already true, built tense logics on branching time: moments in a tree, the past a single line, the future a fan of continuations. Nuel Belnap later made the branching compatible with relativity. A linear past behind a branching future is the branching diagram exactly, and it has been a rigorous object since the 1960s.

Time itself may be less basic than the picture assumes. In 1983 Don Page and William Wootters showed that a universe whose total state never changes—as the Wheeler-DeWitt equation of quantum gravity seems to demand—can still contain change. Condition the rest on what a clock inside it reads, and the rest evolves by the Schrödinger equation. Time becomes a correlation between a clock and everything else. Time that loops is stranger still: general relativity permits closed timelike curves in principle, and Deutsch’s 1991 quantum model of them demands only self-consistency—a tree obliged to agree with its own roots. None has been observed.

## The Future Reaching Back

Retrocausality usually enters through John Wheeler’s delayed-choice experiment, proposed in 1978. Decide whether to learn a photon’s route through an interferometer or let the routes interfere—but decide after it has passed the first splitter. It behaves according to the later choice. In 2007 a group near Paris did it with single photons and a random-number generator far enough away that no signal could have warned the photon. Wheeler declined to say the past had been changed.

> no elementary phenomenon is a phenomenon until it is a registered (observed) phenomenon
>
> — —John Archibald Wheeler, Law Without Law, 1983

The quantum eraser sharpens it. In the version run by Yoon-Ho Kim and colleagues, published in 2000, a laser lights two regions of a crystal, A and B, either of which can emit an entangled pair of photons. The signal photon travels a short way to a scanning detector, D0. Its twin, the idler, travels about two and a half metres further and arrives some eight nanoseconds later at one of four detectors.[^3] D3 and D4 can be reached only from A or only from B, so a click there records which region the pair came from. D1 and D2 sit behind a beam splitter that mixes the routes, so a click there records nothing about the path.

*Figure: the-eraser-timing — The delayed-choice quantum eraser, schematically. The signal photon reaches D0 first; its idler reaches D1 to D4 about eight nanoseconds later. Sorted by which idler detector fired, D0’s featureless hump splits into fringes and anti-fringes.* (drawn in the essay: https://epimystic.com/essays/the-node-we-call-now/)

What the experiment shows is that the correlation between the photons depends on what is done to the later one. It shows no signal to the past. D0’s record, taken alone, is the same featureless hump whatever the idlers do. The fringes exist only in subsets of the data, picked out by coincidence counting once the idler record has arrived at ordinary speed. The eraser erases nothing that was ever written at D0. It sorts.

Standard quantum mechanics needs no retrocausation to explain this. Some serious interpretations use it anyway, because the laws themselves prefer no direction of time. In 1964 Yakir Aharonov, Peter Bergmann and Joel Lebowitz showed that for a system prepared in one state and later found in another, the probabilities of results in between are time-symmetric. Aharonov built this into the two-state vector formalism, where the complete description at a moment is a pair: a state arriving from the past and one arriving from the future. Its ‘weak values’ can lie far outside the ordinary range—a spin component of 100 for a particle whose spin can only be plus or minus a half.[^4]

John Cramer’s transactional interpretation, from 1986, makes each quantum event a handshake between a wave sent forward and a wave returned backward. Huw Price and Ken Wharton go further. Bell’s theorem assumes that a particle’s hidden state is independent of the measurement it will later meet. Drop that—let the state depend on its future measurement, through influences running back along one light cone and forward along another—and the Bell correlations can be reproduced without action at a distance. Matthew Leifer and Matthew Pusey proved in 2017 that, under plausible assumptions, a time-symmetric realist quantum theory must be retrocausal. These remain minority views, none yet as complete as Bohm’s or Everett’s.

For the branching picture the lesson is precise. Retrocausality is the failure of the Markov screen. In a retrocausal world the present node cannot fix the odds of what comes next without a boundary condition from the future, and the tree becomes a web pinned at both ends. A universe in which each state suffices for the next is exactly a universe without retrocausation. The two ideas one might hope to join are here in tension, and the tension is informative.

> **Retrocausality is not the future sending messages. It is the present failing to be the whole story.**

## The Observer in the Tree

What would multiversal thinking look like from inside? First: what can probability mean if every outcome happens? David Deutsch in 1999 and David Wallace after him argued that a rational agent who believes the wavefunction is all there is, and who meets a few modest principles of decision theory, must weight branches by their Born weights. Others locate the probability in self-locating uncertainty: after the branching and before you look, you do not know which branch you are in. Both derivations are contested; both conclude that [the mathematics of choosing under uncertainty](https://epimystic.com/essays/the-mathematics-of-the-frightened-primate/) is unchanged, and you should bet as a single-world believer would.

Jorge Luis Borges had the image sixteen years before Everett had the physics, in a story about a labyrinth that is a novel in which every choice is taken.

> Time forks perpetually toward innumerable futures.
>
> — —Jorge Luis Borges, The Garden of Forking Paths, 1941

The darker corollary is quantum immortality: if every branch is real, then in some branch you survive every danger, and you can only find yourself in those. It fails on its own terms: the surviving branches carry ever smaller weight, and the survivor is not spared but merely lasts, through ever less likely rescues. And if Deutsch and Wallace are right, a rational agent cares about weight, not mere existence, so a many-worlds believer has no licence for recklessness. Retrocausal thinking is more modest still: in none of these models can an agent message its own past, since whatever runs backward hides in correlations. Believing in it changes what counts as an explanation, not what you can do.

Which brings us to the central question: does thinking this way change the probability matrix? The answer has three layers. The first is that in standard quantum mechanics beliefs do not touch amplitudes; the Hamiltonian has no term for what anyone thinks. The idea that consciousness collapses the wavefunction, associated with London and Bauer and with Eugene Wigner, who later abandoned it, remains a minority view, and the brain is a poor candidate for delicate quantum influence: its relevant states decohere many orders of magnitude faster than neurons act.[^5] Nor is there evidence that attention biases chance. Princeton’s PEAR laboratory reported tiny deviations in random-number generators under human intention; a three-laboratory replication in 2000 failed to find them, and a 2006 meta-analysis found an effect small enough to be explained by selective publication.

The second layer reverses the first. An observer is a physical system, and its actions are part of the dynamics. In the decision-process picture the transition the world actually follows is P(s′|s) = Σₐ π(a|s) P(s′|s, a): the policy sits inside the sum. An agent that models the branches—runs futures in imagination, notices which lead where, and acts on it—has a different policy, and so a different distribution over what happens next. This is ordinary causation through action, and it is enormous. To call it mysterious is to [confuse the agent’s map with the territory’s amplitudes](https://epimystic.com/essays/the-map-that-insists-it-is-the-territory/); to call it trivial is to forget that most of the future that matters to us is made this way.

> **Thinking changes the odds the way anything does: by moving matter. That is not mind over matter. It is mind as matter.**

The third layer belongs to QBism. If the probabilities are the agent’s own, updating them is the whole story: an agent that revises its expectations has, by definition, changed its matrix. QBists add that the world’s answer to an action is not up to the agent. Only the bets are.

## The Whole Picture

Put the pieces together. Established: a state that evolves deterministically, screening off its own past; probabilities that enter only where records are made; and decoherence, which turns the branching of amplitudes into a coarse-grained tree whose weights multiply and add. Interpretation: whether one branch is real or all, whether the pruning is physics, and whether the probabilities belong to the world or to whoever bets. The branching model—one world at a time, stochastic transitions over a decoherent tree, agents whose policies weight them—is a coherent position, with a place on the map beside the collapse theories and Ellis’s growing block.

It needs no reward in the laws, because agents supply their own. If its pruning is objective, collapse experiments may one day see it. And it must live with two neighbours it cannot fully accept: a relativity with no universal now, which a one-state-at-a-time universe needs, and the retrocausal models, which deny that the present is ever the whole story.

*Figure: the-map-of-worlds — The interpretations on two axes: how many worlds are real, and whether the dynamics is deterministic or stochastic. The branching model sits with the collapse theories. Retrocausal readings carry a backward loop; QBism sits on the crossing.* (drawn in the essay: https://epimystic.com/essays/the-node-we-call-now/)

So the node we call now is real in a stronger sense than the picture needed. It is where amplitudes become records, where the complex tree becomes a probability tree, and where agents, themselves parts of the state, choose the actions that tilt the next set of weights. Nothing in the physics lets a mind reach into the equation. Everything in it lets a mind reach into the world, which is the only place the equation was ever being solved.


---

[^1]: A small part of a larger system can behave non-Markovianly—its future depending on its own past—because the memory is stored in its surroundings. The whole, system plus environment, is Markovian again. The screen holds for the universe even when it leaks for a piece of it.
[^2]: For histories α and β built from projections at successive times, the decoherence functional is D(α, β) = Tr(C_α ρ C_β†). When its off-diagonal entries vanish the diagonal ones, D(α, α), are probabilities that obey the ordinary sum rules. Griffiths (in Further reading) develops it with examples.
[^3]: In Kim and colleagues’ apparatus the idler’s optical path is about 2.5 metres longer than the signal’s, so any information from the idler arrives at least eight nanoseconds after the signal photon has been recorded—long against the detectors’ one-nanosecond response.
[^4]: The paper is Aharonov, Albert and Vaidman, 1988: a weak measurement of a spin component, suitably pre- and post-selected, returns an average of 100 for a quantity whose ordinary eigenvalues are ±½. The number is an average over many weak, noisy readings, not a single impossible outcome.
[^5]: Tegmark’s estimates, in Physical Review E in 2000, put the decoherence of neural firing states at between 10⁻¹³ and 10⁻²⁰ seconds, against neural dynamics of 10⁻³ to 10⁻¹ seconds. By the time a neuron has done anything, the superposition has long since become a mixture.

## Further reading

- **The Emergent Multiverse** — David Wallace. The most careful modern case for Everett, including the decision-theoretic derivation of the Born rule this essay leans on. Dense but honest about what the argument needs, and the place to test whether ‘every branch is real’ can be made to mean something exact.
- **Consistent Quantum Theory** — Robert B. Griffiths. The textbook of the histories approach by the physicist who founded it. It shows, with worked examples, exactly when a family of quantum histories can be given probabilities that add like a tree—the formal version of the picture this essay starts from.
- **Time’s Arrow and Archimedes’ Point** — Huw Price. A philosopher’s argument that our intuitions about cause and time are built on an asymmetry the laws do not contain, and that a time-symmetric quantum world may need influences running backward. The book that made retrocausality respectable enough to argue with.
- **The Evolving Block Universe and the Meshing Together of Times** — George F. R. Ellis. Annals of the New York Academy of Sciences, 2014. A cosmologist’s proposal that spacetime grows as quantum outcomes become definite, with the present as the growing edge—the nearest published relative of a universe that exists one state at a time.
- **Delayed ‘Choice’ Quantum Eraser** — Yoon-Ho Kim, Rong Yu, Sergei Kulik, Yanhua Shih and Marlan Scully. Physical Review Letters, 2000. Four pages, and the most misquoted experiment in popular physics. Read the figure captions and the coincidence plots before any account of what the future supposedly did to the past.
- **Reinforcement Learning: An Introduction** — Richard S. Sutton and Andrew G. Barto. The standard account of states, actions, policies, rewards and the Bellman equation. Its first chapters are the cleanest statement of the machinery that the universe-as-decision-process picture borrows, and so of exactly what it would be claiming.