What Does Bell's Theorem Really Tell Us About Reality?

In a basement laboratory in Delft in 2015, physicists fired photons at electrons trapped in diamonds 1.3 kilometers apart, timing everything so precisely that no signal — not even one traveling at light speed — could get from one measurement to the other before both were finished. The result matched what quantum mechanics had predicted for eighty years and what a certain kind of common sense had always insisted must be false. Two particles, separated by more than a kilometer of Dutch countryside, behaved as though each somehow knew, instantaneously, what was happening to its partner.

This was not the first such experiment, nor the last. By now Bell tests have been run with photons, atoms, ions, and superconducting circuits, closing loophole after loophole, and the verdict has never wavered. Yet ask ten physicists what the result means, and you risk a decade-long argument. Everyone agrees the data rule something out. Almost no one agrees on what to conclude instead. That gap — between an experimentally airtight theorem and a metaphysically wide-open aftermath — is one of the strangest situations in all of science, and it is nowhere near closed.

svg[The CHSH experiment: entangled particles, independent choices, correlations that outrun any local hidden-variable theory]

By now these experiments have closed the major "loopholes" that skeptics once used to argue the violations might have mundane explanations. The locality loophole — the worry that a signal might sneak between detectors before measurements complete — was closed by ensuring the choice of settings and the detection events are space-like separated, as in Aspect's later experiments and definitively in the 2015 tests. The detection loophole — the worry that only a biased subsample of particles gets detected — was closed using highly efficient detectors, notably in experiments with electron spins in diamonds (Delft, 2015) and with atoms and photons (NIST, Vienna, 2015). The freedom-of-choice loophole, addressed by generating random settings from sources as independent as possible from the particle source — including, in a striking 2018 experiment, from the light of distant quasars, so that any conspiracy would have to have been arranged billions of years ago — has been narrowed dramatically, though, as we'll see, never in principle fully closed. Local hidden-variable theories, in the straightforward Bell sense, are dead. That is settled science.

Here is where the settled science ends and the real argument begins. Bell's derivation invokes several assumptions at once, and violating the inequality only tells you their conjunction fails — not which conjunct to reject. Three main candidates have dominated the debate for sixty years, and each has serious, sophisticated defenders.

Give up locality. This is the most common textbook gloss: Bell violations show the world is "nonlocal." On this reading, measuring particle A really does instantaneously affect the state of distant particle B, in some sense that outruns the speed of light, even though quantum mechanics conspires — through the no-signaling theorem — to prevent this influence from being used to send an actual message faster than light. This is the reading closest to Bell's own later writings; Bell himself, especially after examining David Bohm's pilot-wave theory, leaned toward believing nature really is nonlocal, and thought this was the more natural lesson of his own theorem. The trouble is that "nonlocal but not signaling" sits awkwardly with special relativity. It's consistent with relativity's letter — no faster-than-light communication is possible — but arguably not its spirit, since in different reference frames observers can disagree about which measurement "happened first," making any nonlocal causal story frame-dependent in an uncomfortable way.

Give up realism. This is the more common reading among mainstream physicists sympathetic to the Copenhagen tradition: particles simply do not have definite pre-existing values for unmeasured quantities. There is no fact about what a spin measurement along an unmeasured axis "would have shown." On this view, Bell's theorem is not so much a shock about nonlocality as a confirmation of something quantum mechanics had been saying all along — that talk of definite unmeasured values is meaningless, and reality is, at bottom, contextual and observer-relative in a way classical intuition never anticipated. The appeal here is that it lets you keep locality and thus stay comfortably inside relativity's framework. The cost is metaphysical: you must accept that physical properties, in some robust sense, do not exist until measured, which raises the further question — central to debates over whether the wavefunction is ontic or merely epistemic — of what measurement even is, and why it should have this world-making power.

Give up measurement independence. This is the least popular option among working physicists, but logically it is exactly as available as the other two, and it has determined, careful defenders. The assumption says that the hidden variable $\lambda$ describing the particles is statistically independent of the experimenters' choice of measurement settings — that Alice and Bob are, in some meaningful sense, "free" to choose their settings however they like, uncorrelated with whatever was fixed at the particle source in the past. If this fails — if there is some correlation, however faint and however mediated, between the hidden state of the particles and the "random" choices made light-years away or nanoseconds later — then Bell's inequality can be violated by an entirely local, entirely realistic mechanism. This is sometimes called a "superdeterministic" loophole, and it is deeply disliked because it seems to require a conspiracy: the universe arranging, from the Big Bang onward, for measurement choices to correlate with particle properties in exactly the way needed to mimic quantum statistics, without any actual quantum nonlocality or indefiniteness at all.

None of the three options is refuted by any experiment we have. That is the uncomfortable core of the puzzle. Every loophole-closing experiment tightens the noose around specific alternative explanations — signaling, detection bias — but the fundamental interpretive trilemma is not an experimental loophole in that sense. It is baked into the logical structure of the theorem itself. No experiment, run inside spacetime by embedded observers who must themselves choose settings somehow, can force a demonstration that those choices were truly free of all prior correlation with the rest of the universe. This isn't a temporary technological gap; some researchers argue it is a principled limit on what any experiment, however refined, could show.

Given this genuine openness, the physics and philosophy communities have organized, loosely, around a handful of live research programs, each taking one horn of the trilemma seriously and building physics around it.

Nonlocal, deterministic hidden variables. The most fully worked out example is Bohmian mechanics (also called pilot-wave theory), developed by David Bohm in 1952, building on an earlier idea of Louis de Broglie. In this theory, particles have definite positions at all times — genuine realism, in the strongest sense — guided by a real physical wave, the same wavefunction quantum mechanics uses, which pushes and pulls particles along deterministic trajectories. Entanglement in this picture reflects a single wavefunction living in a high-dimensional configuration space that genuinely connects distant particles, so that an action on one particle instantaneously affects the "guiding wave" acting on its distant partner. Bohmian mechanics reproduces every prediction of standard quantum mechanics exactly — it is empirically indistinguishable from orthodox quantum theory in all tested regimes — while being explicitly, unapologetically nonlocal. Its appeal is enormous: it removes the measurement problem, offers a picture of a single objective world with no observer-dependence, and vindicates Einstein's instinct that beneath quantum randomness lies something more like classical determinism, just of an oddly holistic kind. Its trouble is exactly the tension with relativity mentioned earlier, along with the technical challenge of extending it cleanly to relativistic quantum field theory and to particle creation and annihilation, where Bohmian trajectories become far less natural.

Rejecting realism, embracing contextuality or radical indefiniteness. This is less a single theory than a family of stances, from Copenhagen-style instrumentalism to more modern relational and QBist approaches. QBism (Quantum Bayesianism), developed since the 2000s chiefly by Christopher Fuchs, Ruediger Schack, and collaborators, holds that quantum states represent an agent's personal degrees of belief about the consequences of their own future actions on the world, not objective features of external reality. On this view, Bell correlations are not spooky at all: they simply reflect that Alice's update of her beliefs about Bob's system, upon learning her own measurement outcome, need not respect any classical notion of pre-existing values, because there were never any "values" — objective or subjective — to violate locality with. There is no action at a distance because there is, in the relevant sense, no distant "thing" being acted upon independent of an agent's experience. This dissolves the puzzle rather than solving it in the traditional sense, which strikes some physicists as intellectually honest and others as changing the subject. Relational quantum mechanics, developed by Carlo Rovelli from the 1990s onward, takes a related but distinct tack: physical facts are always relative to a "reference system" or interacting system, and there's no absolute, observer-independent fact about outcomes at all until systems interact — dissolving the classical picture of one determinate reality that realism demands.

Superdeterminism and retrocausality. A minority but growing camp, including physicist Sabine Hossenfelder and, in different form, Tim Palmer, and the retrocausal program associated with Huw Price and Ken Wharton, takes seriously the possibility that measurement independence fails. Superdeterminists argue that if the universe is fully deterministic — as classical physics assumed and as many physicists still find aesthetically compelling — then in the strictest sense there was never any "free" choice of measurement settings at all; everything, including the experimenter's decisions, traces back through an unbroken causal chain to the same initial conditions that fixed the particles' hidden states. On this reading, Bell's theorem doesn't teach us anything is nonlocal; it just reveals that the universe's initial conditions were fine-tuned in a way that makes apparent quantum correlations emerge from ordinary local physics. Retrocausal approaches offer a subtler variant: rather than a conspiracy reaching back to the Big Bang, they suggest influences can propagate backward in time from the measurement event to the particle's preparation, restoring both locality and realism at the cost of allowing cause to sometimes follow effect. Superdeterminism's chief appeal is conceptual economy — it needs no nonlocality, no indefinite properties, no observer-dependence, just ordinary deterministic local physics operating on correlated initial conditions. Its chief cost is that it seems to undermine the very possibility of controlled experimentation: if what you "choose" to measure is secretly correlated with the systems you're measuring, how could science's basic method of randomized, independent testing be trusted anywhere? Critics, including most mainstream physicists, regard this as too high a price, verging on unfalsifiable; defenders respond that "too weird to be true" is not a scientific refutation, and note that fine-tuning arguments already play a legitimate role elsewhere in physics and cosmology.

Between these three poles lie further variants and hybrids — objective collapse models like GRW, which modify the Schrödinger equation itself to make collapse a real, spontaneous physical process, sidestepping some aspects of the debate while introducing new, in-principle testable deviations from standard quantum predictions; and many-worlds approaches, in which Bell correlations are explained not by any influence between branches but by the peculiar way in which the entire superposition, including all outcomes, evolves locally, with apparent nonlocal correlation emerging only when observers on different branches later compare notes. Many-worlds proponents argue their picture is fully local at the level of the universal wavefunction, though critics contend the "locality" achieved this way redefines the terms so thoroughly that it isn't answering the same question anymore.

This is the most uncomfortable part of the story: it is not clear that any experiment could settle the matter, at least not in the direct way physicists usually hope for.

Consider the freedom-of-choice loophole. Experimenters have pushed the "randomness" behind measurement settings further and further back — from quantum random number generators, to human choices in the 2018 "BIG Bell Test" involving over 100,000 volunteers worldwide, to light from quasars billions of light-years away, chosen so that any correlation between the settings and the particle source would have had to be established before those quasars' light was emitted, deep in cosmic history. Each of these experiments genuinely strengthens the case against superdeterminism, by making the required "conspiracy" ever more absurdly early, ever more astronomically vast in scope. But none of them can achieve a logical proof, because the experimenters themselves, the quasars, and the particle source all share a common causal past in the Big Bang. Superdeterminism, taken to its most extreme and least falsifiable form, can always retreat further back in time. Whether this counts as a live scientific hypothesis or an unfalsifiable escape hatch is itself a matter of ongoing, sometimes heated, debate between physicists like Hossenfelder, who argues superdeterminism deserves serious research investment rather than dismissal, and the majority who consider it explanatorily empty until it makes distinct, testable predictions.

So a full experimental refutation of the freedom-of-choice option looks structurally difficult, perhaps impossible, for any experiment run inside spacetime. What could move the needle is theoretical: if superdeterministic models could be built with genuine predictive power — making distinct predictions about, say, small deviations from quantum statistics in particular regimes, or explaining fine-tuned features of the Standard Model as a byproduct of the same underlying mechanism — they would become properly falsifiable rather than merely logically available. So far, no fully worked-out superdeterministic model reproduces all of known quantum mechanics while making new, testable predictions; this remains an active but small research program, and its proponents openly acknowledge the burden is on them to produce one.

For the locality-versus-realism question, the situation is a bit better, though still short of decisive. Different interpretations that reject different assumptions are, by design, empirically equivalent to standard quantum mechanics in all currently tested regimes — that's precisely why they've survived so long as live options. But some variants make distinguishable predictions at the edges. Objective collapse models, for instance, predict tiny deviations from unitary quantum evolution that grow with the size and mass of the system involved — deviations in principle detectable in sufficiently isolated, sufficiently large quantum superpositions, such as levitated nanoparticles or macroscopic mechanical oscillators cooled near their ground state. Experiments testing quantum superposition at ever larger mass scales — with molecules, then larger clusters, and eventually perhaps micron-scale objects — are underway in groups like those of Markus Arndt in Vienna, and could in coming decades either detect a genuine breakdown of standard quantum linearity (vindicating collapse models) or push the boundary of confirmed quantum behavior so far up in scale that collapse models are effectively ruled out in their simplest forms.

Similarly, if Bohmian-style hidden variable theories are ever extended rigorously into full relativistic quantum field theory in a way that makes distinct predictions — rather than merely reproducing the same statistics as orthodox theory by construction — that would open a genuine experimental front. So far this hasn't happened in a testable way; Bohmian mechanics remains empirically indistinguishable from standard quantum mechanics wherever both are well-defined.

The honest assessment, then, is this: the interpretive trilemma may not be a temporary gap awaiting better instruments, but a structural feature of any theory that reproduces quantum statistics while being tested by embedded, spacetime-bound observers. What could settle it is not necessarily a sharper Bell test — those have already done what they can do — but either a genuinely novel theoretical framework that makes distinguishable predictions (a testable superdeterministic model, a detectable collapse mechanism, a relativistically consistent nonlocal dynamics with new phenomenology), or perhaps a deeper conceptual argument, still unwritten, that finally makes one option unambiguously less costly than the others. Some physicists and philosophers hold out hope that progress on the nature of quantum information — treating information-theoretic principles as the bedrock from which quantum mechanics itself can be derived, as in the "reconstruction" programs pursued since the 2000s by Lucien Hardy, Rob Spekkens, and others — might dissolve the trilemma rather than resolve it, by showing that "locality," "realism," and "independence" were never quite the right categories in the first place. That, too, remains a hope rather than an achievement.

It would be easy to file this under abstract philosophy of physics, safely walled off from anything practical. That would be a mistake, on several counts.

First, technologically, Bell violations are not just a curiosity — they are the working principle behind device-independent quantum cryptography and randomness generation. Because the violation of a Bell inequality can be certified without trusting the internal workings of your devices, protocols exist that guarantee security or genuine randomness based purely on observed statistics, regardless of what an adversary might have done inside the hardware. These technologies work today because of Bell's theorem, entirely independent of which interpretation you favor — a nice reminder that operationally, quantum mechanics is settled even where its meaning is not.

Second, for foundational physics, the resolution matters enormously for the search for a theory of quantum gravity. If nature is genuinely nonlocal in some robust sense, that has implications for how spacetime itself might emerge from more fundamental, non-spatiotemporal structures — an idea taken seriously in some approaches to quantum gravity, where entanglement is conjectured to be the "glue" from which spacetime geometry emerges (as in the ER=EPR conjecture relating wormholes to entangled pairs). If instead realism is what falls, then any successful theory of everything will have to build indefiniteness into its most basic ontology, not just as an approximation for large ensembles but as a fundamental feature of what exists — reshaping what we even mean by "fundamental." If measurement independence is what fails, the implications ripple outward to the philosophy of scientific method itself, since virtually all experimental science assumes that experimenters can, at least approximately, choose their protocols independently of the systems under study.

Third, and perhaps most personally: this question sits at the intersection of physics and something like self-understanding. Bell's theorem, whichever horn you take, forces you to give up a piece of naive realism about the everyday world — either that distant things are truly independent of each other, or that things have properties whether or not anyone looks, or that your choices are free of cosmic-scale entanglement with everything that came before. Historically, physics has repeatedly forced such reconsiderations — of absolute space and time, of a clean separation between observer and observed, of determinism itself. Bell's theorem may be the sharpest such forcing yet, precisely because it is proven, not merely suggested by data open to gentler reinterpretation.

physicist[Alain Aspect|born 1947|French physicist whose experiments in the early 1980s at Orsay were the first to test Bell inequalities with time-varying analyzer settings, closing an important early version of the locality loophole; shared the 2022 Nobel Prize in Physics]

Aspect, along with John Clauser and Anton Zeilinger, received the 2022 Nobel Prize in Physics "for experiments with entangled photons, establishing the violation of Bell inequalities and pioneering quantum information science" — a prize that recognized experimental confirmation of the phenomenon while pointedly not endorsing any particular reading of what it meant. Clauser, working with Stuart Freedman in Berkeley in 1972, performed the first experimental test of any Bell-type inequality at all. Zeilinger's group in Vienna, along with teams led by Ronald Hanson in Delft and Sae Woo Nam and Krister Shalm at NIST, carried out the loophole-free tests of 2015 that many consider the decisive experimental chapter of this story — decisive, that is, for ruling out local hidden variables, not for choosing among the interpretations that remain.

David Bohm, working under political exile after refusing to testify before the House Un-American Activities Committee in 1949, developed his pilot-wave reformulation of quantum mechanics in Brazil and later in England, keeping alive a minority tradition that insisted realism and determinism need not be casualties of quantum weirdness. Today that tradition continues through physicists like Sheldon Goldstein, Detlef Dürr, and Nino Zanghì. On the opposing side of the ledger, Christopher Fuchs and the QBist school have pushed the anti-realist reading furthest, treating quantum probabilities as fundamentally about an agent's beliefs rather than the world's furniture. And Sabine Hossenfelder, alongside Tim Palmer, has argued publicly and often provocatively that superdeterminism deserves more serious research attention than the mainstream currently grants it — a minority position, but one that keeps the third horn of the trilemma from being forgotten.

There is something bracing about a question this well-posed and this unresolved. Most open problems in physics are open because we lack data, or lack a workable theory, or both. This one is different: the data are as clean as data get, replicated across a dozen experimental platforms and multiple Nobel-winning research programs, and the theorem connecting them to its conclusion is rigorous mathematics that any graduate student can verify line by line. What's missing isn't information. It's a way of adjudicating between metaphysical pictures that are, so far, equally compatible with all of it.

Some physicists find this state of affairs intolerable and expect it to resolve, eventually, through some future theory — of quantum gravity, of quantum information, of measurement itself — that will make the choice obvious in retrospect, the way relativity made the "aether" question obviously wrongheaded once you saw the right framework. Others suspect the trilemma is telling us something permanent: that classical intuitions about locality, definite properties, and free experimental choice were never going to survive contact with a genuinely quantum universe intact, and that the honest answer is to let go of the demand for a single, tidy resolution altogether, treating quantum mechanics's formalism as complete and its interpretation as, in some sense, underdetermined by design.

Both attitudes are respectable. Neither is proven. What's certain is that Bell's theorem did something remarkable: it took a disagreement that Bohr and Einstein could argue about for thirty years without any experiment ever needing to intervene, and turned it into a question the universe itself has now weighed in on, decisively, again and again — while still leaving room, at the very center of the answer, for us to keep arguing about what the universe actually said.

• Is measurement independence a testable assumption at all, or is it a permanent blind spot for any observer embedded in spacetime?
• Can Bohmian-style nonlocal hidden variables be extended into a fully relativistic quantum field theory with genuinely new, testable predictions?
• Could quantum gravity research — treating entanglement as the source of spacetime geometry — settle the locality question from an entirely different direction?
• Do information-theoretic reconstructions of quantum mechanics dissolve the trilemma, or merely relabel it?
• Would a confirmed detection of spontaneous collapse (or its confirmed absence at ever larger mass scales) change which horn of the trilemma looks most costly?

Ask the tutor: If a future experiment definitively ruled out objective collapse models by confirming quantum superposition at macroscopic scales, what would that actually tell us about the locality-versus-realism choice — and would it tell us anything at all?