Counting the Worlds in the Machine: What Quantum Computing’s Successes Do — and Don’t — Tell Us About Reality

Abstract

Quantum computing has entered a remarkable phase. Between 2019 and 2026, laboratories crossed thresholds once considered distant: Google demonstrated “quantum supremacy” and later, with its Willow chip, the long-sought “below-threshold” error correction; trapped-ion, neutral-atom, and photonic platforms produced their first genuine logical qubits; and corporate roadmaps now target large fault-tolerant machines by the end of the decade. These advances have revived a tantalizing question first posed by physicist David Deutsch: when a quantum computer factors a number too large for the visible universe to hold, where was the computation performed? Deutsch’s answer — in parallel universes — is the most famous claim that quantum computing proves the Many-Worlds interpretation of quantum mechanics. This article surveys, for a general audience, both the real state of quantum-computing progress and the interpretational debate it has reignited. Its central conclusion reflects the mainstream expert consensus: a working quantum computer is consistent with every major interpretation of quantum mechanics, because all of them make identical experimental predictions. Quantum computing, however impressive, does not — so far — favor any single picture of reality. We close by examining which experiments, if any, could ever break that tie.

Part One: The State of Quantum Computing Success

What a quantum computer actually is

A classical computer stores information in bits, each definitely 0 or 1. A quantum computer uses qubits, which exploit two genuinely quantum phenomena. The first is superposition: a qubit can be in a combination of 0 and 1 at once, described by a mathematical object called the wavefunction. The second is entanglement: two or more qubits can be correlated so tightly that their joint state cannot be described as the states of each separately — Einstein’s “spooky action at a distance.”

It is tempting to say a quantum computer “tries all answers at once,” but this is the single most misleading sentence in popular science. As Scott Aaronson, a leading quantum-computing theorist, puts it at the top of his blog: quantum computers won’t solve hard problems instantly by just trying every solution in parallel. The reason is measurement: when you read a quantum computer, you get only one outcome. The art of quantum algorithm design is to arrange interference — the same wave phenomenon that makes ripples reinforce or cancel — so that wrong answers cancel out and the right answer is amplified before you look. scottaaronson

The enemy of all this is decoherence: the leakage of quantum information into the environment, which destroys superposition and entanglement. A stray photon, a thermal vibration, or a cosmic ray can collapse a delicate quantum state. This is why most quantum computers are cooled to near absolute zero or held in ultrahigh vacuum.

Quantum error correction, logical qubits, and the threshold theorem

Because physical qubits are so fragile, useful computation requires quantum error correction (QEC). The idea, pioneered by Peter Shor in 1995, is to encode one robust logical qubit across many noisy physical qubits, so that errors can be detected and corrected without disturbing the underlying quantum information. The threshold theorem is the foundational result that says: if the error rate of the physical components can be pushed below a certain critical value, then adding more physical qubits per logical qubit makes the logical error rate fall exponentially. Below threshold, scaling up improves reliability; above threshold, scaling up makes things worse. Crossing that line has been the central goal of the field for three decades.

The milestones, 2019–2026

Google’s supremacy claim (2019). Google’s 53-qubit Sycamore processor performed a contrived benchmark called random circuit sampling — essentially generating samples from a probability distribution that is hard for classical computers to reproduce — and claimed it would take a supercomputer 10,000 years. IBM contested the comparison, arguing a better classical algorithm could do it in days, and skeptics noted the task had no practical use. This established a pattern: supremacy demonstrations are narrow, carefully chosen benchmarks, not useful computation. scottaaronson

Google’s Willow chip (December 2024). This was the more scientifically important milestone. On a 105-qubit processor, Google demonstrated below-threshold error correction: scaling its surface-code logical qubit from a 3×3 to a 5×5 to a 7×7 array of physical qubits cut the logical error rate roughly in half each time. The peer-reviewed result (published in Nature, from the team’s arXiv paper “Quantum error correction below the surface code threshold”) reports a logical error rate of 0.143% ± 0.003% per cycle for a 101-qubit distance-7 code, suppressed by a factor of Λ = 2.14 ± 0.02 for each increase of code distance by two, with a logical qubit that exceeded its best physical qubit’s lifetime by a factor of 2.4 ± 0.3 — “beyond break-even.” John Preskill of Caltech called the demonstration “a notable milestone.” Willow also ran a fresh random-circuit-sampling benchmark Google said would take a classical supercomputer 10 septillion (10²⁵) years. scottaaronson

It was in this announcement that Google Quantum AI’s founder, Hartmut Neven, wrote that Willow’s speed “lends credence to the notion that quantum computation occurs in many parallel universes, in line with the idea that we live in a multiverse, a prediction first made by David Deutsch” — the spark for the interpretational debate in Part Two. TechHQ

IBM’s roadmap. IBM has pursued raw scale and a transparent roadmap: the 127-qubit Eagle (2021), the 433-qubit Osprey (2022), the 1,121-qubit Condor (2023), and the high-quality 156-qubit Heron family that now anchors its fleet. In June 2025, IBM laid out a detailed path to Starling, described in its own words as “a large-scale, fault-tolerant quantum computer capable of running quantum circuits comprising 100 million quantum gates on 200 logical qubits,” targeted for delivery by 2029 at a new IBM Quantum Data Center in Poughkeepsie, New York. Starling uses efficient quantum low-density parity-check (qLDPC) codes that IBM says cut physical-qubit overhead by up to 90%. IBM projects “quantum advantage” by the end of 2026 and a later machine, Blue Jay, with 2,000 logical qubits by 2033. (These are corporate roadmap targets, not delivered results; IBM notes it has met its interim milestones to date, but the 2029 and 2033 systems remain forecasts.)

Trapped ions (Quantinuum, IonQ). Trapped-ion machines manipulate individual charged atoms with lasers and boast the highest gate fidelities in the industry. In November 2025, Quantinuum launched Helios, a 98-physical-qubit machine delivering 48 fully error-corrected logical qubits at a 2:1 encoding rate — a density the company called “previously considered unattainable” and the best encoding efficiency demonstrated by any platform — alongside a quantum-volume record of 33,554,432. According to Quantinuum’s arXiv paper “Helios: A 98-qubit trapped-ion quantum computer,” the machine uses barium-137 hyperfine qubits and achieves average infidelities of 2.5×10⁻⁵ for single-qubit gates and 7.9×10⁻⁴ for two-qubit gates (i.e., two-qubit gate fidelity of about 99.92%). IonQ reported a world-record two-qubit gate fidelity of 99.99% in 2025 and projects aggressive scaling toward roughly two million physical qubits and tens of thousands of logical qubits by 2030.

Neutral atoms (QuEra, Atom Computing). These trap neutral atoms in “optical tweezers,” a flexible architecture well-suited to error correction. A Harvard/MIT/QuEra collaboration demonstrated 48 logical qubits in 2023; in 2025 QuEra, with Harvard, MIT, and Yale collaborators, reported a 3,000-qubit array operating continuously for over two hours and executed algorithms with up to 96 logical qubits. Atom Computing, partnering with Microsoft, demonstrated 24 logical qubits from 1,180 physical atoms.

Photonics (PsiQuantum, Xanadu). These encode qubits in particles of light, operating largely at room temperature and leveraging existing semiconductor manufacturing. PsiQuantum is betting on a single leap to a roughly one-million-qubit fault-tolerant machine, raising $1 billion in a 2025 Series E round and targeting utility-scale sites in Brisbane and Chicago. Xanadu demonstrated on-chip generation of error-resistant “GKP” states (published in Nature) in 2025, a building block for fault-tolerant photonic computing.

What does “success” mean — and the skeptics

It is crucial to distinguish three things: (1) benchmark supremacy — beating classical computers at a contrived task like random circuit sampling, achieved but of no practical value; (2) quantum advantage — beating classical computers at a useful task, not yet convincingly demonstrated; and (3) fault-tolerant, large-scale computation — running algorithms like Shor’s on cryptographically relevant numbers, which requires on the order of millions of physical qubits and remains years away.

We are, by consensus, still in what Preskill named the NISQ era — Noisy Intermediate-Scale Quantum — characterized by devices of roughly 50 to a few thousand qubits that are too noisy for sustained error correction. In a 2025 essay (“Beyond NISQ: The Megaquop Machine”) Preskill argued the field is approaching the threshold of the next era, which he dubbed the “megaquop machine” — a device capable of a million reliable quantum operations. TechTarget Caltech

Not everyone believes the destination is reachable. The mathematician Gil Kalai is the most prominent skeptic, arguing that scalable quantum computing is impossible in principle. His case is that noise in quantum systems is not merely an engineering nuisance but is correlated in ways that will always defeat error correction — that NISQ-scale devices produce “primitive” distributions too weak to bootstrap into fault tolerance. “I hold the view that building scalable quantum computers — and even achieving some early milestones on the path to quantum computing — is impossible,” Kalai wrote in 2025. Most physicists disagree, and Willow’s below-threshold result is widely seen as direct evidence against Kalai’s strongest claims, but the debate continues; Kalai publicly debated quantum researcher Matthias Christandl at the Learned Society of the Czech Republic in 2025. The honest assessment as of mid-2026: the engineering trend is strongly positive, several independent platforms have crossed key thresholds, but no machine has yet performed a useful computation beyond classical reach, and large-scale fault tolerance remains a forecast, not a fact.

Part Two: Which Interpretation Does Quantum Computing Support “So Far”?

Why interpretations exist

Quantum mechanics is among the most precisely tested theories in science. Yet the equations leave open a stubborn question: what is the wavefunction, and what physically happens when we measure it? Before measurement, a system is in superposition; after, we see one definite outcome. This is the measurement problem, and the various interpretations of quantum mechanics are competing answers to it. The critical fact for this article is that all the mainstream interpretations make the same experimental predictions. They are, in the regimes we can currently access, empirically equivalent — which is precisely why the choice among them is so contested.

The major interpretations, neutrally surveyed

Copenhagen (Bohr, Heisenberg): the wavefunction represents our knowledge or the outcomes of measurements, not necessarily an underlying reality; measurement causes “collapse,” and one need not ask what happens between measurements. It is less a single doctrine than a family of related views.

Many-Worlds / Everettian (Everett, DeWitt, Deutsch, Wallace): the wavefunction is real and never collapses; it simply evolves smoothly forever. Every possible outcome occurs, each in a branch of a vast, continually splitting multiverse. There is no special measurement process — just entanglement and decoherence sorting the universal wavefunction into non-interacting branches.

de Broglie–Bohm / pilot-wave theory (Bohm): particles have definite positions at all times, guided by a real “pilot wave.” It is deterministic and reproduces all quantum predictions, at the cost of explicit non-locality. There is only one world.

QBism (Fuchs, Schack): the quantum state represents a rational agent’s subjective degrees of belief, updated upon measurement. “Collapse” is just an agent revising expectations — an epistemic, not physical, event.

Objective collapse (GRW; Penrose; Diósi): the wavefunction is real and collapse is a genuine physical process, governed by new dynamics that make superpositions of large objects collapse spontaneously and quickly while leaving microscopic systems alone. Crucially, these theories make different predictions from standard quantum mechanics and are therefore testable — more on this in Part Three.

Relational quantum mechanics (Rovelli): physical quantities are not absolute but relative to the system doing the observing. As Rovelli frames it, just as velocity is meaningful only relative to another object, quantum states are relative to an interacting system; there is no observer-independent set of facts.

Consistent (decoherent) histories (Griffiths, Gell-Mann, Hartle): quantum mechanics assigns probabilities to entire self-consistent sequences of events (“histories”) rather than relying on measurement collapse.

A unifying scientific insight cutting across all of these is decoherence, developed by Wojciech Zurek and others. Decoherence explains why we never see macroscopic superpositions: interaction with the environment rapidly “einselects” stable “pointer states” and destroys interference between them, imposing, in Zurek’s words, “an effective ban on the vast majority of the Hilbert space, eliminating especially the flagrantly non-local ‘Schrödinger cat’ states.” Decoherence is interpretation-neutral — it is solid physics accepted by all camps — but it does not by itself say whether the other branches are real (Many-Worlds) or merely apparent (Copenhagen). It explains the appearance of collapse without settling its reality.

Deutsch’s challenge: the case that quantum computing proves Many-Worlds

David Deutsch, who in 1985 conceived the universal quantum computer, is the most forceful proponent of the view that quantum computing supports Many-Worlds. His argument, from The Fabric of Reality (1997, p. 217), is worth quoting at length:

“To predict that future quantum computers … will work … one need only solve a few uncontroversial equations. But to explain exactly how they will work, some form of multiple-universe language is unavoidable. … When a quantum computer delivers the output of such a computation, we shall know that those intermediate results must have been computed somewhere, because they were needed to produce the right answer. So I issue this challenge to those who still cling to a single-universe world view: if the universe we see around us is all there is, where are quantum computations performed? I have yet to receive a plausible reply.” arxiv

Sharpened for Shor’s factoring algorithm, the challenge runs: factoring a large number may require manipulating more intermediate results than there are atoms in the visible universe (about 10⁸⁰). If a quantum computer factors a number using “10⁵⁰⁰ or so times the computational resources that can be seen to be present,” then — Deutsch asks — “where was the number factorized? … Who did factorize it, then? How, and where, was the computation performed?” His answer: across the branches of the multiverse. This is the intuition Hartmut Neven echoed in 2024 when he said Willow “borrowed” its computation from parallel universes.

Why most experts reject the inference

Here is the crux, and it must be stated plainly: the mainstream view among physicists, computer scientists, and philosophers of physics is that quantum computing does not favor Many-Worlds over the alternatives. The reasons are several and reinforce one another.

1. Empirical equivalence. Every interpretation predicts exactly the same behavior for a quantum computer. A Bohmian, a QBist, a Copenhagenist, and an Everettian all calculate identical outputs for Shor’s algorithm. So a working quantum computer cannot be evidence for one and against another — it confirms quantum mechanics, which they all share. Scott Aaronson made this point directly in response to the Willow claim on his blog Shtetl-Optimized (December 2024): the experiment “doesn’t add anything new to this old debate. It’s yet another confirmation of the predictions of quantum mechanics. What those predictions mean for our understanding of reality can continue to be argued as it’s been since the 1920s.” He added that updating one’s interpretation on the basis of the experiment “is a matter of psychology rather than logic” — anyone moved by the philosophical case for Many-Worlds should already have accepted it from the double-slit experiment, while a Bohmian or QBist “should’ve made exactly the same prediction for Google’s experiment and therefore had no update.” scottaaronson + 2

2. The “you can’t read out the parallel results” rebuttal. The popular image of a quantum computer harvesting answers from parallel worlds is, technically, wrong — because you only ever extract one answer, shaped by interference. Andrew Steane made this the centerpiece of a 2003 paper bluntly titled “A quantum computer only needs one universe” (Studies in History and Philosophy of Modern Physics 34B: 469–478). Steane argues that the impression of “vast parallel computation … is a false impression engendered by an imperfect mathematical notation,” and that since the concept of parallel universes “implies a computational power which is not in fact present in quantum computation, … such an image obscures more than it illuminates.” His direct reply to Deutsch’s “where” question: the computation happens “in the region of spacetime occupied by the quantum computer.” Steane even argues the parallel-universe picture is testable and wrong — if a machine were “really ‘doing 2ⁿ computations,’” it would be sensitive to errors at the level 1/2ⁿ, which it is not. arxiv arxiv

3. A tension between “computational worlds” and “decoherent worlds.” Many-Worlds defines its branches using decoherence — branches are worlds precisely because they have decohered and no longer interfere. But a quantum computation depends on keeping its qubits coherent and interfering; decoherence is the very thing engineers fight to prevent. The philosopher of physics Michael Cuffaro has emphasized this prima facie tension: the “worlds” a quantum computer supposedly exploits are not the same as the decoherent worlds of the Many-Worlds interpretation, and the identification of the two breaks down especially in alternative models like measurement-based (cluster-state) quantum computing. stanford

4. Even leading Everettians are cautious. David Wallace, the foremost academic defender of Many-Worlds, does not rest his case on quantum computing. He notes that the “quantum parallelism” picture may not even apply to most quantum algorithms — Shor’s algorithm seems to fit it (“Shor’s algorithm, at least, does seem to operate in this way,” he writes in the 2010 volume Many Worlds?), but many others do not. Wallace believes Many-Worlds is true on independent grounds (decoherence and the measurement problem), treating quantum computing as at most a vivid illustration, not a proof.

5. Critics of Many-Worlds on other grounds. Adrian Kent of Cambridge, a prominent skeptic, has argued at length that Everettian accounts struggle to explain probability and the Born rule and to justify why ordinary “Copenhagen” quantum theory appears confirmed — concluding in “Against Many-Worlds Interpretations” that “no plausible set of axioms exists for an MWI that describes known physics.” His critique targets the foundations of Many-Worlds itself, undercutting the broader inference from quantum computing to the multiverse. (Other named skeptics of the parallelism argument include Itamar Pitowsky, Jeffrey Bub, and Armond Duwell, who accepts that quantum computers compute “in parallel” in some sense but denies this uniquely supports Many-Worlds.) arXiv arXiv

What experts actually say when asked

It is worth noting the sociology, carefully. The Stanford Encyclopedia of Philosophy observes that Many-Worlds enjoys disproportionate support among quantum-information researchers and quantum cosmologists, because the multiverse picture meshes intuitively with “parallel processing” — though the same entry stresses that “the usefulness of the MWI in explaining the speedup of quantum computation has been questioned.” Sean Carroll, a physicist and articulate Many-Worlds advocate, is candid that the interpretation is not empirically favored: in his words it “makes exactly the same predictions” as its rivals, and he argues for it on grounds of simplicity and taking the equations literally, not on grounds of any decisive experiment. On the other side, Sabine Hossenfelder dismissed the Willow multiverse claim, noting that random circuit sampling’s result “has no practical use” and that the calculation is “exactly the same” type Google ran on a circa-50-qubit chip in 2019. The pattern is clear: scientists’ interpretational preferences track their philosophical tastes, not the data — which is exactly what we should expect if the interpretations are empirically equivalent. Podcasts

Part Three: What Would Greater Success Suggest?

If quantum computers scale to running Shor’s algorithm on huge numbers

Suppose, by the 2030s, a fault-tolerant machine factors a 2,048-bit number, breaking real-world encryption. Does that settle the interpretation debate? The mainstream answer is no. The logic of Part Two is unchanged by scale: a bigger quantum computation is still a confirmation of quantum mechanics, which all interpretations share. Deutsch’s challenge becomes rhetorically more vivid at 10⁵⁰⁰ “resources,” but it does not become more logically forcing. As Aaronson and Steane argue, the resources are never independently observed — they collapse, interfere, or branch (depending on your interpretation) into the single answer you read out. The honest nuance, which Deutsch himself half-concedes, is that large-scale quantum computing might make Many-Worlds more psychologically compelling to some. In his words, complex quantum computation “adds nothing to a case [for Many-Worlds] that is already unanswerable. But it does add psychological impact. With Shor’s algorithm, the argument has been writ very large.” Psychological impact is not logical proof — a distinction at the heart of this whole debate.

Experiments that genuinely could distinguish interpretations

Here the picture changes, and it is the most important point for an honest treatment. Some alternatives to standard quantum mechanics are not empirically equivalent — they make different predictions and are therefore testable.

Objective collapse theories (Penrose–Diósi, GRW/CSL). These predict that superpositions of sufficiently massive objects collapse spontaneously. The Diósi–Penrose model attributes collapse to gravity: a mass in two places at once creates a superposition of two spacetime geometries, which Penrose argues is unstable and decays on a calculable timescale. Crucially, this is falsifiable. A 2020 experiment at Italy’s Gran Sasso underground laboratory searched for the faint radiation that collapse models predict charged particles should spontaneously emit. As reported by Donadi, Curceanu, Bassi, Diósi and colleagues in Nature Physics (“Underground test of gravity-related wave function collapse”), the germanium-detector search set a lower bound on the effective size of the mass density of nuclei “about three orders of magnitude larger than previous bounds,” a result that “rules out the natural parameter-free version of the Diósi-Penrose model” — though modified versions with a free length parameter survive. A generation of experiments — matter-wave interferometry with ever-larger molecules, levitated nanoparticles, and tabletop optomechanics — is steadily tightening the bounds. This is the clearest example of interpretation-related physics being tested, and partly falsified, in the laboratory.

Larger and larger superpositions. Every time physicists place a bigger object into a verified superposition, they push back the frontier where any spontaneous collapse could hide. If macroscopic quantum coherence holds to arbitrarily large scales, collapse theories are squeezed out; if a genuine, unexplained collapse were ever observed, it would be revolutionary evidence for them — and against the unitary-only picture of Many-Worlds.

Wigner’s friend and the Frauchiger–Renner theorem. These probe whether different observers can share a single objective set of “facts.” In 2018, Daniela Frauchiger and Renato Renner proved a no-go theorem titled “Quantum theory cannot consistently describe the use of itself” — arguing that certain assumptions (a single world, observer-independent facts, and the universal validity of quantum theory) cannot all hold together. In 2019, Massimiliano Proietti and colleagues at Heriot-Watt University realized an extended Wigner’s-friend scenario. As reported in Science Advances (“Experimental test of local observer independence”), from 1,794 six-photon coincidence events recorded over 360 hours of measurement, the team violated the associated Bell-type inequality by five standard deviations, concluding that “if one holds fast to the assumptions of locality and free choice, this result implies that quantum theory should be interpreted in an observer-dependent way.” These experiments do not crown a single interpretation, but they sharpen the constraints, pressuring views that insist on absolute, observer-independent facts and lending comfort to relational and QBist views. (Critics, including Tim Maudlin, note that six photons “obviously do not experience anything,” so the “observers” are stand-ins, not conscious agents — a real limitation on what such experiments can show.) arxiv

Would a failure of quantum computing favor objective collapse?

This is the most intriguing honest question. If scalable quantum computing turned out to be impossible — if, as Gil Kalai argues, noise always wins — would that support collapse theories? Potentially, yes, in a limited way. Aaronson has made a striking observation: “If quantum computing is impossible, I would say that quantum mechanics as physicists have taught and understood it for 80 years is wrong,” because it would imply some efficient classical description of quantum systems that standard theory says shouldn’t exist. A fundamental, unavoidable breakdown of large-scale quantum coherence is exactly what objective-collapse theories predict and what standard quantum mechanics (and Many-Worlds) denies. So a persistent, principled failure of error correction — not mere engineering difficulty — could be read as evidence for new collapse physics. This is why Kalai himself suggests that even if he is right, the outcome “could still lead to important scientific discoveries in physics.” Importantly, the field’s recent below-threshold successes push in the opposite direction, against any such fundamental barrier.

What it would take, even in principle, to favor one interpretation

The lesson is that interpretations divide into two classes. Those that are strictly empirically equivalent to standard quantum mechanics — Copenhagen, Many-Worlds, Bohm, QBism, relational QM, consistent histories — cannot, even in principle, be distinguished by any experiment that merely confirms quantum mechanics, quantum computers included. To favor one would require either finding a regime where they secretly disagree, or appealing to non-empirical criteria like simplicity, locality, or explanatory elegance — philosophical, not experimental, grounds. Those that are not equivalent — the objective-collapse theories — can be tested, and are being tested, precisely because they dare to predict something standard quantum mechanics does not. The irony is sharp: the only interpretations a quantum computer could ever bear on are the ones that predict the quantum computer might fail.

Recommendations: How to Read the News Responsibly

For the intelligent lay reader trying to navigate quantum-computing headlines, a few staged, concrete guidelines follow — along with the benchmarks that should change your assessment.

Separate the three kinds of “success.” When a company announces a milestone, ask which it is: a contrived benchmark (supremacy/random circuit sampling — impressive engineering, no practical value), a useful speedup (quantum advantage — not yet credibly achieved), or fault-tolerant computation (still years off). Treat “10 septillion years” comparisons as benchmark theater, not utility. Threshold that would change this: a peer-reviewed demonstration of a useful, verifiable computation that no classical computer can match.
Treat below-threshold error correction as the milestone that matters. Willow (2024), Quantinuum’s Helios (2025), and QuEra’s logical-qubit arrays are genuinely significant because they show error rates falling as systems grow — the precondition for everything else. Benchmark to watch: logical qubits that stay reliable across millions of operations (Preskill’s “megaquop” machine), and roadmaps like IBM’s Starling (2029) or PsiQuantum’s million-qubit machine actually delivering on schedule.
Discount any claim that a chip “proves the multiverse.” This is the firmest recommendation in this report. No experiment that simply confirms quantum mechanics — and every quantum computer does exactly that — can favor Many-Worlds over Copenhagen, Bohm, QBism, or relational QM. When you see such a claim (as in the Willow announcement), recognize it as metaphysics or marketing, not a scientific result. The genuine achievement stands on its own.
Watch the genuinely decisive experiments instead. The places where reality could actually adjudicate are objective-collapse tests (Gran Sasso-style searches, large-molecule interferometry, levitated nanoparticles) and extended Wigner’s-friend experiments. Benchmark that would be revolutionary: an observed, reproducible collapse of a macroscopic superposition that standard quantum mechanics cannot explain — or, conversely, verified coherence of ever-more-massive objects, which would steadily kill collapse models.
Hold the skeptics in view. Kalai’s challenge has not been formally refuted, only outpaced by experiment so far. If error correction were to stall at a hard wall despite improving hardware, take that seriously — it would be the most interesting possible outcome, with implications for physics itself.

Caveats

Corporate roadmaps are promises, not results. IBM’s Starling (2029), IonQ’s and PsiQuantum’s million-qubit ambitions, and similar targets come from companies with commercial and fundraising incentives. Several analysts (and IBM’s own framing) note these are iterative plans that may slip. Delivered, peer-reviewed milestones (Willow’s Nature paper, the Helios arXiv paper, the Gran Sasso Nature Physics result) carry far more evidential weight than projected dates.
“Quantum advantage” remains unclaimed in any rigorous, useful sense as of mid-2026. Benchmark supremacy has been demonstrated; practical advantage has not been convincingly shown, and earlier supremacy claims (Google 2019) were partly contested by improved classical algorithms.
The empirical-equivalence claim has a boundary. It holds for the interpretations that are mathematically equivalent to standard quantum mechanics. Objective-collapse theories are genuinely different physical theories, not mere interpretations, and are testable — a distinction this article treats as central rather than incidental.
Wigner’s-friend experiments use proxy “observers.” The photonic realizations involve no conscious agents, so their bearing on interpretations that hinge on observers (QBism, some Copenhagen variants) is suggestive, not conclusive.
Expert opinion is genuinely divided and partly sociological. Surveys of physicists show no consensus interpretation, and community preferences (e.g., Many-Worlds among quantum-information researchers) appear to track training and taste rather than evidence — consistent with the underdetermination this article describes.

Epilogue: The Cartographer’s Temptation

There is a very human temptation, when standing before a machine that does something miraculous, to read our favorite story of reality directly off its workings. Google’s Willow chip performs in five minutes what would take a classical computer longer than the age of the universe, and it feels almost irresistible to conclude that the extra room must have come from somewhere — from other worlds, other universes, a hidden vastness.

But the deepest lesson of a century of quantum mechanics is that the mathematics is exquisitely clear while its meaning remains radically open. The equations that let engineers build Willow, Helios, and Starling are the same equations whether you believe the universe is one or uncountably many, whether the wavefunction is a real physical object or a bookkeeping device for an agent’s expectations. The quantum computer is gloriously indifferent to our metaphysics. It computes the same way in Deutsch’s multiverse, in Bohm’s single guided world, in Rovelli’s web of relations, and in Fuchs’s community of believing agents.

That indifference is not a failure of physics; it is a discipline. It reminds us to separate what we have measured from what we have chosen to imagine — and to hold the second loosely. The genuine triumphs of quantum computing are extraordinary enough without conscripting them as evidence in a debate they cannot adjudicate. The worlds, if there are worlds, will not be counted by any chip. They remain, for now, a question for the wakeful mind rather than the running machine — and perhaps that is exactly where, until an experiment can do better, they belong.