Quick read · plain language
A Call for Honest Computing
A plain-language version of “Manifesto for Honest Computation”
The short version
A group of writers says that quantum computing has failed to deliver on its promises. They want money moved to other kinds of computing. They ask everyone to demand proof before believing claims.
What went wrong
The writers say that one type of computing has eaten up too much money. This type is called gate-model quantum computing. It has taken about sixty out of every hundred dollars spent on new kinds of computing. The total spent is about thirty-five billion dollars.
The problem? No quantum computer has solved any real business problem better than a normal computer. The writers say this is not a small delay. The whole system is broken. People who fund research get rewarded for big claims, not for real results. No one checks the claims in time to stop bad spending.
The writers also say that the basic building block of this computing, the qubit, is misunderstood. It is not a thing like a tiny ball. It is more like a relationship between things. Thinking of it as a thing leads to bad choices.
What works better
Other kinds of computing already work or show real promise. Ising machines solve hard scheduling problems and are sold today. Neuromorphic chips use far less power than normal chips for some tasks. Optical computers use light and work at room temperature. Analog quantum simulators help scientists study materials.
The writers say the best way forward is to spread money across many types of computing. They suggest only five to ten dollars out of every hundred should go to gate-model quantum computing. Most should go to thermodynamic, optical, and neuromorphic computing instead.
Five rules for honest work
The writers offer five rules. First, match the physical machine to the problem. Do not force every problem into the same shape. Second, focus on connections between parts, not on single parts. Third, measure success by energy used per answer. This is the only fair way to compare different machines. Fourth, every funded project must say ahead of time what would count as failure. If that failure happens, funding stops. Fifth, people who make claims must not be the ones who check them. Independent groups must do the checking.
What this paper does not show
The writers do not prove that gate-model quantum computing will never work. They say it might work someday for a few special problems like breaking codes. They do not claim that any one alternative is ready to replace all normal computing. They also do not show that their suggested new agency would actually fix the problems. They call for it, but they do not test the idea.
What the writers ask of everyone
The writers ask investors to demand energy-per-answer numbers before putting in money. They ask researchers to register their claims publicly and to welcome failed tests as useful knowledge. They ask government funders to start a new agency modeled on DARPA, to require independent checks, and to buy computing power based on energy used per answer. They ask the public to ask hard questions and to follow the energy numbers, not the press releases.
The writers themselves promise to report energy-per-answer for every claim, to set deadlines they can fail, and to publish independent checks even when those checks prove them wrong. They challenge others to make the same promise.
Words used here
- qubit
- the basic unit in quantum computing, here described as a relationship between states rather than a physical object
- gate-model quantum computing
- a design for quantum computers that uses step-by-step operations, like a recipe with many stages
- joules per solution
- a measure of how much energy a computer uses to finish one task
- falsifiability
- the quality of a claim that lets someone prove it wrong if it is wrong
Learn more in plain language
- DARPASimple English Wikipedia
- Neuromorphic computingWikipedia (English, more technical)
- Optical computingWikipedia (English, more technical)
- Research ethicsWikipedia (English, more technical)
- QubitSimple English Wikipedia
This plain-language version was written by an AI model from the text of the paper and checked against plain-language rules: everyday words, no mathematics, no unexplained names. It leaves details out and can still be wrong; the paper is the source of record. To report a problem, open the paper, select the passage and choose Report.
For specialists: the technical digest, quoted from the paper
In the paper's words
Five papers have traced an arc from diagnosis to prescription. The qubit is a scaffold, not an invariant. Alternative paradigms exist.
The abstract, verbatim, and its opening sentence is the claim. Every line below is quoted from the text.
Numbers it reports
We propose a specific resource reallocation --- from the ~60% currently consumed by gate-model quantum computing toward thermodynamic, optical, neuromorphic, and analog quantum simulation --- and a Falsification Pledge by which institutions commit to pre-registering computational claims with time-bound success criteria.
from AbstractAllocation Substrate What Success Looks Like 30% Analog quantum simulation 10,000-atom simulators; catalyst/pathway design 25% Optical processors 1,000×1,000 photonic circuits in production 20% Neuromorphic + memristive Memristive crossbar at commercial scale 15% Reversible/adiabatic CMOS Processor at 1% Landauer limit 10% Photonic quantum (room-temp) Logical qubit with error rate $\lt 10^{-6}$
from 3.2 Medium-Term (3-7 Years): Scale What Shows PromiseApproximately 45% thermodynamic and analog (Ising machines, p-bit networks, analog quantum simulators), 25% optical (integrated photonic matrix processors), 20% neuromorphic, and only 5-10% fault-tolerant quantum --- a near-inversion of the current allocation where gate-model QC absorbs roughly 60% of advanced computing investment.
from 1.4 The Portfolio (Phase IV)Allocation Substrate Problem Class Status 25% Ising machines / thermodynamic solvers Combinatorial optimization Commercially available 20% Optical matrix processors Linear algebra (inference) Research prototypes 20% Neuromorphic processors Probabilistic inference, edge AI Commercially available 15% p-bit probabilistic networks Sampling, optimization Research prototypes 10% Analog quantum simulators Quantum many-body physics Research demonstrations 10% Conventional CMOS + algorithms General-purpose Dominant incumbent What to STOP funding at near-term scale: Universal fault-tolerant gate-model QC on superconducting or trapped-ion platforms.
from 3.1 Near-Term (1-3 Years): Deploy What WorksThese platforms should be maintained as fundamental research programs at ~5% of current funding levels, not as commercial ventures at ~60%.
from 3.1 Near-Term (1-3 Years): Deploy What WorksAllocation Substrate Rationale 30% Reversible classical computing Path to Landauer-limit general-purpose computation 25% Photonic quantum (MBQC, CV) Room-temperature; avoids cryogenic penalty 20% Analog quantum simulation at scale Pharmaceutical and materials discovery 15% Fault-tolerant QC (crypto only) Only path for factoring; government security interest 10% Field-theoretic computation Speculative; fundamental research only
from 3.3 Long-Term (7-15+ Years): Invest in the Physics
How it is done
We propose five principles: (1) the substrate IS the algorithm; (2) correlation over particle; (3) joules per solution as the universal metric; (4) falsifiability as a condition of funding; and (5) institutional independence of verification.
from AbstractThe Landauer bound does not privilege quantum over classical computation.
from 1.3 The Limits (Phase III)Five reforms are necessary: an Advanced Computing Projects Agency (ACPA) modeled on DARPA; mandatory independent verification for computational claims above a materiality threshold; structural separation of quantum hardware and software; joules-per-solution procurement for all government computational contracts; and a public registry of computational claims with time-bound falsification criteria.
from 1.5 The Institutions (Phase V)The honest approach: match the problem class to the physical substrate whose natural dynamics solve it most efficiently.
from Principle 1: The Substrate IS the Algorithm
What would falsify it
No computational research program should receive public funding without pre-registered, time-bound, falsifiable milestones.
from Principle 4: Falsifiability as a Condition of FundingFalsifiability is not an optional feature of honest science.
from Principle 4: Falsifiability as a Condition of FundingEvery prediction we make about future computational capability will be accompanied by a specific deadline and a specific falsification criterion.
from 5. The Falsification Pledge
Where the claim stops
The laws of physics do not negotiate.
from 6. Conclusion: Honest, or NothingThe roadmaps have been revised outward, repeatedly, from "five years" to "ten years" to "the 2030s" --- and those timelines assume flawless execution of error-correction protocols that currently consume thousands of physical qubits per logical qubit.
from AbstractThe Landauer bound does not privilege quantum over classical computation.
from 1.3 The Limits (Phase III)
Where to read
- 1. What We Have Learned504 words · 2 min
- 2. The Five Principles of Honest Computation624 words · 3 min
- 3. The Honest Portfolio: A Concrete Reallocation232 words · 1 min
- 4. A Call to Action538 words · 2 min
- 5. The Falsification Pledge266 words · 1 min
- 6. Conclusion: Honest, or Nothing377 words · 2 min
Read from the paper itself: 25 sections, 2,993 words. Nothing in this digest is written by a model; a line that is not a sentence of the paper is not shown. Have a specific question? Ask this paper.
The full text · Ask the whole corpus
Keep exploring
One body of research, three doors.