The 'Cut 95% of Science' Argument Gets the Disease Right and the Cure Wrong

"Cut 95% of research and papers in academia, and it would make no difference to scientific progress." The claim circulates widely because it is half statistical fact and half emotional release. An analysis of roughly 172 million papers found that more than half have never been cited. The most common citation count for a paper is zero. Many papers are cited exactly once, by their own authors in another paper. Nobel laureate Eric Betzig put it more bluntly: if 95% of academia were abolished tomorrow, the world would notice nothing except that paper mills stopped operating. The emotional half is also real. Small modifications, repeated validations, and papers produced for graduation, promotion, and grant completion do fill journal pages. In 2026, large language models make it even easier to manufacture vast quantities of unread papers. The figure "95%" may soon be too low.

The right direction does not mean the right tool. "Cut 95%" is the right diagnosis attached to the wrong question. It turns a surgery that requires precision into indiscriminate bombing.

1. Zero citations do not mean zero value, but pollution must be separated from trial and error

Zero citations are normal, not exceptional. A large number of papers are published and then ignored. But there is a logical leap between "zero citations" and "harmless to cut." Scientific discovery is extremely unpredictable. Many papers that later had major impact were ignored when published, or rejected by top journals as "not important enough." Large numbers of seemingly minor incremental studies form the cognitive map by which later researchers judge "this road is blocked" or "there is an anomaly here." Edison said, "I have not failed. I've just found 10,000 ways that won't work." In research, that is not inspirational chicken soup; it is the cost structure. The "useless" data produced by trial and error have value because they eliminate wrong options and make later steps lighter.

Two things must be separated: redundant waste and garbage pollution. A zero-cited paper that honestly reports a negative result is fundamentally different from a paper that manipulates data to make a quota. The former is an early-warning signal for later researchers. The latter actively distorts knowledge, reinforces false models, and suppresses innovation. Cutting 95% indiscriminately will not precisely remove the tumor. It will destroy healthy tissue and the warning system along with it. Science is built through massive trial and error. But trial and error builds a map, not a garbage dump. Burning the map does not make the road easier to walk.

2. The root of low-value research is not "too much," but "coercion" and "fraud"

The first root of large-scale low-value research is the evaluation system. "Papers only, titles only, degrees only, awards only." Quantified evaluation has turned publication itself into the goal. When paper counts and journal rankings become hard currency for promotion and resource allocation, research is alienated from truth-seeking into quantity-seeking. Researchers are forced to become paper-production machines. The "up-or-out" pre-tenure system in universities compresses this pressure into extremely short cycles, forcing researchers to choose short, fast, safe topics. Betzig points to another absurdity: current grant application systems require scientists to promise in advance "what I can accomplish in the next three years," which fundamentally violates the nature of exploration. If results can be precisely scheduled, it is not exploration, only execution. The result: high-risk ideas cannot get funding. People write proposals based on results they have already obtained, and attach labels like "AI" or "quantum" to cater to hot topics.

The evaluation system is not the only lesion. For-profit publishing capital is another engine. The four major commercial publishers together earned more than $7 billion in 2024, with profit margins above 30%. The output of paper mills doubles roughly every 1.5 years. Publishers present themselves as saviors of research integrity, but as Csaba Szabo argues, they are in fact the biggest beneficiaries and enablers of the crisis. An industry that runs on publication volume has no internal incentive to kill publication volume. Szabo even argues that the for-profit publishing system must be ended. Not repaired, not polished: dismantled. Funding agencies should directly take responsibility for scholarly communication and data dissemination, build open infrastructure, and decouple certification from commercial brands.

There is also something deeper than institutional design: selective enforcement of norms. Academic ethics committees, peer review, and data-sharing requirements have long existed, but enforcement is selective and person-dependent. People who point out methodological problems are told they "don't understand the rules," are "too nitpicky," or "undermine solidarity." Those who report misconduct are sometimes pushed out harder than the misconducters themselves. The sink is right there, but whether people wash their hands depends on who they are. This produces a kind of black humor: hygiene standards are public and verifiable, yet some repackage "not washing" as "preserving probiotics." They describe rule-breaking as flexibility, innovation space, or disciplinary characteristics. The problem is integrity and power. The metaphor "toilet paper is mostly unused but cannot be omitted" is still too clean. A more accurate picture: methods are crude. Review runs on connections. Data get processed into shape. Conclusions follow a story. Everyone knows it is nonstandard but does it anyway. Sometimes people wash, sometimes they don't. When criticized, they say there are beneficial bacteria in it.

3. "Major breakthrough" cannot be the only yardstick

Can the value of research be judged only by whether it produces a "major breakthrough"? No. Using major breakthroughs as the sole yardstick is epistemologically arrogant and practically harmful.

The definition of a major breakthrough is retrospective. At the moment it happens, it is often not consensus but non-consensus. When the National Natural Science Foundation of China established "major non-consensus projects," it explicitly listed "high controversy" as a sentinel signal of major breakthroughs. If research is filtered through a sieve that demands "foreseeable major breakthroughs," work with disruptive potential will be systematically killed at the source. Precisely because it is disruptive and temporarily incomprehensible to peers, it looks like a waste of money. Betzig's criticism of peer review is sharp: peer review is not quality control at all. It is a tool by which the mediocre protect themselves from genius. When you have a revolutionary idea, you have no peers.

The value of research is distributed. A paper may provide a key negative result that saves a field years of detours. It may train a student who later does important work. It may be the starting point of a serendipitous discovery whose initial form looks anything but "major." PNAS research on the role of serendipity in scientific progress reminds us that many seeds of breakthrough look indistinguishable from weeds when planted. Admitting that "major breakthrough" is not the only standard is precisely how we protect the soil in which the important 5% grows, not how we grant legitimacy to data manipulation and mass-produced academic noise. Value pluralism is not a defense of garbage. It is a refusal to use a ruler so crude that it condemns what has not yet grown.

4. What should be cut is not "what looks unimportant," but three concrete lesions

Why is "cut 95%" the wrong question? Because it uses a crude arithmetic solution to evade a structural problem. What needs to be cut has never been "research that looks unimportant," but three things.

First, papers that actively commit fraud and the paper-mill networks that support them. This is pollution. Recover funding from clearly fraudulent projects, audit fraudsters' current grants, and make ethics committees and funding agencies actually function rather than exist only on paper. A report of misconduct sitting unopened in an ethics committee inbox is far more serious than a few extra low-quality papers.

Second, the systemic incentive to publish for publication's sake, knowingly producing no contribution while consuming public resources. Separate teaching from research. Let undergraduate assignments take the form of assignments rather than casual publications. Reduce the weight of publication counts in evaluating everything. Pull publication down from its position as "evaluation currency." This redirects money to projects that need it. Research funding and positions stay.

Third, a publishing model that runs on publication volume, enjoys profit margins above 30%, and externalizes integrity costs onto academia. Funding agencies can attach reproducibility, verifiability, and research-integrity requirements and push responsibility for best use of funds onto institutions. Establish independent training systems for "scientific discovery" and "scientific integrity," making experimental design, statistical analysis, data integrity, and independent review a professional capability. Replace high-margin commercial intermediaries with open preprints, community-managed peer review, and data/code repositories. Wellcome tried this and failed for lack of teeth. The lesson is not that it can't work. The lesson is that voluntary measures don't.

Conclusion: Don't let righteous anger become the wrong knife

Behind the slogan "cut 95%" there is anger aimed at something specific. It is not anger at scientific trial and error itself, but at trial and error being alienated into performance, publication into currency, and integrity into "not understanding the rules." That anger is right.

Righteous anger must not become the wrong knife. Science needs redundancy and large numbers of inconspicuous small studies slowly surfacing. What science does not need is active fraud, institutionally enforced mass production, and a publishing machine that profits from dysfunction. What should be done is not to wave a giant cleaver at academia and shout "cut 95%," but to aim a scalpel at the three lesions: fraud networks, coercive production incentives, and for-profit publishing. Cut fraud. Cut noise. Redirect resources to exploration that has a chance of going somewhere. Stop treating people who wash their hands as troublemakers. This is far harder than a satisfying numerical slogan. But the 5% needs soil. Right now we're paving over it. The sink is right there. People walk past it every day.