As I was reading Karen Hao’s magnificent Empire of AI I came across an interesting footnote—the tech world is frequently publishing on the preprint server arXiv, and foregoing peer reviewed publication. Hao frequently consulted this server to keep abreast of developments on the AI world.
arXiv was established in 1991 and covers physics, math, and computer science-related topics. It is not peer-reviewed—the manuscripts are not checked for accuracy, logic, or scientific soundness. And its website interface literally looks like it has not been updated since 1991.
In Empire of AI, Hao repeatedly asserted that this fast-paced scientific publishing was surpassing and crowding out, longer, better-vetted academic studies. This publication trend further exemplifies the carelessness of the Big Tech world that would rather beg forgiveness than ask permission.
Yet, the preprint is gaining traction in the scientific world outside of AI. Once, the preprint was the prelude to the published peer-reviewed paper, a formality to bypass. Since the Covid-19 pandemic, when academic journals were bottlenecked (since they rely on unpaid peer-reviewers), preprints have rocketed into the public eye. And they fed into the public appetite for free, un-paywalled Scientific content—but at what cost?
The Brokenness of Peer-Review
Peer reviewers are selected by journal editors often by cold-emailing individuals with relevant expertise. When I’ve submitted manuscripts, I’ve also been asked to list potential peer reviewers, and literally I do a Google Search.
Rejection rates for review invitations are high. Those who do agree are often overburdened faculty, who may delegate reviews to graduate students in the guise of instructing them about the publication process. While journals provide general guidelines, reviewers are largely expected to understand the process independently. In my own experience assisting with peer review for my PI, I Googled how to conduct a review effectively. Journals may or may not choose to hide the author’s names from a reviewer, and in many cases, I could see the list of authors’ names.
It is very difficult to combat the naked bias baked into the peer-review system that disadvantages those who are not well-networked in the scientific community. Multiple studies have proved that authors with “prestigious” affiliations are more likely to be accepted for publication regardless of journal type. In single-blind review processes, where reviewers are aware of the author names and affiliations, papers submitted to Computing Machinery International Conference were more likely to be recommended for acceptance from top authors, universities, or companies (Tomkins, Zhang, and Heavlin 2017).
The “friendship” effect (prior affiliation or proximity between authors and reviewers) has been documented in neuroscience manuscripts submitted to PLOS One (2011–2012). Specifically, authors with closer connections to reviewers received an average 0.11-point increase on a 4-point review scale (Tomkins, Zhang, and Heavlin, 2017).
Another study re-submitted 12 high-impact published research articles affiliated with prestigious institutions, under fictious names and institutions (Peters and Ceci 1982). Eight out of 9 that were selected for peer review were rejected.
This trickles down to lowered citations and publications for early career researchers, often times women or people of color. Given the extreme selectivity of the tenure track, women and people of color are less likely to be tagged as reviewers. Arguments against compensating authors and reviewers often rely on the magical-thinking assumptions of objectivity and neutrality in science. Given the obscure and tenuous aspect of the publication process, fallacious papers are published and publicly praised in peer-reviewed journals all the time [1] (Glenza 2022; Piller 2022).
This bias has not been remedied by the rise of so-called ‘open access’ publishing.
The Rise of Open Access
In the 1990s, “open access” journals emerged, touting their potential to “democratize” scientific publication, since many journals have hefty subscription fees and paywalls. Hence, any person with Internet can click on the article and read it. By 2011, approximately 11% of the world’s scientific articles were published in open-access journals (Van Noorden, 2013).
While, open access journals are free to read, they charge authors thousands of dollars for publication. They are still peer-reviewed, and many of these journals are under the umbrella of a legacy publication like Elsevier. Often, a university will be double-charged by publisher: for open-access fees and then for access to the paywalled publications to be hosted in their university library.
Acceptance rates vary widely across open-access journals and thier promise to ‘democratize’ publication has not be fulfilled. The Public Library of Science (PLOS), founded in 2000, is a prominent example. In 2013, with their publication fee of $1,350, about 70% of submitted articles were accepted. The Physical Review Letters with a $2,700 fee accepted roughly 35%. In contrast, Nature, which is not open access, had an acceptance rate of about 8% (Van Noorden, 2013).
A 2018 study examined the role of gender and nationality in peer review outcomes at eLife, an open-access journal in the life and biomedical sciences. Of 23,867 submissions from 2012 to 2017, the acceptance rate was a selective 14%. However, acceptance rates favored male authors, particularly when reviewed by all-male panels. There were advantages based on nationality—especially when at least one reviewer shared the same country as the last or corresponding author. Women and authors from outside North America and Europe were underrepresented among accepted papers (Murray et al., 2018).
This bias can overlook scientifically unsound research. John Bohannon wrote and submitted a fake “wonder drug” paper with obvious errors and unsound conclusions to 300 open-access journals, with about half accepting them, including Sage and Elsevier journals (Bohannon 2013). Ultimately, prestige attracts prestige, where “prestigious” scientists are more likely to attract more resources, graduate students, and citations.
The University of California, San Francisco, Library collections posted changes in average journal pricing over time. They report: '“The UCSF Library has struggled over the last decade to maintain our journal subscriptions as well as add new resources to our collections, much to the frustration of UCSF’s prolific scholars at this world-renowned institution.”
Publication fees & subscription fees provide a barrier to submission and acquisition
“Open-access” journals, rather than democratizing the publication process have shifted the burden of payment to the author, who might be desperate for tenure or academic promotion. This disadvantages marginalized and early scholars—not only the fees, but the lowered visibility can result in lowered citation counts. Also, if they can only ‘afford’ to pay for paywalled journals, this makes it less likely for libraries to acquire the scholarship.
The Dutch publisher Elsevier maintains a near-monopoly in scientific publishing, managing over 3000 open-access and non-open access journals as of 2026. Publication fees in Elsevier journals can exceed $9,000 (Elsevier, 2023) with additional charges for features such as color figures. A 2017 Guardian investigation revealed that Elsevier had profit margins of 36%, bigger than Apple. And Elsevier submissions have only continued to increase in the intervening years.
Unbelievably, APS journals may charge up to $800 for a single color figure (APS, 2023). Beyond publication costs, publishers also charge institutions substantial subscription fees to access content which was often produced by those same institutions. Paying for individual academic articles can cost $30–50 a pop (Hagve, 2020).
Pre-prints, the Rogue Disrupter
Given the pay-to-play model that rapidly reshaped scientific publishing in the 2010s, the field is now ripe for disruption. This has emerged in the form of cost-free preprint servers such as arXiv, bioRxiv, and medRxiv, the giants of the pre-print world. An early version of a manuscript might be posted on the preprint server to establish the topic’s novelty by researchers concerned about other labs ‘scooping’ their ideas; to ensure the work is publicly available in some form online; and to obtain early feedback. This shift is appealing particularly for those in non-academic settings who cannot afford to subscribe to legacy journal libraries like Wiley or Elsevier or pay $50 per article hidden behind a paywall.
Screenshots of bioRxiv, medRxiv, and arXiv pre-print sever cover pages (April 11, 2026). Notably, only medRxiv displays a prominent red ‘warning label’ to viewers. In contrast, arXiv includes a more understated disclaimer—‘Materials on this site are not peer-reviewed by arXiv’—without further clarification. Both bioRxiV and medRxiv thank The Chan Zuckerberg Initiative for financial support.
By the late 2010s, pre-prints were perceived as a way to boost a researcher’s citations: high citations are often a common requirement within the academic research community for promotion. A 2019 bioRxiv article seemingly confirmed that preprints can significantly enhance citation counts for researchers, despite only 67.7% of 2013-2017 pre-prints being published in peer-reviewed journals. However, non-academics might not be able to distinguish the difference between peer review and pre-prints. Only medRxiv prints a warning label on the top of their site:
Caution: Preprints are preliminary reports of work that have not been certified by peer review. They should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
A 2026 peer-reviewed study in Research Integrity and Peer Review provided an overview of the preprint-to–peer-reviewed publication pipeline. The authors found a median lag of 11.5 months between preprint posting and searches for a corresponding peer-reviewed publication. Only a median of 42% preprints were eventually published. The study also found that citation growth for preprints was substantially smaller than for peer-reviewed publications, suggesting that citation benefits from preprints are more modest than often touted.
bioRxiv, and medRxiv have fewer than a dozen staff who “lightly vet” these papers. Editing is minimal, and rigorous scientific examination of manuscripts is non-existent. bioRxiv reports a 2-day turnaround process for publication.
bioRxiv medRxiv, and arXiv also have loose and inconsistent mechanisms for the disclosure of conflicts of interest during the submission process, as well as for ethical approval of human subjects and funding disclosures that are not verified.
Their use surged during the COVID-19 pandemic, as researchers rushed to share findings that could accelerate medical treatments and public health responses. medRxiv, which had been fortuitously launched in 2019, experienced an exponential leap in downloads after March 2020. The Chan Zuckerberg Initiative Awarded $2 Million to medRxiv in 2020.
Even though COVID-19-related preprints on medRxiv declined in the next few years from 80% to just 7%, other topics grew; there were 12,863 preprints posted on medRxiv in 2024 close to the 2020 peak, while bioRxiv’s preprints exceeded its pandemic high. Recently, medRxiv and bioRxiv have joined forces to form a non-profit, called OpenRxiv perhaps ushering in a new era of move fast and break-things publication (note that OpenAI started as a non-profit too).
Even with these fees, there is minimal personalized guidance for authors in the publication process. The authors are still responsible for adhering to the journal’s strict formatting guidelines for both the text and figure generation. The paper must also be written to a high, uniform standard of syntactical and grammatical correctness. If the paper fails the “pre-screening” process, which involves determining whether the paper fits the journal’s aim and scope and adheres to formatting guidelines, it is then sent to peer review. If detailed feedback or amendments are required to the paper, it is rare that the author is assisted in revision. And if rejected, the paper is sent off to another journal with different formatting standards, and a new round of peer-review must be conducted. Non-English speakers usually have to pay for extra editing. Furthermore, authors are largely responsible for promoting and publicizing their own paper.
So why not just put it on a preprint server?
No editing. While the peer-review system is broken, removing editing and review is the wrong move.
They do not meet standards of scientific responsibility to society.
They do not necessarily accelerate discovery; a common but weak argument is that, for example, a startup in San Francisco can immediately build on an MIT preprint on arXiv, yet there is limited evidence that this occurs in practice.
They can contribute to the spread of misinformation.
Post-publication peer review alone is often inadequate.
They can contribute to the spread of misinformation.
Post-publication peer review alone is often inadequate.
They could be faked by AI or bad actors.
The way forward
By 2023, a group of neuroscientists, frustrated over high publishing fees and charges, formed Imagining Neuroscience through the nonprofit MIT press. This began in 2022 when the editors of Neurolmage asked Elsevier to lower the article processing charge from $3.450 to under $2000. The editors resigned and found those Imagining Neuroscience. This came at a time when the Biden administration issued guidance that ferally funded research projects be made freely and immediately available to the public (Zahneis 2022). As of 2026, it is on Volume 4 and rivalling its predecessor. However, despite their extensive list of internal editors, they do rely on external, unpaid reviewers. And they charge $1,400 per article but do offer fee wavers in some circumstances.
Even the noblest ideals in scientific publishing fall short.
Scientific Publishing should be:
Readily available and accessible to the public
Efficient and timely in publishing
Rigorously vetted
Accurate
Embracing diverse voices and perspectives.
Non-profit
Compensate fairly for labor
Support and amplify authors
Transparent and reproducible
[1] In 2022, a highly cited Alzheimer’s paper, published in Nature in 2006 was found to have had fraudulent data and tampered images. Hundreds of millions of NIH dollars went to funding project proposals that were based on the science of this paper. Skeptics of the 2006 paper were dismissed for years before the fraudulence was suspected in 2021. When the fraudulence was discovered, the FDA initially refused to halt clinical trials of drugs that were based on the science from this paper. Ultimately, this fraudulence set back Alzheimer’s research by years. I wrote about it here.
Please consider following me on my social media platforms:
References
APS. 2023. “Announcement: New Prices and Payment Policy for Color Figures in the Print Journal.” PHYSICAL REVIEW LETTERS. https://journals.aps.org/prl/edannounce/PhysRevLett.98.260001 (May 7, 2023).
Bohannon, JOHN. 2013. “Who’s Afraid of Peer Review ?” Science 342(October): 60–65. https://www.science.org/doi/10.1126/science.342.6154.60
Elsevier. 2023. “Article Publishing Charge (APC) Price List.” Elsevier. https://www.elsevier.com/__data/promis_misc/article-publishing-charge.pdf (June 9, 2023)
Glenza, Jessica. 2022. “Critical Elements of Leading Alzheimer’s Study Possibly Fraudulent.” The Guardian. https://www.theguardian.com/society/2022/jul/23/alzheimers-study-fraudulent
Hagve, Martin. 2020. “The Money behind Academic Publishing.” Tidsskrift for Den norske legeforening 140(11): 1–5. https://tidsskriftet.no/en/2020/08/kronikk/money-behind-academic-publishing
Merton, R. K. 1968. “The Matthew Effect in Science: The Reward and Communication Systems of Science Are Considered.” Science 159(3810): 56–63.
Murray, Dakota et al. 2018. “Gender and International Diversity Improves Equity in Peer Review.” bioRxiv: 400515. https://www.biorxiv.org/content/early/2018/08/29/400515
Peters, Douglas P., and Stephen J. Ceci. 1982. “Peer-Review Practices of Psychological Journals: The Fate of Published Articles, Submitted Again.” Behavioral and Brain Sciences 5(2): 187–95.
Piller, Charles. 2022. “Blots on a Field?” Science 377(6604): 358–63.
Tennant, Jon. 2018. “Elsevier Are Corrupting Open Science in Europe.” The Guardian (June 29). https://www.theguardian.com/science/political-science/2018/jun/29/elsevier-are-corrupting-open-science-in-europe.
Tomkins, Andrew, Min Zhang, and William D. Heavlin. 2017. “Reviewer Bias in Single- versus Double-Blind Peer Review.” PNAS 114.
Van Noorden, Richard. “Open Access: The True Cost of Science Publishing.” Nature, vol. 495, no. 7442, 2013, pp. 426–29, doi:10.1038/495426a
Zahneis, Megan. 2022. “‘A Historic Moment’: New Guidance Requires Federally Funded Research to Be Open Access.” The Chronicle of Higher Education. https://www.chronicle.com/article/a-historic-moment-new-guidance-requires-federally-funded-research-to-be-open-access?sra=true&cid=gen_sign_in
Oh god! Finally someone writing about this! Thank you, thank you, thank you!