Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Comment

Generative artificial intelligence and scientific publishing:


urgent questions, difficult answers
Azeem Azhar describes, in Exponential: Order and generative AI undermines education and is nothing more
Chaos in an Age of Accelerating Technology, how human than high-tech plagiarism, and many feel similarly about
society finds it hard to imagine or process exponential AI models trained on work without upholding copyright.9
growth and change and is repeatedly caught out by this Plagiarism is a problem in scientific publishing, but those
Yuichiro Chino/Getty Images

phenomenon.1 Whether it is the exponential spread of a concerned with research integrity are also considering
virus or the exponential spread of a new technology, such a post-plagiarism world,10 in which hybrid human–AI
as the smartphone, people consistently underestimate writing becomes the norm and differentiating between
its impact. Azhar argues that an exponential gap has the two becomes pointless. In the ideal scenario, human
Published Online developed between technological progress and the pace at creativity is enhanced, language barriers disappear, and
March 6, 2024
https://doi.org/10.1016/
which institutions are evolving to deal with that progress. humans relinquish control but not responsibility.10 Such
S0140-6736(24)00416-1 This is the case in scientific publishing with generative an ideal scenario would be good. But there are two urgent
artificial intelligence (AI) and large language models questions for scientific publishing to address.
(LLMs). There is guidance on the use of generative AI from First, how can scientific publishers and journal editors
organisations such as the International Committee of assure themselves that the research they are seeing is real?
Medical Journal Editors.2 But across scholarly publishing Researchers have used generative AI to create convincing
such guidance is inconsistent. For example, one study fake clinical trial datasets to support a false scientific
of the 100 top global academic publishers and scientific hypothesis that could only be identified when the raw
journals found only 24% of academic publishers had data were scrutinised in detail by an expert.11 Papermills
guidance on the use of generative AI, whereas 87% of (nefarious businesses that generate poor or fake
scientific journals provided such guidance.3 For those scientific studies and sell authorship) are a huge problem
with guidance, 75% of publishers and 43% of journals and contribute to the escalating number of research
had specific criteria for the disclosure of use of generative articles that are retracted by scientific publishers.12 The
AI.3 In their book The Coming Wave, Mustafa Suleyman, battle thus far has been between papermills becoming
co-founder and CEO of Inflection AI, and writer more sophisticated in their fabrication and ways
Michael Bhaskar warn that society is unprepared for the of manipulating the editorial process and scientific
changes that AI will bring.4 They describe a person’s or publishers trying to find ways to detect and prevent
group’s reluctance to confront difficult, uncertain change these practices. Generative AI will turbocharge that race,
as the “pessimism aversion trap”.4 For journal editors but it might also break the papermill business model.
and scientific publishers today, this is a dangerous trap When rogue academics use generative AI to fabricate
to fall into. All the signs about generative AI in scientific datasets, they will not need to pay a papermill and will
publishing suggest things are not going to be ok. generate sham papers themselves. Fake studies will
Within months, LLMs could pass medical examinations exponentially surge and nobody is doing enough to stop
and have now been trained to evaluate text medical this inevitability.
conversations with trained actors as patients.5 Preliminary Prevention lies in big and small changes that will be costly
research indicates generative AI shows greater diagnostic and difficult. A major step is to recognise that although
accuracy than clinicians in some diseases and better few people can disagree with the open science movement
empathy for text conversations than their human and its many benefits,13 the implementation of the open
counterparts.6 In 2023, Springer Nature became the first access business model in scientific publishing has generally
scientific publisher to create a new academic book by worked in opposition to limiting fabrication and research
empowering authors to use generative AI.7 Researchers misconduct. To maintain profits and a sustainable business
have shown that scientists found it difficult to distinguish responsible to shareholders, scientific publishers have
between a human generated scientific abstract and one focused on the quantity of papers over quality, because
created by generative AI.8 Noam Chomsky has argued that the main revenue has been through article-processing

1118 www.thelancet.com Vol 403 March 23, 2024


Comment

charges (APCs)—the more papers that are published, and hallucinations or not having enough training data
the more APCs publishers secure.14,15 This approach in the system to identify a scientific advance.22–24 And
needs to change. Part of that process will involve finding although some commentators have argued that it is
new ways to finance the open access model, of which surely easier to remove bias from an algorithm than
there are some credible alternatives, such as new types from a person,25 there remain many aspects to a journal
of group deals with countries and institutions that that still require human general intelligence. Journals are
remove the focus from APCs.16 Another major step will much more than the research they publish. At this stage,
be reforming academic environments to reward quality AI can help improve the efficiency, transparency, and
research over quantity and diminish the link between objectivity of peer review26 to help human editors make
publication and promotion, with efforts underway better decisions, but it cannot replace human decision
towards this goal.17,18 Some of the smaller actions involve making. At their best, scientific journals are a creative,
technology to prevent fraud, including developing AI inclusive, and leading expression of science woven
tools that can identify discrepancies and errors in raw together with the richness of human interpretation. To
datasets, increasing the use of tools that detect image live up to this ideal—and to survive—research integrity,
manipulation and duplication,19 and developing security risk management, and the value of human thought
log-ins for authors that confirm they are humans and leadership must be every journal’s new priority.
linked to an institution. Other solutions lie in more I am Senior Executive Editor at The Lancet.
robust editorial processes that scrutinise studies with Jessamy Bagenal
fabrication in mind and detect signs of data fabrication. jessamy.bagenal@lancet.com
A transformative re-evaluation of what can be trusted The Lancet, London EC2Y 5AS, UK

in the scientific literature should mean that more value 1 Azhar A. Exponential: order and chaos in an age of accelerating technology.
London: Penguin, 2022.
is placed on journals and publishers that prioritise 2 International Committee of Medical Journal Editors. Updated ICMJE
recommendations (January 2024). https://www.icmje.org/news-and-
trustworthiness. editorials/updated_recommendations_jan2024.html (accessed
Second, to what extent do we want scientific journals Feb 26, 2024).
3 Ganjavi C, Eppler MB, Pekcan A, et al. Publishers’ and journals’ instructions to
to continue being for human readers and led by human authors on use of generative artificial intelligence in academic and scientific
editors? Selecting, reviewing, curating, and optimising publishing: bibliometric analysis. BMJ 2024; 384: e077192.
4 Suleyman M, Bhaskar M. The coming wave. London: Bodley Head, 2023.
scientific content in journals has always been a human 5 Kung TH, Cheatham M, Medenilla A, et al. Performance of ChatGPT on
endeavour. Scientific journals change personality from USMLE: potential for AI-assisted medical education using large language
models. PLoS Digit Health 2023; 2: e0000198.
one editorship to the next. They rightly come under 6 Van den Heuvel M. AI excels in patient interaction and diagnosis in pilot
scrutiny for matters relating to diversity and inclusion, study. Medscape. Jan 30, 2024. https://www.medscape.com/viewarticle/ai-
excels-patient-interaction-and-diagnosis-pilot-study-
because journals so acutely represent the priorities 2024a100022e?form=fpf (accessed Feb 26, 2024).
7 Springer Nature Group. Springer Nature and authors successfully use
and interests of editors. But alongside the human generative AI to publish academic book. Springer Nature. Oct 18, 2023.
face of journals, the scientific community also faces https://group.springernature.com/gp/group/media/press-releases/first-ai-
generated-book/26189712 (accessed Feb 26, 2024).
an enormous information overload challenge with far 8 Else H. Abstracts written by ChatGPT fool scientists. Nature 2023; 613: 423.
too much research waste.20 This is a problem human 9 Stewart J. Noam Chomsky says ChatGPT is “high-tech plagiarism”.
My Modern Met. Feb 17, 2023. https://mymodernmet.com/noam-
editors and publishers have not been able to solve. AI is chomsky-chat-gpt/ (accessed Feb 26, 2024).
well placed to wade through millions of data points and 10 Eaton SE. Postplagiarism: transdisciplinary ethics and integrity in the age of
artificial intelligence and neurotechnology. Int J Educ Integr 2023; published
decide what is relevant. Pharmaceutical companies are online Oct 12. https://doi.org/10.1007/s40979-023-00144-1.
already using AI to trawl through thousands of abstracts 11 Naddaf M. ChatGPT generates fake data set to support scientific hypothesis.
Nature 2023; 623: 895–96.
and full-text articles to gain new insights into diseases.21 12 Van Noorden R. More than 10,000 research papers were retracted in 2023—a
new record. Nature 2023; 624: 479–81.
What if the solution to our bloated scientific information 13 Rafols I, Meijer I, Molas-Gallart J. The benefits of open science are not
problem is to use AI as the main readers and curators of inevitable: monitoring its development should be value-led. LSE Impact Blog.
Aug 14, 2023. https://blogs.lse.ac.uk/impactofsocialsciences/2023/08/14/
scientific advances? Robot journals for robot readers? the-benefits-of-open-science-are-not-inevitable-monitoring-its-
Nobody can predict the future, but for now, robot development-should-be-value-led/ (accessed Feb 26, 2024).
14 Horton R. Offline: The scramble for science. Lancet 2022; 400: 983.
editors would be a mistake. There are important 15 Ioannidis JPA, Pezzullo AM, Boccia S. The rapid growth of mega-journals:
concerns about the use of closed LLMs for peer review threats and opportunities. JAMA 2023; 329: 1253–54.
16 Sanderson K. Who should pay for open-access publishing? APC alternatives
and editorial processes in terms of compounding bias emerge. Nature 2023; 623: 472–73.

www.thelancet.com Vol 403 March 23, 2024 1119


Comment

17 The Lancet. Safeguarding research integrity. Lancet 2024; 403: 699. 23 Hosseini M, Horbach SPJM. Fighting reviewer fatigue or amplifying bias?
18 Declaration on Research Assessment (DORA). About. 2024. https://sfdora. Considerations and recommendations for use of ChatGPT and other large
org/ (accessed Feb 26, 2024). language models in scholarly peer review. Res Integr Peer Rev 2023; 8: 4.
19 Jones N. How journals are fighting back against a wave of questionable 24 Stokel-Walker C, Van Noorden R. What ChatGPT and generative AI mean for
images. Nature 2024; 626: 697–98. science. Nature 2023; 614: 214–16.
20 Kleinert S, Horton R. How should medical science change? Lancet 2014; 25 Lobel O. The equality machine: harnessing digital technology for a brighter,
383: 197–98. more inclusive future. London: Hachette UK, 2022.
21 The Medicine Maker. How AI is changing pharma’s approach drug 26 Bhosale U, Kapadia A. AI integration can future-proof peer review—but only if
development. The Medicine Maker. July 7, 2023. https://themedicinemaker. we define and drive the right principles. Enago Academy. Sept 29, 2023.
com/discovery-development/how-ai-is-changing-pharmas-approach-drug- https://www.enago.com/academy/ai-integration-in-peer-review/ (accessed
development (accessed Feb 26, 2024). Feb 26, 2024).
22 Donker T. The dangers of using large language models for peer review.
Lancet Infect Dis 2023; 23: 781.

UK Public Health Science 2024: a call for abstracts


Published Online We are delighted to invite abstract submissions for UK research bodies. This cross-sectoral approach will make
March 18, 2024
https://doi.org/10.1016/
Public Health Science: A National Conference Dedicated for a fascinating and valuable event where participants
S0140-6736(24)00527-0 to New Research in UK Public Health that will be held in can be inspired by novel world-class research.
Cardiff, UK, on Nov 29, 2024. This annual conference Alongside the overall focus on public health science, this
showcases leading, innovative public health research year’s conference will explore the connection between
from across the UK and Ireland with accepted submissions healthy neighbourhoods, the built environment, and
published online by The Lancet. The conference will bring public health from a multidisciplinary perspective.
together researchers, practitioners, policy makers, and Cities have a central role in shaping population health.1
experts by experience to share and discuss contemporary Understanding and optimising the built environment
issues, perspectives, and challenges in public health within urban settings is vital for cultivating health-
science and explore their impacts for practice, policy, promoting spaces. Submissions are invited that address
health services, and research. evidence-based interventions and policies relating
Abstracts are welcomed from people and groups to not only the physical aspects of neighbourhoods,
working across all sectors and disciplines on topics but also to the quality of the air, reduction of noise
related to any aspect of public health in the UK or Ireland. pollution, and mitigation of other environmental
Relevant research is invited from the National Health stressors.2–4 By highlighting these issues, the conference
Service (NHS), local authorities, charities and advocacy seeks to stimulate discussions that can inform policies,
services, government bodies, universities, or other interventions, and research initiatives dedicated to
fostering healthier communities and mitigating the
impact of environmental pollutants on population
health.
The conference will consist of oral paper presentations,
chaired poster sessions, and keynote speeches.
Submissions are now open for oral and poster
presentations. Abstracts can be submitted under any
Vitaly GolovatyUK/Amazing Aerial agency/Science Photo Library

of the three core categories: creativity in public health


science; new methodological approaches to public
health science; or impact and implementation of public
health science in policy and practice. Research described
in the abstracts may have been undertaken anywhere
in the world but must be relevant to public health
science, practice, and policy in the UK or Ireland. We
particularly encourage submissions that include experts
by experience or early career researchers.

1120 www.thelancet.com Vol 403 March 23, 2024

You might also like