The Dispatch
The price of innovative science
Novel research directions are declining on a global scale; with ongoing concerns over funding and peer-review alongside new challenges with ChatGPT’s use in science, who’s to blame?
By Julie Hoeflinger
“In general, in scientific research, the data and the facts matter more than being creative. But at the same time, being creative can lead you to some really good ways of visualising [the data] and also of being able to ask new questions.”
In a study published recently, ChatGPT outperformed university students in an undergraduate psychology exam. What’s more, the essays written by artificial intelligence went undetected 94 percent of the time by exam markers. Such studies seem to be growing in number, leading to strengthened fears that ChatGPT and other artificial intelligence technologies will be the death of human creativity.
Are we already becoming less creative problem solvers? In certain fields, we have seen a quantifiable reduction in innovation. A study published last year in Nature reported a decline in disruptive science over time, with disruptive science being defined as, “work that breaks with the status quo and carves out new directions for scientific inquiry”. Another paper published in 2020 also touches on the same issue, with the title, “Are ideas getting harder to find?”
The authors found that current research findings are “less likely to connect disparate areas of knowledge” and more likely to “rely on a narrower set of existing knowledge”, suggesting that one way researchers could be more productive is by using more diverse work.
However, other experts point to a flawed funding system and the growing bureaucracy of science. Despite grant applications eating up more and more of researchers’ time, grant approval rates continue to drop: the overall success rates for research grant applications hover just around 20 percent.
So, then, who’s to blame for declining creativity in research? ChatGPT? Scientists? Funders? Is it true that ideas are getting harder to find?
Pressure paves the road to AI’s assistance
Researchers face extreme competition, with the number of annual publications surpassing a record-breaking 9 million articles in 2023, and are under additional pressure to get their papers published in prestigious scientific journals in order to receive high evaluations and further funding. They also spend a hefty amount of time just doing administrative tasks, with grant writing taking percent of their time. As a result, many scientists are being drawn to ChatGPT.
Not long after ChatGPT became open to the public in 2022, scientific journals began seeing papers submitted with ChatGPT listed as a co-author. Several journals, including Springer-Nature and Science, quickly took action by prohibiting AI-generated text from being included in submissions. And for good reason: scientific papers that included ChatGPT’s writing posed a number of problems, such as the tendency of AI to combine sources, including factual inaccuracies and using repetitive, homogenised language.
Regardless, ChatGPT isn’t necessarily to blame for declining innovation in scientific research. Kanta Dihal, a Lecturer in Science Communication at Imperial College London with a specialisation in narratives about artificial intelligence, sheds light on the use of ChatGPT in research.
While noting the detriment of cases where ChatGPT has written entire papers, Dihal points out one stage of the research process that could potentially benefit from AI: grant writing.
“There's a lot of criticism about the grant application process stifling creativity and originality, especially super original, wildly out-there research,” Dihal explains. “But on the other hand, it also is a process that can be highly formulaic and that disadvantages people who don't write eloquently, especially if they have to write eloquently when English is their third or fourth or fifth language. And I think that there are some opportunities for generative AI, just like there were with the invention of word processors and programs like Grammarly, to level the playing field regarding applying for funding.”
Furthermore, ChatGPT can help brainstorm ideas at the beginning of the research process, facilitating the generation of more novel approaches.
In response to the pressing question of whether ChatGPT is causing a decrease in creativity, Dihal responds, “AI is only as creative as all the human creativity that it was trained on.”
While AI training on other AI-generated content may become a larger issue down the line, it doesn’t seem to be the blame for the current decline in scientific breakthroughs.
Fostering creativity in a data-based field
ChatGPT aside, intense pressure can stifle creative thinking. To address this, some scientists have taken matters into their own hands.
One such initiative is the Good Science Project at Imperial College London. The Good Science Project aims to foster conversation about modern research culture and gives scientists a collaborative, artistic space free from the typical restraints of research.
Elena Corujos-Simón, Translation and Research Manager of the Institute for Molecular Science and Engineering at Imperial College London, says: “This whole project is being able to bring researchers into a room and ask them to be creative and forget about all the constraints they normally have to face, like the constraint of time and short contracts. .
“There are no reviewers. There's no feedback. It's all very positive. So, I think in that sense, it's really [about] embracing creativity from otherwise research minds.”
Also called into the conversation was the value of science communication and the cyclical effect it can have on creativity in one’s research. “In general, in scientific research, the data and the facts matter more than being creative. But at the same time, being creative can lead you to some really good ways of visualising [the data] and also of being able to ask new questions,” she shares.
Not only can asking new questions lead to disruptive science, but it’s well known that raising awareness about a particular area of research through strong science communication can secure additional funding.
Yet despite the benefits of engaging the public with research, many scientists express challenges in their ability to do so, particularly that it’ll take away time and attention dedicated towards their research.
Hiring third party science communicators trained in translating complex science to a lay audience in a compelling, digestible way is a great alternative for scientists to raise awareness about their research. But of course, creating successful science communication campaigns requires more funding.
Yet calls for more funding are being met. UNESCO reported that investment in global research has grown faster than the economy. In the United States alone, funding for research increased by $51 billion in 2020 to $717 billion, and, just to maintain consistent growth in GDP per person, the amount of research effort must be doubled every 13 years to counteract the growing challenge of producing fresh ideas. The UK Parliament has similarly invested a record-breaking amount of money into research and development, with a goal of reaching £20 billion per annum financial year 2024/25.
Initiatives like the Good Science Project and science communication efforts along with ChatGPT’s ability to assist with novel idea generation and governments investing exponentially more funding in science should collectively lead to an increase in more creative research, then, which is sure to result in more scientific breakthroughs – right?
Creativity blocked by bureaucracy
Spoiler: the massive increases in funding haven’t yet led to a proportionate rise in scientific breakthroughs.
It turns out that proposing something far outside of what is typical is extremely risky and could potentially end up being an enormous waste of time, when time is of the essence. According to a paper published in 2016, more novel research proposals are linked to lower evaluations.
This suggests that scientists are being evaluated on their ability to secure funding rather than on the quality of their research. As a result, they may opt for safer proposals that are more likely to be approved, which contributes to an overall decline in the integrity of scientific research. American physicist Alvin Weinberg expressed this concern over half a century ago after the creation of the National Science Foundation in the US, stating, “science dominated by administrators is science understood by administrators” and “such science quickly becomes attenuated if not meaningless”.
For these reasons, many scientists complain that an increased research and development budget won’t put an end to the reduction in disruptive science, and some argue that it may even make the situation worse.
“The decline in innovative work is probably the product of many factors, only some of which we can control,” says Manuel Ribes in a 2023 article for the Bioethics Observatory at the Catholic University of Valencia, Spain. “Undoubtedly among them is the fact that current science is more dedicated to publishing papers and seeking funding than to doing the kind of in-depth work that leads to breakthroughs.”
Criticisms against the bureaucratisation of science is nothing new: scientists have been complaining about it for years. But it only seems to be worsening, with ever-growing paperwork requirements and poor allocation of funding accelerating the issue.
Stephen Turner and Daryl Chubin said it best in an article published in Scientific American in 2019: “We need to ask whether we have lost something precious: the possibility of young scientists with new ideas taking risks and pursuing hunches that advance discoveries, not just careers.”
Beyond funding: How UK Parliament is setting out to improve research productivity
In 2021, Professor Adam Tickell, the Vice-Chancellor at the University of Birmingham, led an independent review of research bureaucracy in the UK in order to “identify and eliminate unnecessary bureaucracy that hinders research productivity” to ensure that research funding is used more efficiently. The final report was submitted in 2022 and included key recommendations on how to decrease administrative tasks.
The government officially responded to the report earlier this year, announcing several reforms in alignment with the report’s suggestions. Among these reforms includes introducing new, shorter guidelines for preparing business cases, limiting them to 12 pages, along with other measures that will accelerate the approval process and reduce time spent on paperwork so that more can be spent on doing the actual research.
Additionally, in 2023, UK Parliament created the Department for Science Innovation and Technology, which sets out to reduce science bureaucracy by 2030, along with the Advanced Research and Invention Agency’s (ARIA), which plans to fund “high-risk, high-reward” scientific research.
All things considered, ChatGPT is certainly not the death of human ingenuity – so long as it’s being used in an ethically regulated way – as is the case for nearly all new technologies. Initiatives like the Good Science Project and the growing field of science communication point to the idea that people aren’t yet becoming less creative in the face of artificial intelligence.
And more funding hasn’t resulted in more novel research directions. Responsibility for the ongoing decline in disruptive research cannot be placed in the hands of scientists alone, but rather governments must take concrete actions to reverse this trend. While the UK’s R&D plan is a good start, further evaluation will be needed to determine whether it lives up to its promises of increasing scientific innovation in the UK.