Foreword

Along with growing computing power and digital storage capacity of the hardware, the AI technology has entered a new phase of development in both the academic and business worlds. The advent of ChatGPT at the end of 2022 is a landmark event related to the development of AI. Nature has recently published two articles discussing the impact of ChatGPT on academia so far and suggesting possible countermeasures. In this article, we will discuss the opportunities and challenges brought by ChatGPT to scientific researchers in the light of latest studies and discussions on ChatGPT. Any comment from the readers will be appreciated.

I. What benefits are brought by ChatGPT to scientific researchers?

(i) Work more efficiently

ChatGPT is a large language model (LLM), a machine-learning system that autonomously learns from data and can interact naturally with users after training on a massive data set of text. It is able to "answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests".

Usually, researchers spend a lot of time on literature search, experimental data analysis, report writing and other tasks. ChatGPT can quickly find relevant information from the massive literature database and quicken the research process. Moreover, ChatGPT can also automate the process of experimental design and data analysis, helping researchers obtain experimental data and move on to the next step faster. Since the advent of ChatGPT, more and more researchers have started to use it to write literature reviews and draft papers, find any loophole in the design of experiments and even polish their papers. In this way, scholars can save a lot of time spent on doing literature search, writing literature reviews, making comparisons, and drawing conclusions, and shorten the time before their research findings can be published. The resulting uplift in their efficiency is tangible.

(ii) Spur scientific innovation and tap into unexplored fields

Because ChatGPT enables researchers to write more fluently and work more efficiently, an immediate effect is accelerated innovation. Since part of the basic tedious work is taken over by AI, scholars have more time to focus on designing new experiments and researching new topics and have the possibility to deliver more breakthroughs in science.

As early as in 1911, some pioneering academic paper pointed out that "intelligent partnerships" between men and intelligent technologies could outdo human intelligence alone. Now the emergence of ChatGPT foreshadows the probability of the said intelligent partnerships excelling human capabilities and advancing innovations in scientific research to a level that we could not imagine. For example, in medicine, ChatGPT can assist doctors in making diagnosis and treatment decisions and support pharmaceutical research & development; in biology, ChatGPT can help researchers in areas such as gene editing and DNA sequencing; in addition, ChatGPT can be used in conjunction with other technologies, such as computer vision and robotics, to provide all-around support for research in various fields, bridge interdisciplinary research, and encourage mutual understanding and collaboration between different disciplines.

(iii) Increase research resources available for use

As a model that can convincingly interact with users in English and other languages on a wide range of topics in natural sciences, sociology, and more, ChatGPT is not only available for free, but also accessible and easy to use. If used properly, ChatGPT can help create an equitable academic climate by, for example, removing language barriers for scholars and enabling more scientists who do well in research to write academic papers of better quality. When it comes to scientific research, ChatGPT is a rich and robust reservoir of resources. In general, it can provide assistance in the following aspects:

II. What risks and challenges will be posed by using ChatGPT as a research assistant?

(i) Threaten to distort scientific facts and spread misinformation

While ChatGPT, a conversational AI, has potential enough to be a good "research assistant," participating in multiple stages of the research process and giving help, exactly for the same reason, it threatens to undermine the transparency and quality of the research process by researchers. A critique in Nature said, " ChatGPT and other LLMs produce text that is convincing, but often wrong, so their use can distort scientific facts and spread misinformation."

Some scholars have found, in asking ChatGPT questions, that it often produces false and misleading texts for questions that require a further look-up into the literature or deep-going research. These texts are seemingly convincing and well-founded at first glance, but do not stand up to further scrutiny and proofing. Wrong answers to researchers’ questions contain several factual mistakes, misrepresentations, and erroneous data. These problems prove that the use of conversational AI in academic research has the potential to give rise to factual mistakes, deepen researchers' biases about an issue or even derail them from the right research direction. It would be a disaster for the academic community if researchers were confused by ChatGPT-generated answers with flowery rhetoric and incorporated misinformation into scientific findings and published them.

(ii) Call originality in question and beget academic dishonesty

It is widely believed in academia that AI cannot be the author of an academic paper because it is impossible to hold AI responsible for the contents it produces. However, as many researchers involve AI increasingly in their research process, in some cases, ChatGPT may even be able to complete the important parts of an academic paper based on the guidance and hints received from the researcher. In this case, it is difficult to make a distinction between the engagement and contribution of the researcher and AI to the research results. Moreover, given the limitations of the existing censorship practices and procedures, there is a big chance that reviewers or the specially designed censorship procedures fail to completely or accurately identify whether and to what extent researchers have used ChatGPT. This problem will challenge the recognition of originality of research results. In addition, these ambiguities in terms of originality, authorship identification, and sources of cited information may also provide a hotbed of academic dishonesty. The importance and urgency of this issue is well proven by the fact that some European and American universities have announced a ban on the use of ChatGPT in students’ academic papers or coursework.

(iii) Make researchers’ creativity degenerate

ChatGPT has the potential to erode the free will of researchers. Once researchers have personally enjoyed the benefits brought by ChatGPT, it is very likely that they will develop an addict to this technology. With the aid of ChatGPT, researchers’ abilities in literature retrieval, data statistics and analysis, and other fields, may be weakened or even gradually disappear. The degeneration of scholars' abilities in these areas may not be critical and will not be fatal to the advancement of science and technology, but there are still a lot of skills that are indispensable to researchers, such as generating ideas, designing experiments and practicing them, writing papers and publishing research findings. If AI goes unchecked to take over more work from scholars, as AI evolves further, researchers' personal contribution to academic results may become increasingly limited and vague. Although there is a long way to go to reach there, how to keep the creativity of researchers from degeneration is an important issue faced by the academic community.

Conclusion

Given the tremendous convenience ChatGPT can bring to scientific work, the use of ChatGPT by researchers is an inevitable trend, and it is barely possible for the academic community to ban or avoid the use of this technology. That said, it is inappropriate to allow researchers to indulge in using ChatGPT. In the future, we should stick to strict censorship of research results, and even consider adding special steps to verify the participation of ChatGPT and the accuracy of the information it contributes. We should uncompromisingly uphold the principle and concept that researchers are responsible for their research results and ensure transparency in scientific research.

The academic community should grab at the opportunities while managing the risks properly. From researchers to universities, publishers of academic papers, every participant in scientific work should take responsibility for their work. As Dostoyevsky said, "Everyone is responsible for everyone and everything". We believe that science will eventually find a way to benefit from conversational AI.

References:

1.Eva A. M. van Dis. ChatGPT: five priorities for research. Nature 614, 224-226 (2023).

2.Chris Stokel-Walker. ChatGPT listed as author on research papers: many scientists disapprove. Nature 613, 620-621 (2023)

3.Salomon, G., Perkins, D. N. & Globerson, T. Edu. Res. 20, 2–9 (1991).