The paper by Cavalcante-Silva et al.[1] is highly significant. They have addressed the merits and demerits of generative
artificial intelligence (AI), such as ChatGPT (OpenAI Inc., San Francisco, CA, USA),
in academic writing. Since ChatGPT's introduction, I have consistently reviewed papers
focused on this topic.
While researchers who emphasize the merits of ChatGPT claim that its use in writing
is time-saving and can lower the language barrier for non-native English authors,
those who focus on its demerits point out issues such as ChatGPT's inaccuracies and
the risk of unintentional plagiarism.[2] Studies on ChatGPT in writing have focused on specific specialties, for example,
gynecologists focused on ChatGPT-writing of gynecologic conditions, and psychiatrists
covered psychiatric conditions. However, there are only a few truly specialty-specific
issues in ChatGPT writing, leading to similar contexts and findings across different
fields. Discussions about detecting ChatGPT-generated manuscripts and the ethical
implications of its use further complicate the debate. The experiments conducted so
far are monotonous: “input this and then output that”, resulting in repetitive discussions.
Cavalcante-Silva et al. have moved beyond such monotonous discussions. By stepping
out of the conventional framework, they made important statements that may only be
fully appreciated by those with deep knowledge of the subject. I would like to highlight
these points and offer some additional thoughts. First, they insightfully noted that
“the ability to ask the right question and give the correct command may be the human
skill that will be most valued in the AI era”. Secondly, they raised concerns that
ChatGPT-writing might impact our ability to read, comprehend information, and critically
analyze it. Third, they argued that clear rules and objectives must be established
for the use of AI, including ChatGPT.
First, as previously reported, ChatGPT can generate readable manuscripts, including
letters[3]
[4] and case reports.[5] The key is that better inputs lead to better outputs.[5] For instance, some researchers have tried to generate abstracts by inputting only
the paper title.[6] However, neither humans nor ChatGPT can produce a “good” abstract from a title alone.
Crafting a detailed and effective input for manuscript writing requires human effort
and skilled experience. Completing this task typically leads to manuscript completion,
so this use of ChatGPT can be seen as “human first, ChatGPT as a helper”. Skilled
writers who can create complete prompts may prefer to write themselves, finding it
simpler and more effective. In this context, ChatGPT's role is similar to the transition
from handwriting to word processing we experienced 30 years ago. Given these points,
I fully agree with Cavalcante-Silva et al.'s statement.
The second point is the most crucial: it concerns the future impact of ChatGPT on
human writing, thinking, and cognitive abilities. Historically, the introduction of
new technologies has often prompted similar concerns. For example, automobiles were
thought to reduce human physical activity, and personal computers (PCs), by replacing
handwriting, were expected to diminish our spelling memory. Despite these worries,
both automobiles and PCs have become indispensable without causing notable inconvenience.
The question now is whether AI and ChatGPT will have a similar degree of impact. This
issue may be considered individually, but I believe that ordinary physicians may not
reach a definitive answer. Therefore, I hope that brain scientists, including those
in sleep science, will provide objective data—such as differences in brain activity
between self-writing and ChatGPT-assisted writing. For example, as cited by Cavalcante-Silva
et al.,[1] nighttime exposure to short-wavelength blue light from smartphone screens has been
shown to alter circadian rhythms,[7] potentially affecting long-term mental conditions. Similarly, ChatGPT-assisted writing
might impact mental or cognitive conditions over time. Accumulating such data will
help us scientifically predict the effects of ChatGPT on human cognition.
The third point concerns the regulation of ChatGPT use in academic writing. This issue
is closely tied to the answer to the second point. If the “safety” of ChatGPT use
is scientifically validated, the current “soft” regulation requiring the declaration
of ChatGPT use may remain reasonable. However, if any harmful effects are suggested,
let alone proven, stricter regulations should be considered. We must lean towards
caution because the potential impact of ChatGPT on human writing and thinking abilities
is groundbreaking.
The latest smartphones are equipped with generative AI, indicating that this technology
has already permeated everyone's daily life. Some authors may be tempted to rely heavily
on ChatGPT for paper writing. However, let us consider a hypothetical statement: “The
use of ChatGPT in academic writing is believed to deteriorate the writing and thinking
abilities of future generations and should thus be avoided except as a linguistic
checker”. This would serve as a warning similar to “global warming”. I believe that
doctors, concerned with not only current but also future health and welfare, will
adhere to such regulations once they are established. A detector of ChatGPT-generated
manuscript and the ethical matter of ChatGPT-writing will naturally affect such regulations:
I herein did not touch on these topics to simplify the story.
If such a statement is proven incorrect and ChatGPT use in writing is deemed safe,
then the aforementioned regulation should be abolished, and new regulations should
be established. We should always err on the side of caution. One might ask, “Why did
not earlier generations regulate its use more strictly?”. We could end up facing criticism
from future generations, which might be a “nightmare” for us. I hope that sleep science
researchers will provide new insights into this issue.
Bibliographical Record
Shigeki Matsubara. Beyond the Monotonous Discussion of ChatGPT Use in Academic Writing:
Expectations for Sleep Science Researchers. Sleep Sci ; : s00451805062.
DOI: 10.1055/s-0045-1805062
