Keywords
generative artificial intelligence - ChatGPT - education - institutional guide
The emergence of artificial intelligence (AI) has revolutionized many fields, including
natural language processing. Among these AI technologies, ChatGPT, a large language
model trained by OpenAI based on the GPT-3.5 architecture, has garnered attention,
especially in relation to its potential to assist in writing text in general. The
GPT in the acronym stands for generative pre-trained transformer, which is a type
of neural network-based model that generates natural language text. Generative pre-trained
transformer-3 was released in 2020 and became public in the same year, while ChatGPT
was made available to the public in 2022 and is constantly updated to improve its
capabilities. The last two versions of ChatGPT have significantly impacted generative
writing by expanding the possibilities of using the tool as an aid in writing scientific
papers. While AI can offer benefits, such as streamlining the writing process, ensuring
consistency, and saving time, there are concerns regarding the accuracy and reliability
of information generated by AI. This letter aims to provide a scientific-based overview
of ChatGPT and to highlight the challenges and opportunities associated with AI-assisted
scientific writing and AI-powered applications in medicine.
Recent research (published on a preprint server) comparing original scientific abstracts
with those generated by ChatGPT reported that the latter produced credible scientific
abstracts with no plagiarism detected.[1] The scientific texts were evaluated, and 66% of the AI-generated abstracts were
identified by an AI output detector and 68% by human reviewers. However, the human
reviewers incorrectly identified 14% of the original abstracts as being generated
by an AI, while the AI output detector only mistakenly identified one (2%) original
article as being generated.[1] Despite the difficulty in differentiating the articles in the two formats, the reviewers
pointed out that generated abstracts were vaguer and presented stereotyped wording
(which may be related to the clarity of the command/richness of information given
to the AI). The authors highlighted the possible unethical use of the tool to falsify
research (as it generates credible numbers) and stressed the importance of clear disclosure
when the technology is used to support the researcher's scientific knowledge.[1]
In academic work, the ability to concatenate information, present results, and discuss
ideas in scientific texts is a valued and welcomed skill, and the submission of the
written work (course completion paper, monograph or thesis) is part of the requirements
for obtaining a degree in most undergraduate and graduate programs. As generative
AI becomes more prevalent, we may be on the cusp of a paradigm shift in how we evaluate
knowledge. With this in mind, we invite teachers and researchers to consider the implications
of the use of this tool in the current scientific landscape.
This debate about AI takes place against a background of a general debate about the
impact of a range of technologies. For example, a 2019 study by Zinnatullina et al.,
which looked at the relationship between cinema and literature in the 21st century, discussed the positive and negative impacts of cinema on literature.[2] In her book Reader, Come Home: The Reading Brain in a Digital World, the American cognitive neuroscientist Maryanne Wolf discussed the impact of digital
technology (cell phones and tablets) on the ability of humans to read and comprehend
information, understand complex arguments, and critically analyze different points
of view addressed in texts.[3] Although this author argued that the nature of digital media does not automatically
condemn the practice of deep reading, and can even enhance it, the digital universe
does present a threat that can erode this form of attention and that we need to teach
our brains to become all-encompassing in the age of electronic technology.
It should not be forgotten that, over recent decades, the way that individuals gather
information and connect with each other has been shaped by internet access and an
increasingly technology-integrated daily life.[4] In this scenario, the focus should not be on the technological advances, but rather
on the potential consequences of deviating from traditional patterns in areas such
as natural intelligence and interpersonal interactions. Regarding the use of AI generative
technology, we believe it should not be evaluated in a purely dichotomous way, either
blaming it for dumbing down the human species or praising it for accelerating scientific
development—through, for example, facilitating the writing of manuscripts. Rather,
the main consequences related to the use of this technology need to be properly evaluated,
and institutional principles must be created to guide when generative AI should and
should not be used. These might include factors, such as the criteria for authorship,
the appropriate use of AI content, and ensuring accuracy, completeness, and originality
of the content produced.
Although it is possible for an author to write a text on a subject that they do not
know, or produce a manuscript from the input of data using generative AI, it should
be noted that the type of request or command given to the AI directly impacts the
quality of the text generated; asking the right question and using proper reasoning
are central points to consider when trying to solve a problem - as Albert Einstein
said, “If I had an hour to solve a problem, I would spend 55 minutes thinking about
the problem and 5 minutes thinking about solutions”. In this context, could the ability
to ask the right question/give the correct command be the human skill that will be
most valued in the AI era?
As sleep researchers, we are particularly interested in the possible impact of the
use of generative AI on sleep; however, to the best of our knowledge, no studies have
yet been published on this topic. Over recent decades, there has been marked growing
awareness and evidence concerning the importance of sleep; it has been shown that
cognitive performance,[5] memory retention,[6] and productivity (a desired aspect for AI users) depend on sleep in sufficient quantity
and quality. Moreover, sleep has a widespread impact on homeostasis and health. Yet,
despite this knowledge, insufficient sleep has become commonplace, and sleep deprivation
presents a global health problem.[7]
[8] It is already well documented that increased access to technologies can result in
sleep curtailment.[9]
[10]
[11] One example of the negative effects of technology on sleep is nighttime exposure
to the short wavelength blue light produced by smartphone screens, which causes a
dose-dependent suppression of melatonin production and is associated with altered
circadian rhythms.[12]
[13] Regarding the use of generative AI-powered applications, they might prove to be
so engaging and addictive that they could become an additional factor contributing
to sleep disruption. However, technology can also positively affect sleep, since the
development of improved wearable devices that use AI to assess sleep can provide helpful
feedback and help those who sleep poorly to improve their sleep quality.[14] One of the most common and health-threatening sleep problems, obstructive sleep
apnea (OSA), with a prevalence of one third of the population,[15] can now be detected and quantitatively measured using a wearable watch device that
uses photoplethysmography to detect pulse intervals and an automated algorithm to
detect heart rate rhythms characteristic of OSA.[16] In the future, wearable devices using AI may be able to assess parameters that can
currently only be evaluated by polysomnography, a gold-standard test that requires
specific equipment and qualified technical support, thus aiding in the assessment
of sleep quality and the diagnosis of sleep disorders.
The use of AI will yield both positive and negative outcomes, but its potential applications
are clear. For example, it could be a valuable tool in scientific writing, especially
in respect of meta-analysis studies, systematic reviews and other complex evaluations.
It could also be of great benefit to educators by performing routine, time-consuming
tasks and thus freeing up their time to be spent in more productive ways. However,
in relation to both education and the writing of scientific texts, clear rules and
objectives must be established for the use of AI. This technology also has potential
practical applications in medicine in areas such as drug discovery[17] and clinical decision making.[18] Moreover, the use of AI could give professionals more time to focus on basic research
questions, thus accelerating the speed of scientific progress.
Regarding sleep, AI can also have both positive and negative effects; AI-powered technologies
have immense potential to advance our understanding of sleep patterns and disorders;
however, AI may also have the capability to disrupt sleep routines due to its captivating
and potentially addictive attributes. Striking a balance between the benefits and
potential drawbacks of integrating AI into society demands ongoing research by experts,
the wide dissemination of the scientific results, as well as continued public discourse
on the subject.
