ChatGPT: The transformative influence of generative AI on science and healthcare

Scientific News

Varghese and Chapiro rechently published in Journal of Hepathology an interesting article detailing how Large Language Models have the potential to transform scientific communication but need to be handle with the utmost care to guarantee accuracy and quality of the scientific informations described. Semisupervised generative artificial intelligence has shown promise in creating humanlike text and may have various applications in science and clinical practice. Large language models (LLMs) like ChatGPT are being used for tasks such as text generation and patient education. However, there are concerns about bias, accuracy, and transparency in the use of LLMs. Editorial policies are being updated to address the use of LLMs in research publications. Overall, LLMs have the potential to improve communication in healthcare but require careful oversight and validation.

Varghese, J., & Chapiro, J. (2024). ChatGPT: The transformative influence of generative AI on science and healthcare. Journal of Hepatology, 80(6), 977–980

In an era of rapid technological evolution, the liver community has a unique opportunity to embrace breakthroughs in generative AI with an open mind . These advancements have the potential to transform the landscape of liver disease research and practice, sparking optimism and intrigue among researchers and healthcare professionals.

The Journal of Hepatology , recognizing the significance of novel developments in artificial intelligence (AI), has taken proactive steps to stay at the forefront of this evolving field. By recruiting experts in the AI field to serve on the Editorial Board, the journal is demonstrating its commitment to fostering innovation and driving progress in liver disease research.

Publications introducing novel AI technology are no longer uncommon in the journal. They are among the most highly debated and possibly practice-changing papers across a broad range of scientific disciplines, united by their focus on liver disease.

As AI is rapidly evolving, this expert paper will focus on educating the readership on large language models and their possible impact on the research practice and clinical outlook in liver disease. We will outline both the challenges, such as the need for large amounts of high-quality training data, and the opportunities, such as the potential for more accurate and personalized diagnoses, that LLMs present in the field.

What is generative AI?

Generative artificial intelligence (AI) creates new content, such as text, images, or music, based on patterns in the data it has been trained on.

McCarthy coined the term AI in the 1950s and describes a system that can mimic human behavior.

This can be realized via expert-driven rule-based systems or by data-driven training via machine learning.

A significant recent advance has been semisupervised generative AI, which is trained on unlabelled data and fine-tuned for specific supervised tasks.

It can create more complex output based on input prompts and – in its most advanced version – may generate entirely new data contexts.

ChatGPT utilizes transformer neural networks that have been pre-trained on unlabelled large text corpora. This process involves exposing the model to a vast amount of text data without any specific labels or annotations, allowing it to learn the underlying patterns and structures of the language.

At this stage, ‘large language models’ or LLMs, refer to transformer neural networks that have been pre-trained on unlabelled large text corpora to acquire a comprehensive understanding of language patterns.

New realistic images can be generated based on user prompts, as showcased by Dall-E, another creation of OpenAI.

Generative AI Generating flexible output for new content

With resources and the immense amount of available human-written text that serves as training data, ChatGPT generates human-like text better than most other natural language processing tools.

In addition to its potential uses for text generation, some early applications have been systematically tested for specific tasks.

These include supporting computer programmers in advanced tasks such as converting or generating programming source code, proofreading code, and bug-fixing, classifying hate speeches on Twitter, or passing the USMLE medical exam.

One key advantage that LLMs could have compared to classical supervised deep learning is their chatbox functionality, which can communicate with user questions more intuitively than a complex graphical user interface that requires structured input from the user.

This type of free-text communication has the potential to improve human-machine interaction, e.g., by increasing the technology’s accessibility for patients, particularly visually impaired or blind people, and healthcare professionals or researchers.

While this measure may be adequate to prevent protected health information from being widely disseminated, there is currently no safeguard to monitor the quality of the learning experience effectively and direct the future ChatGPT output toward a verifiable and accurate level of medical information.

Implications for scientific activities

While ChatGPT can accelerate the writing process, one must be concerned about output and bias amplification.

An illustrative example of artificial hallucination in academic writing occurs when ChatGPT is asked to provide literature sources for scientific statements or one of its generated answers.

If a significant amount of an author’s text is generated via an LLM, this should be mentioned as a further note or acknowledgment of the exact LLM version.

It will depend on the editorial policies of each publisher to specify what a significant amount means.

While the Science Journal and the International Conference for Machine Learning have recently published a ban on submissions using ChatGPT or other LLMs, many other Journals are currently considering updates to their editorial policies. Yang H. (2023) also published in the journal Nature a guide on how he uses Generative AI responsibly in his teaching.

While it should be emphasized that LLMs may never be used to replace a seasoned peer-reviewer professional judgment call concerning the level of credibility and quality of the submitted original research, other highly time-consuming tasks that bear a lower level of risk may very well be “outsourced” to LLMs if an appropriate level of supervision can be guaranteed. This could free up healthcare professionals’ time for more complex tasks requiring expertise and judgment.

Practical opportunities

LLMs have enormous potential for improving communication in healthcare in various ways.

If appropriately trained and validated, such models may excel at patient education due to their unparalleled ability to provide varying degrees of medical information to patients interactively and iteratively. This could lead to improved patient understanding of their condition and treatment options, and ultimately, better patient outcomes.

This feature may significantly improve access to care and allow for improved resource utilization in interactions between patients and healthcare professionals.

LLMs may quickly become the bridge, interpreting complex or lengthy sub-specialty reports, e.g., from pathology or radiology for patients and general practitioners, easing the linguistic barriers and providing language at the level requested by the end-user

Regulatory challenges

Generative AI and LLMs, in particular, are no exception to the widely debated risks of applying AI in healthcare. These risks include potential biases in the data used for training, lack of transparency in the decision-making process, and the need for regulatory approval for medical purposes.

Even if the software version is implemented within the internal hospital network, such software will require regulatory approval for medical purposes. This process can be complex and time-consuming, and may pose a significant barrier to the adoption of LLMs in healthcare.

Each software application that aims to support clinical decision-making and affects the diagnosis, treatment, or prevention of disease fulfills the status of a medical device.

AI software to improve clinical decision-making will require quality and risk management systems. Such software will need to demonstrate medical benefits while preserving patient safety. Healthcare professionals can be held accountable when using such unapproved systems, as is the case when administering drugs without approval.

The use of AI and Large Language models also imply political regulation which are underway such as the proposal for regulation being discussed at the European Parliament

Conclusions and outlook

ChatGPT has tremendous potential to improve the way authors generate or edit text and increase information accessibility.

There will be a variety of novel LLMs or generative AI applications, all of which will be further tailored to user-specific needs and will most likely outperform current versions in specialized tasks.

This technology is still in its infancy and requires critical human oversight due to amplification bias, artificial hallucination, and lack of model transparency or explainability. It’s important to note that LLMs are not a panacea and have their limitations, such as the potential for over-reliance on the technology and the need for continuous human validation of the generated outputs.

Was this article helpful?