Review

Generative Pre-Trained Transformer-Empowered Healthcare

Conversations: Current Trends, Challenges, and Future
Directions in Large Language Model-Enabled Medical Chatbots
James C. L. Chow 1,2, * , Valerie Wong 3 and Kay Li 4

1 Department of Medical Physics, Princess Margaret Cancer Centre, University Health Network,
Toronto, ON M5G 1X6, Canada
2 Department of Radiation Oncology, University of Toronto, Toronto, ON M5T 1P5, Canada
3 Department of Physics, Toronto Metropolitan University, Toronto, ON M5B 2K3, Canada;
[email protected]
4 Department of English, University of Toronto, Toronto, ON M5R 2M8, Canada
* Correspondence: [email protected]

Abstract: This review explores the transformative integration of artificial intelligence (AI) and
healthcare through conversational AI leveraging Natural Language Processing (NLP). Focusing
on Large Language Models (LLMs), this paper navigates through various sections, commencing
with an overview of AI’s significance in healthcare and the role of conversational AI. It delves into
fundamental NLP techniques, emphasizing their facilitation of seamless healthcare conversations.
Examining the evolution of LLMs within NLP frameworks, the paper discusses key models used
in healthcare, exploring their advantages and implementation challenges. Practical applications
in healthcare conversations, from patient-centric utilities like diagnosis and treatment suggestions
to healthcare provider support systems, are detailed. Ethical and legal considerations, including
patient privacy, ethical implications, and regulatory compliance, are addressed. The review concludes
by spotlighting current challenges, envisaging future trends, and highlighting the transformative
Citation: Chow, J.C.L.; Wong, V.; Li, K. potential of LLMs and NLP in reshaping healthcare interactions.
Generative Pre-Trained
Transformer-Empowered Healthcare Keywords: AI; humanistic AI; ethical AI; machine learning; large language models; natural language
Conversations: Current Trends, processing; medical chatbot; transformer-based model; ChatGPT; healthcare
Challenges, and Future Directions in
Large Language Model-Enabled
Medical Chatbots. BioMedInformatics
2024, 4, 837–852. https://doi.org/ 1. Introduction
10.3390/biomedinformatics4010047
The trajectory of artificial intelligence (AI) development spans decades, with machine
Academic Editors: Carson K. Leung learning (ML) emerging as a pivotal force in propelling AI’s evolution [1–4]. The adoption
and Alexandre G. De Brevern of AI and ML in the medical field has experienced significant growth, particularly in
Received: 29 January 2024
ML-enabled medical devices. Joshi et al. focused on 691 FDA-approved AI/ML-enabled
Revised: 4 March 2024
medical devices, revealing a substantial surge in approvals since 2018, predominantly
Accepted: 12 March 2024
in radiology. The prevalence of the 510(k) clearance pathway, relying on substantial
Published: 14 March 2024 equivalence, is notable [5]. This review focuses on a specific ML facet: the Large Language
Model (LLM) within Natural Language Processing (NLP) [6,7]. Particularly, we delve into
the integration of LLMs like Chat Generative Pre-trained Transformer (ChatGPT, version
3–4) into chatbots, augmenting their capacity for seamless user engagement [8–10].
Copyright: © 2024 by the authors. Chatbots, AI-driven conversational agents prevalent in online interactions, have found
Licensee MDPI, Basel, Switzerland. extensive utility in disseminating healthcare information and enhancing customer ser-
This article is an open access article
vices [11–15]. Table 1 summarizes the general features that medical professionals would
distributed under the terms and
expect a medial chatbot to have. These features encompass accurate information retrieval,
conditions of the Creative Commons
symptom assessment, and diagnosis support to help in understanding and addressing
Attribution (CC BY) license (https://
health concerns. Moreover, the chatbot is expected to provide treatment guidance, medica-
creativecommons.org/licenses/by/
tion information, and assistance with appointment scheduling, ensuring a comprehensive
4.0/).

BioMedInformatics 2024, 4, 837–852. https://doi.org/10.3390/biomedinformatics4010047 https://www.mdpi.com/journal/biomedinformatics

BioMedInformatics 2024, 4 838

healthcare experience. Health monitoring features, emergency response capabilities, and

patient education contribute to a holistic approach. Privacy and security measures, multi-
lingual support, and integration with electronic health records uphold standards of confi-
dentiality and accessibility. Personalized recommendations, follow-up mechanisms, and
a user-friendly interface tailor the chatbot experience to individual needs, while features
like adherence support and mental health resources further enhance its utility. Continuous
feedback mechanisms ensure ongoing improvement, making the chatbot a valuable tool to
promote patient well-being. The advent of ChatGPT has notably elevated the appeal of
chatbots, facilitating more human-like interactions through adaptive text learning [16,17].
However, the precision of healthcare information dispensed by ChatGPT still raises some
concerns, prompting inquiries into potential user misguidance [18–20].

Table 1. Features of AI chatbot expected by the medical professional.

Feature Description
Accurate Information Retrieval Provide accurate and up-to-date medical information from reliable sources.
Symptom Assessment Analyze and assess user-described symptoms to suggest potential health conditions.
Offer preliminary assistance in suggesting potential diagnoses, understanding its
Diagnosis Support
limitations.
Provide general information on treatments, medications, and lifestyle
Treatment Guidance
recommendations.
Offer details about medications, including dosage, side effects, and potential
Medication Information
interactions.
Assist users in scheduling appointments with healthcare providers and send
Appointment Scheduling
reminders.
Support users in tracking and monitoring health metrics like blood pressure or blood
Health Monitoring
sugar.
Recognize urgent situations and provide emergency response information or facilitate
Emergency Response
contacts.
Offer educational content to enhance users’ understanding of medical conditions and
Patient Education
prevention.
Ensure strict adherence to data privacy regulations and maintain the confidentiality of
Privacy and Security
user health information.
Multilingual Support Provide communication in multiple languages to cater to diverse patient populations.
Integration with EHR Facilitate integration with existing healthcare systems to access relevant patient data.
Personalized Recommendations Offer personalized health advice based on user data, preferences, and lifestyle.
Implement features for follow-up interactions, reminders, and maintaining continuity
Follow-up and Continuity of Care
of care.
User-Friendly Interface Ensure an intuitive and user-friendly interface for easy interaction.
Adherence Support Assist patients in adhering to prescribed treatment plans and medications.
Include features for mental health assessments, stress management, and access to
Mental Health Support
mental health resources.
Feedback and Improvement Incorporate mechanisms for users to provide feedback on the chatbot’s performance.

This review underscores the dual role of AI-assisted healthcare chatbots, exploring
their potential to educate the public with accurate information sourced from medical
institutions [21], while acknowledging the risks associated with misinformation [22]. Fur-
thermore, the exploration extends to uncharted territory, considering the role of chatbots
in aiding disabled individuals and the elderly. Additionally, the paper contemplates the
nuanced function of chatbots as temporary emotional outlets, particularly relevant given
the upsurge in depression cases during the recent pandemic [23,24].
 in aiding disabled individuals and the elderly. Additionally, the paper contemplate
nuanced function of chatbots as temporary emotional outlets, particularly relevant g
the upsurge in depression cases during the recent pandemic [23,24].
BioMedInformatics 2024, 4 In this comprehensive examination, we scrutinize the advantages that 839 AI chat
bring to the healthcare system while addressing inherent challenges. The discussion
compasses the intricate dynamics of AI chatbots, their potential to positively im
In this comprehensive
healthcare examination, we and
information dissemination, scrutinize
their the advantages
pitfalls. As wethat AI chatbots
delve into the future
bring to the healthcare system while addressing inherent challenges. The discussion
jectory, the review aims to illuminate potential advancements in AI chatbots within
encompasses the intricate dynamics of AI chatbots, their potential to positively impact
healthcareinformation
healthcare sector. dissemination, and their pitfalls. As we delve into the future
trajectory, the review aims to illuminate potential advancements in AI chatbots within the
2. Fundaments
healthcare sector. and Evolution of Language Models
2.1. Fundamentals of Natural Language Processing
2. Fundaments and Evolution of Language Models
NLP stands
2.1. Fundamentals as a cornerstone
of Natural in the realm of ML, a subset of AI that learns from
Language Processing
to approximate
NLP stands ashuman expectations
a cornerstone [25,26].
in the realm of ML,Particularly,
a subset of AI NLP playsfrom
that learns a pivotal
data to role in f
itating AI’s comprehension of the diverse languages used by individuals. Chatbots
approximate human expectations [25,26]. Particularly, NLP plays a pivotal role in facilitat-
ing AI’s comprehension
grated of the diverse
with NLP capabilities languages
excel used byand
in learning individuals. Chatbots the
understanding integrated
natural lang
with NLP capabilities excel in learning and understanding the natural language
patterns employed by users in textual communication, enabling them to respond in patterns
employed by users in textual communication, enabling them to respond intelligibly [27,28].
gibly [27,28]. Figure 1 shows a typical chatbot architecture, including the user inter
Figure 1 shows a typical chatbot architecture, including the user interface, user message
user message
analysis analysis
component, dialogcomponent,
management dialog management
component, responsescomponent, responses genera
generation component,
component,
and and[28].
the database theThe
database
NLP is[28].
mainlyThe NLPto
linked is the
mainly linked
message to the
analysis message to
component analysis c
ponent to analyze the context information.
analyze the context information.

Figure1.1.General
Figure General chatbot
chatbot architecture.
architecture. Source:Source:
AdaptedAdapted
from [28].from [28].

Given the inherent variability in how individuals communicate, lacking a standard-

Given the inherent variability in how individuals communicate, lacking a stand
ized template or exact pattern, ML, especially NLP, strives to analyze free-text and speech
ized template
through linguistic orand
exact pattern,
statistical ML, especially
algorithms. NLP,aims
This analysis strives to analyze
to extract free-text
discernible pat- and sp
through
terns from linguistic and statistical
the rich tapestry algorithms.
of human expression ThisWhile
[29,30]. analysis aims
pattern to extract
analysis formsdiscernible
the
terns fromthe
foundation, theevolution
rich tapestry of human expression
of AI necessitates its ability to [29,30].
engage inWhile pattern
meaningful analysis form
conversa-
tions with users,
foundation, theprimarily
evolutionexemplified in question-answering
of AI necessitates its ability (QA) scenarios
to engage in [31,32].
meaningful conv
The acquired text patterns are cataloged in a database, empowering
tions with users, primarily exemplified in question-answering (QA) scenarios the AI to match [31,32]
these learned patterns during user interactions—a process akin to pattern matching and
text searching techniques [33]. Crucially, NLP goes beyond mere pattern recognition; it
grapples with the nuances of how individuals articulate ideas. This involves understanding
BioMedInformatics 2024, 4 840

that distinct expressions can convey the same meaning, enabling AI to emulate human-like
responses, thus enhancing the conversational experience [34,35].
It can be seen that the training of the model involves exposing the algorithm to vast
amounts of text data, allowing it to learn the patterns, semantics, and structures inherent in
human language. This process typically utilizes large datasets to train the model on tasks
such as language understanding, sentiment analysis, or question answering. The NLP
model undergoes iterative adjustments during training, refining its ability to recognize
and generate meaningful language output. The ultimate goal is to enhance the model’s
proficiency in understanding and generating human-like text, enabling it to perform diverse
linguistic tasks with accuracy and relevance.
In the healthcare domain, NLP demonstrates its prowess by extracting pertinent in-
formation from free-text documents such as electronic health records. Beyond symptom
examination, NLP’s ability to compare, classify, and recommend actions based on vast
sets of textual data contributes significantly to disease symptom classification and patient
guidance [36,37]. NLP emerges as a linchpin in the intersection of AI and healthcare, foster-
ing a nuanced understanding of language patterns, enhancing conversational dynamics,
and contributing invaluable insights in the medical field [38]. The following exploration
further delves into the applications and implications of NLP in healthcare conversations.

2.2. Evolution of Large Language Models

LLMs have emerged as transformative components within the NLP, significantly
influencing the evolution of AI [39,40]. LLMs, belonging to the broader category of ma-
chine learning, excel in processing and generating human-like text by leveraging extensive
datasets. Their remarkable ability to capture intricate language nuances and generate
coherent responses has positioned them as integral players in advancing NLP [41]. In
healthcare conversations, LLMs play a crucial role in enhancing the conversational capabil-
ities of AI systems [42]. By understanding and generating contextually relevant responses,
LLMs contribute to the humanization of interactions, creating more engaging and effective
healthcare dialogues.
Several key LLMs have made a significant impact on the healthcare conversations,
revolutionizing the way AI engages with users. One noteworthy exemplar is OpenAI’s
GPT (Generative Pre-trained Transformer) series, with models like GPT-3 and 4 demon-
strating exceptional language understanding and generation capabilities [43,44]. GPT-4,
in particular, has garnered attention for its versatility in various applications, including
healthcare-related tasks [45]. BERT (Bidirectional Encoder Representations from Transform-
ers) is another influential LLM that has left an indelible mark on NLP [46,47]. Renowned
for its bidirectional training approach, BERT excels in grasping contextual nuances, making
it particularly adept at understanding the intricacies of medical language and information.
Furthermore, models like XLNet [48], T5 (Text-to-Text Transfer Transformer) [49], and BART
(Bidirectional and Auto-Regressive Transformers) [50] have played instrumental roles in
advancing the sophistication of LLMs in healthcare applications. These models exhibit
enhanced capabilities in processing medical literature, extracting relevant information, and
generating coherent responses tailored to healthcare-related inquiries [51]. Table 2 shows
some popular LLMs used in healthcare conversations.
The utilization of LLMs in healthcare conversations signifies a paradigm shift, enabling
AI systems to comprehend and respond to user queries with a depth of understanding
akin to human-like interactions. As we navigate through the evolutionary trajectory
of LLMs, their continued refinement and integration into healthcare dialogue systems
hold promise for further augmenting the efficacy and user experience in the realm of
medical conversations.
 Model Description Applications in Healthcare
OpenAI’s powerful LLM with
Medical documentation, question answering, text-based
GPT-3 and 4 [43–45] strong natural language
interactions
BioMedInformatics 2024, 4 understanding. 841
Clinical text analysis, medical literature understanding,
Bidirectional processing makes
BERT, BioBERT and biomedical text mining, information extraction from
BERT suitable for clinical text
ClinicalBERT [46,47] Table 2. Some LLMs and their medical
potential texts, clinical
applications note understanding,
in healthcare conversations. medical
analysis.
question answering
Model Description Applications in Healthcare
OpenAI’s model capable of
XLNet [48] Medical literature analysis, clinical documentation
GPT-3 and 4 [43–45] capturing OpenAI’s powerful
bidirectional LLM with strong
context. Medical documentation, question answering,
natural language understanding. text-based interactions
Text-to-Text Transfer Transformer,
T5 [49] Summarization of medical
Clinical documents,
text analysis, medicalquestion
literature generation
designed for various NLP tasks.
BERT, BioBERT and Bidirectional processing makes BERT understanding, biomedical text mining,
ClinicalBERT [46,47] Bidirectional and for
suitable Auto-Regressive
clinical text analysis. information extraction from medical texts, clinical
BART [50] Transformer, used for text Text summarization, document medical
note understanding, generation, paraphrasing
question answering
generation.OpenAI’s model capable of capturing
XLNet [48] Medical literature analysis, clinical documentation
bidirectional context.
The utilization of LLMs in healthcare Summarization
Text-to-Text Transfer Transformer,
conversationsofsignifies a paradigm shift, ena-
medical documents, question
T5 [49] bling AI systems to comprehend
designed for various NLP tasks. and respond to user queries with a depth of understand-
generation
ing akin to human-like interactions. As we navigate through the evolutionary trajectory
Bidirectional and Auto-Regressive Text summarization, document generation,
BART [50] of LLMs, their used
continued refinement and paraphrasing
integration into healthcare dialogue systems
Transformer, for text generation.
hold promise for further augmenting the efficacy and user experience in the realm of med-
ical conversations.
The incorporation
The incorporation of of LLMs
LLMs into
into medical
medical chatbots
chatbots introduces
introduces significant
significant advantages,
advantages,
revolutionizing the healthcare interactions. LLMs enhance the understanding contextual
revolutionizing the healthcare interactions. LLMs enhance the understanding of of contex-
nuances
tual in medical
nuances queries,
in medical enabling
queries, enablingchatbots
chatbots totoprovide
providemore
moreaccurate
accurate and
and relevant
relevant
responses [52].
responses This heightened
[52]. This heightened comprehension
comprehension fosters
fosters aa humanized
humanized interaction, with
interaction, with
LLMs proficiently mimicking natural language patterns, creating a
LLMs proficiently mimicking natural language patterns, creating a more engaging andmore engaging and
empathetic user experience. Additionally, LLMs empower medical chatbots
empathetic user experience. Additionally, LLMs empower medical chatbots to efficiently to efficiently
retrieve and
retrieve and disseminate
disseminate precise
precise medical
medical information,
information, positioning
positioning them
them as
as reliable
reliable sources
sources
of up-to-date healthcare knowledge [53]. Figure 2 shows the touchpoints of
of up-to-date healthcare knowledge [53]. Figure 2 shows the touchpoints of a patient’s a patient’s care
journey in which an LLM can be employed to enhance the patient’s
care journey in which an LLM can be employed to enhance the patient’s experience.experience.

Figure
Figure 2.2. A A
standard patient
standard journey
patient in healthcare,
journey encompassing
in healthcare, encompassingthree three
key stages: (1) pre-consul-
key stages: (1) pre-
tation involves
consultation patient patient
involves registration, medicalmedical
registration, consultation, or healthorscreening;
consultation, (2) diagnosis
health screening; includes
(2) diagnosis
patient consultations, examinations, and supplementary investigations; and (3) management com-
includes patient consultations, examinations, and supplementary investigations; and (3) management
prises medication, patient counseling, education, and reimbursement for medical bills. LLMs exhibit
comprises medication, patient counseling, education, and reimbursement for medical bills. LLMs ex-
potential to improve the patient experience at each touchpoint in this journey. Source: Adapted from
hibit potential to improve the patient experience at each touchpoint in this journey. Source: Adapted
[53].
from [53].
However, the implementation of LLMs in medical chatbots is not without challenges.
However, the implementation of LLMs in medical chatbots is not without challenges.
Ensuring the accuracy and trustworthiness of information is paramount, as LLMs may
Ensuring the accuracy and trustworthiness of information is paramount, as LLMs may in-
inadvertently generate inaccurate responses, posing a risk of misinformation [54,55]. Pri-
advertently generate inaccurate responses, posing a risk of misinformation [54,55]. Privacy
vacy and security concerns arise, demanding robust measures to safeguard sensitive
and security concerns arise, demanding robust measures to safeguard sensitive health-
health-related data [56]. Furthermore, interpreting complex medical terminology and
related data [56]. Furthermore, interpreting complex medical terminology and aligning
with user expectations present ongoing challenges. Addressing these hurdles is essential to
fully harness the potential benefits of LLMs in the dynamic realm of medical chatbots [18].
BioMedInformatics 2024, 4 842

3. Results
3.1. Application of AI Chatbot in Healthcare
3.1.1. Healthcare Knowledge Transfer with Chatbot
As mentioned, the general public are more likely to use chatbots to look for answers
to their questions in daily life or even in medicine. So it is important to note that chatbots
might obtain wrong information and mislead users, unless developing a healthcare-based
chatbot that is designed to answer people with accurate medical information [57,58]. People
would not know if they were misled and believe the provided information blindly, which
could lead to accidents. It would be best to develop medical chatbots with professionals
like doctors and nurses to ensure the information is accurate and easy to understand for the
general public. If LLMs could be trained specifically in healthcare then their trustworthiness
would be increased, but this might require professionals to verify that every single piece of
information provided by the LLM is correct [59]. LLMs’ trustworthiness is controversial;
some researchers think they simply gather information from the internet and provide it to
the users, while other researchers think they have the potential to be trained specifically
for healthcare purposes with related journals [60]. On the other hand, using chatbots for
healthcare knowledge transfer can help prevent clinicians from answering similar questions
from various patients repeatedly, and hence could allow clinicians to work on jobs that are
highly prioritized [61]. As such, there remains a concern of accuracy and reliability of the
information that researchers always keep in mind [18].

3.1.2. Symptom Diagnosis

According to the study of Kumar et al. [62], training an AI model with a number
of papers in the healthcare field means that the model is able to analyze and predict the
symptoms of diseases. Aside from simply classifying symptoms, clinicians are responsible
for compiling electronic health records (EHRs), which is a patient information managing
digital system, for every single patient visiting a hospital [63]. Having an LLM trained
for classifying symptoms according to doctor–patient conversations would increase the
efficiency of seeing each patient and thus reduce the chance of overcrowding during busy
hours [63]. In fact, there is already a ChatGPT-like chatbot created for healthcare, Med-
PaLM, which is able to analyze X-ray images according to the examples given in [64].
Moreover, it is possible for the LLM to integrate with telehealth services especially for
those who have a disability [53]. In addition, it is believed that LLMs are even able to
detect later-life depression, which is a kind of major public health concern that occurs in
the older generation, as the name states [65]. NLP analyzes the way people talk and also
the speed and pitch they use in order to understand their speech patterns [65]. Having
understood their speech patterns, it is possible to discover the speech patterns of people
with depression and use these as a kind of template to analyze and compare the template
patterns with the target user in order to determine whether they are possibly experiencing
later-life depression.

3.1.3. Mental Healthcare

Having a conversation with someone is the easiest way to help balance mental health,
as people are able to express their feelings when they talk to someone. It is not difficult to
imagine that people can have conversations with LLM chatbots like ChatGPT especially
when there is no one to talk with [66,67]. Stress could simply keep building up when
something bad happens in a person’s daily life and they cannot talk to someone, and then
the stress reaches a limit and their performance at work may worsen and their health
might also be affected to some extent. Chatbots can help with daily emotional support;
there are studies proving they are capable of helping people get rid of stress and feelings
of depression, which also somehow demonstrate better results than traditional mental
heath treatments [68]. With NLP analyzing the texts posted on different social media
platforms, the AI model is able to perform detection of emotions and monitoring of mental
health [69]. It is a sort of text classification scenario in that the NLP technique allows the
BioMedInformatics 2024, 4 843

AI model to analyze texts and compare them with other similar texts and classify them
into different cases in order to detect mental illness from those [70]. In social media, AI
models most commonly focus on detecting scenarios like suicide; when they detect any
wording that might possibly relate to suicide, it recommends users to contact mentors for
mental support [71]. Other than that, there are many people living alone and some of them
might not have anyone around them with whom they can share their feelings of daily life,
for example, they cannot express the negativity they feel at school or in work. People get
depressed when negativity and stress keep building up without letting it out by having
someone to talk to; so, it would be nice if chatbots can become a sort of a place for people to
let out their stresses [72]. It is hard to live alone in society, especially when there are infinite
factors that can make people have a rough day and stress builds and bursts out when there
is no way to release it; people can get angry at no one without a reason and this might lead
to a fight, which could bring down the quality of life around the community.

3.2. Ethical and Legal Implications

3.2.1. Patient Privacy and Data Security Concerns
The integration of AI, particularly LLMs, into healthcare conversations brings forth
ethical and legal considerations, with the foremost among them being patient privacy and
data security. As medical chatbots process sensitive health information, ensuring robust
measures for data encryption, storage, and transmission becomes paramount [73]. Ethical
considerations demand that patient data are handled with the utmost confidentiality
and that stringent protocols are in place to prevent unauthorized access or breaches,
safeguarding the trust patients place in AI-assisted healthcare interactions [74,75].

3.2.2. Ethical Considerations in AI-Assisted Healthcare Conversations

Beyond privacy concerns, ethical considerations play a pivotal role in the deployment
of AI-assisted healthcare conversations. Ensuring transparency and informed consent
becomes crucial when patients engage with medical chatbots. The ethical development and
use of AI models, including LLMs, involve addressing biases, avoiding discrimination, and
maintaining fairness in the provision of healthcare information [76,77]. Striking the right
balance between technological advancements and ethical principles is essential to build a
foundation of trust between patients, healthcare providers, and AI systems [78].

3.2.3. Regulatory Compliance in AI-Powered Healthcare Applications

Effectively managing regulatory complexities is a multifaceted challenge when in-
corporating AI-powered healthcare applications. It is crucial to uphold adherence to
established healthcare regulations, exemplified by the Health Insurance Portability and Ac-
countability Act (HIPAA) in the United States. Ensuring compliance with these regulatory
standards is not only essential for safeguarding patient rights but also forms the bedrock
for responsible AI deployment [79]. This adherence mitigates legal risks and fosters a
seamless integration of technology into healthcare practices. The evolving dynamics of
health-related conversations, driven by AI chatbots, necessitate a thorough understanding
of ethical and legal implications. As the regulatory landscape continues to evolve, it is
noteworthy to mention the European Union’s AI Act [80], which introduces regulations
specific to AI systems, emphasizing transparency, accountability, and user safety in the
deployment of AI technologies across various sectors [81]. Table 3 succinctly outlines the
concerns related to AI chatbots, encompassing ethical and legal dimensions. It emphasizes
Patient Privacy and Data Security, stressing the need for robust encryption and storage.
The Ethical Considerations highlight transparency, informed consent, and fairness, while
Regulatory Compliance underscores adherence to regulations like HIPAA and the Euro-
pean Union’s AI Act, ensuring a responsible AI deployment framework aligned with legal
and ethical standards.
BioMedInformatics 2024, 4 844

Table 3. Concerns of AI-assisted healthcare conversations with ethical and legal implications.

Concern Description
AI in healthcare raises concerns about patient data
Patient Privacy and Data Security security. Robust measures are needed for encryption
and storage.
Transparency and informed consent are crucial.
Ethical Considerations in AI-Assisted
Addressing biases and maintaining fairness in
Healthcare
healthcare is essential.
Adhering to healthcare regulations like HIPAA and
Regulatory Compliance in AI-powered
the European Union’s AI Act is crucial. It establishes a
Healthcare
framework for responsible AI deployment.

As we delve deeper into the ethical and legal implications, it is evident that maintain-
ing a delicate balance between technological innovation, patient privacy, and regulatory
compliance is crucial for the responsible and sustainable evolution of AI-assisted healthcare
conversations [18]. Subsequent sections will explore specific strategies, best practices, and
ongoing developments aimed at addressing these multifaceted considerations.

4. Discussion
The integration of conversational AI into established healthcare systems introduces
a spectrum of challenges, with technical hurdles at the forefront. Concerns encompass-
ing compatibility, interoperability with electronic health records, and potential resistance
from healthcare professionals pose substantial obstacles to the seamless assimilation of
AI into healthcare workflows. Beyond technical considerations, adherence to regulatory
standards and healthcare laws emerges as a paramount concern. This section explores the
intricate challenges tied to ensuring that AI applications in healthcare align with existing
regulations, taking into account regional variations in compliance requirements. More-
over, the risk of over-reliance on AI-driven solutions looms large, raising questions about
diminished human oversight and decision making. Striking a delicate balance between
leveraging technology for efficiency and preserving the human touch in healthcare becomes
imperative to ensure patients receive care that is both personalized and empathetic. In
addition, the implementation of advanced AI systems, particularly LLMs, bears significant
financial implications. The costs associated with acquiring, implementing, and maintain-
ing these technologies require careful consideration, particularly for healthcare facilities
operating with limited resources. These complex challenges emphasize the nuanced nature
of incorporating AI into healthcare practices.

4.1. Future Development and Challenges

It is believed that AI would evolve much quicker in the industry and especially the
chatbot in healthcare. It could be capable of performing symptom diagnosis more precisely
or effectively helping with people’s mental care [82,83]. However, there are concerns that
remain unsolved like privacy problems. On the other hand, there is a chance that not all
kinds of individuals understand the use of chatbots or are able to use chatbots, such as
people with disabilities and the elderly. Also, there are questions surrounding whether AI
chatbots are responsible for legal problems like plagiarism [84]. Regulations and uses of
LLMs remain controversial between the general public; people are still be concerned about
the above-mentioned aspects, so it is important to make LLMs a convenient and secure
tool for people to use comfortably without worrying about anything. Additionally, LLMs
should not be misused.
People question the security of using LLMs in healthcare when it comes to privacy
concerns, especially applying LLMs in healthcare. As LLMs could be able to handle patients’
health records and some other important personal information, then it is questionable
whether the programming of LLMs is secure enough for personal data management [85].
BioMedInformatics 2024, 4 845

Researchers are urged to look for a way to strengthen the protection of patients’ sensitive
information in order to allow people to use these models comfortably. In addition, the
regulation of the use of data and information for LLMs remains unclear and improper, and
hence any unauthorized access or misuse of LLMs is not preventable [86]. On the other
hand, it is required for the LLM to obtain access to health records so that it is pre-trained
for healthcare knowledge specifically to enhance its precision with information regarding
medicines [87]. Although it is impossible to prevent LLMs from accessing possibly sensitive
information, improving its security in terms of the way it stores or uses that information
needs to be focused on. Therefore, it is important to set up regulations for LLMs in different
fields, not just healthcare. However, regulations could be very different globally as there
are various standards on such controversial topics [85]. It is also believed that transparency
of the development of LLMs is important so that the trust from the general public can be
maintained and so that people might have the chance to understand what kind of data
are necessary and how they would be used for LLM development in order to apply it
into healthcare [88]. In addition, it is about ensuring that the actions of the LLMs using
those patients’ sensitive information for whatever need meet the privacy and security
requirements of the HIPAA [89].
There is nothing that can satisfy every single individual in the world, including the
seemingly perfect LLMs. Then, there is a question as to whether LLMs can satisfy everyone,
even those with disabilities and the elderly. For example, what if people with disabilities or
elderly people live alone and there is no one who can tell or teach them how to use LLMs;
then, they might never know what it is, how to use it, and even why they need to use it.
They might not be able to enjoy the convenience brought by the LLMs. Of course, they do
not have to use it to enhance their daily life as they used to live in the usual way, but it is
unfair to them that they cannot always enjoy what other people can. On the other hand,
they might then exhibit symptoms of later-life depression, unless they receive increased
care from their surroundings, such as having people to talk with, play with, or even just
having a walk around the street; all of these factors would help them to get rid of the stress
from daily life, as people have rough days sometimes and they need a way to let out their
emotions. Ideally, if people with disabilities and the elderly can easily learn how to deal
with new technologies and use them, then it could possibly make everyone’s life easier
and less trouble [90]. Because not all people will accept using new technology, even people
who are not disabled or elderly, we should not force everyone to learn and use LLMs but
information regarding their positive and negative factors should be disseminated to the
general public, as well as how developers would improve LLMs in different aspects to ease
the concerns that people have about them. As a result, people could gradually increase
their acceptance of using such technology. Hence, it is important to have volunteers to
help spread healthcare LLMs’ advantages and teach them to use them; once everyone
learns how to use LLMs correctly, they should be able to explore more functions themselves
without help. Then, people have someone to have a conversation with; even when it is late
at night and their family or friends are not available, they could still have a chatbot to talk
with, get advice from, and comfort their feelings to some extent [91].
The concept of hallucinations in LLMs refers to instances where the model generates
outputs that are factually incorrect, misleading, or unrelated to the input context. For
medical chatbots, hallucinations can have significant implications. If an LLM produces
inaccurate medical information or provides recommendations based on false premises, it
can compromise the reliability of the chatbot. This potential misinformation may lead to
misunderstandings, misdiagnoses, or inappropriate medical advice, posing risks to users’
well-being. Managing and mitigating hallucinations in LLMs is crucial for ensuring the
trustworthiness and safety of medical chatbots, underscoring the importance of ongoing
refinement and validation processes in their development. About the ethical concerns with
academic research, researchers questioned LLMs like ChatGPT about whether the resources
it provides to the users for research uses on academic work are reliable or not. According
to the research of Guleria et al. [92], ChatGPT was asked to write an article with a specific
BioMedInformatics 2024, 4 846

topic and to provide the resources it used in the article. The contents it wrote for the article
were deemed to be correct; however, the resources it listed as those used for the article
could not be found by the researchers themselves. As a result, the researchers believe that
ChatGPT is not as reliable as people think because the resources it provided could be made
up by the program itself, instead of researching and analyzing the scientific literature from
the internet [92]. In addition, the researcher performed this experiment to test ChatGPT’s
accuracy on the information related to academic research, and it seemed to be not accurate
enough in some fields; therefore, it is required to verify the output generated by ChatGPT
to prevent any misinterpretation [93]. Moreover, this shows that people should not blindly
trust the information provided by ChatGPT, especially information related to medicine
and any healthcare knowledge, as it could be putting the health and safety of the general
public in danger and also lead to medical misconduct in serious cases, and it cannot even
be responsible for any incorrect information it provides in any content of the scientific
literature [92]. Moreover, it cannot be proven whether the LLM would provide a similar
article to different users if they ask it to write articles on the same topic, such as academic
essay assignments; then, it could lead to a large scale of plagiarism. Because the LLMs are
trained with scientific studies that are copyrighted, the information or output, like articles it
is commanded to provide, could be copied from those scientific journals and people might
not know about this, causing plagiarism as a result [94]. Therefore, all LLM-generated texts
should be tested with a plagiarism detector. According to the research of Gao et al. [95],
AI-generated output is easily found in generated abstracts instead of original abstracts,
compared with a median of 99.98% for generated texts and a median of 0.02% for the
original texts [95].
On the other hand, researchers are looking forward to seeing improvement in LLMs in
the aspect of explainability, as they found that LLMs lack the ability to explain something
well with the use of detailed, step-by-step explanations for the information it provides
for users [96]. Moreover, it is also believed that the market size of chatbots will have a
significant increase in the future, as they keep improving and evolving [97].
The exploration of health-seeking behavior emerges as a noteworthy aspect in our
study. As individuals increasingly turn to the Internet for health information, the role of
chatbot LLMs in shaping healthcare advice-seeking behavior becomes a compelling avenue
for discussion. With the integration of technologies like Microsoft Bing (new Bing) with
ChatGPT, new opportunities and risks surface in the realm of health information acquisition.
Understanding how users interact with chatbot LLMs, the nature of information sought, and
the potential impact on health decisions is pivotal. Opportunities may include enhanced
accessibility to accurate information, empowering users in self-care. However, inherent
risks, such as the potential for misinformation or misinterpretation, necessitate careful
consideration. A thorough exploration of these dynamics will contribute valuable insights
into evolving the healthcare advice-seeking behavior facilitated by chatbot LLMs.

4.2. Opportunities for Improvement and Advancement

In the ever-evolving AI-assisted healthcare conversations, identifying opportunities for
improvement and advancement is crucial for addressing existing challenges and shaping
the future trajectory of this dynamic field.

4.2.1. Enhancing Precision and Accuracy

Optimizing patient safety involves enhancing the precision and accuracy of AI models,
especially LLMs, employed in medical chatbots [98]. Ongoing enhancements in these
models are achievable through meticulous training on diverse and specialized medical
datasets, effectively minimizing the potential risk of disseminating misinformation and
bolstering the dependability of user responses. Implementing strategies such as fine-tuning
LLMs for specific healthcare contexts and domains holds the potential to elevate their
performance, ensuring a more nuanced understanding of user queries and, consequently,
enhancing patient safety. Rigorous control measures, including continuous validation
BioMedInformatics 2024, 4 847

and refinement, are paramount to mitigating risks and upholding the integrity of medical
information disseminated by chatbots in healthcare settings.

4.2.2. Personalization and Context Awareness

There is significant potential in enhancing the personalization and context awareness
of AI-assisted healthcare conversations [99]. Tailoring responses based on individual
user profiles, medical histories, and preferences can create a more personalized and user-
centric experience. Advancements in contextual understanding, incorporating factors
such as patient context, emotional state, and real-time health data, can contribute to more
nuanced and effective interactions, ultimately improving the overall quality of healthcare
conversations.

4.2.3. Interdisciplinary Collaboration and Research

Opportunities for improvement extend to fostering interdisciplinary collaboration
and research initiatives. Collaborations between AI researchers, healthcare professionals,
ethicists, and legal experts can lead to comprehensive insights into the nuanced challenges
of implementing AI in healthcare [100]. This collaborative approach can facilitate the
development of robust frameworks, ethical guidelines, and innovative solutions that
address the multifaceted aspects of AI-assisted healthcare conversations.

4.2.4. User Education and Engagement

Promoting user education and engagement represents another avenue for improve-
ment. Initiatives focused on educating users about the capabilities and limitations of AI in
healthcare conversations can enhance transparency and trust [101]. Encouraging active user
participation in refining AI models, perhaps through feedback mechanisms, can contribute
to the iterative improvement of medical chatbots, aligning them more closely with user
expectations and needs. The improvement and future direction of AI-assisted healthcare
conversations is summarized in Table 4.

Table 4. An overview of the key opportunities for improvement and advancement in the context of
challenges and future directions in AI-assisted healthcare conversations.

Opportunities for Improvement and Advancement

Continuous refinement of LLMs through targeted training on diverse
Enhancing Precision and Accuracy
medical datasets to reduce misinformation and improve reliability.
Tailoring responses based on user profiles, medical histories, and
Personalization and Context Awareness
preferences for a more personalized and user-centric experience.
Tailoring responses based on individual user profiles, medical histories,
Interdisciplinary Collaboration and Research
and preferences for a more personalized and user-centric experience.
Initiatives to educate users about AI capabilities and limitations in
User Education and Engagement
healthcare, encouraging user feedback for iterative improvement.

As we explore these opportunities for improvement and advancement, it becomes

evident that the future of AI-assisted healthcare conversations is ripe with potential. By
proactively addressing these opportunities, we can pave the way for more sophisticated,
ethical, and user-centric applications, ultimately maximizing the positive impact of AI in
the healthcare domain.

5. Conclusions
The conclusion highlights the dual nature of LLMs, like ChatGPT, acting as versatile
tools across diverse fields while presenting potential drawbacks. Although LLMs enhance
work efficiency, particularly in healthcare, concerns related to privacy and accuracy persist.
NLP techniques, fundamental to LLMs, empower them to analyze diverse text sources,
enabling symptom diagnosis and recommendations for patients. Despite the potential
BioMedInformatics 2024, 4 848

advantages, ethical considerations, including privacy issues, and the need for specific train-
ing to enhance accuracy, remain pertinent. To navigate these challenges, concerted efforts
are required to address privacy concerns, enhance public understanding, and regulate
LLM usage. Future progress hinges on achieving consensus among governments and
developers for effective regulation, with specific considerations for people with disabilities
and elderly. Ethical concerns, including resource authenticity and potential plagiarism,
underscore the need for continuous improvement and oversight. Achieving a balance
between technological advancement and ethical considerations is crucial to foster trust and
ensure the widespread, confident utilization of LLMs in healthcare conversations.

Author Contributions: Conceptualization, J.C.L.C. and K.L.; methodology, J.C.L.C., V.W. and K.L.;
software, J.C.L.C., V.W. and K.L.; validation, J.C.L.C., V.W. and K.L.; formal analysis, J.C.L.C., V.W. and
K.L.; investigation, J.C.L.C., V.W. and K.L.; resources, J.C.L.C. and K.L.; data curation, J.C.L.C., V.W.
and K.L.; writing—original draft preparation, V.W. and J.C.L.C.; writing—review and editing, J.C.L.C.
and K.L.; visualization, V.W. and J.C.L.C.; supervision, J.C.L.C.; project administration, J.C.L.C. and
K.L.; funding acquisition, J.C.L.C. and K.L. All authors have read and agreed to the published version
of the manuscript.
Funding: This work is supported by Planning and Dissemination Grants—Institute Community
Support, Canadian Institutes of Health Research, Canada, under the grant numbers: CIHR PCS—
168296 and CHIR PCS—191021.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: No new data were created.
Acknowledgments: The authors would like to thank the support from Leslie Sanders from the
York University, Toronto, Canada. We acknowledge the use of ChatGPT [https://chat.openai.com/
accessed on 11 March 2024] to assist in proofreading the paper.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Confalonieri, R.; Coba, L.; Wagner, B.; Besold, T.R. A historical perspective of explainable Artificial Intelligence. Wiley Interdiscip.
Rev. Data Min. Knowl. Discov. 2021, 11, e1391. [CrossRef]
2. Kononenko, I. Machine learning for medical diagnosis: History, state of the art and perspective. Artif. Intell. Med. 2001, 23, 89–109.
[CrossRef]
3. Siddique, S.; Chow, J.C. Artificial intelligence in radiotherapy. Rep. Pract. Oncol. Radiother. 2020, 25, 656–666. [CrossRef]
4. Chow, J.C. Internet-based computer technology on radiotherapy. Rep. Pract. Oncol. Radiother. 2017, 22, 455–462. [CrossRef]
5. Joshi, G.; Jain, A.; Araveeti, S.R.; Adhikari, S.; Garg, H.; Bhandari, M. FDA-Approved Artificial Intelligence and Machine Learning
(AI/ML)-Enabled Medical Devices: An Updated Landscape. Electronics 2024, 13, 498. [CrossRef]
6. Raiaan, M.A.K.; Mukta, M.S.H.; Fatema, K.; Fahad, N.M.; Sakib, S.; Mim, M.M.J.; Ahmad, J.; Ali, M.E.; Azam, S. A Re-
view on Large Language Models: Architectures, Applications, Taxonomies, Open Issues and Challenges. IEEE Access 2024,
12, 26839–26874. [CrossRef]
7. Khan, W.; Daud, A.; Khan, K.; Muhammad, S.; Haq, R. Exploring the frontiers of deep learning and natural language processing:
A comprehensive overview of key challenges and emerging trends. Nat. Lang. Process. J. 2023, 4, 100026. [CrossRef]
8. Thirunavukarasu, A.J.; Ting, D.S.J.; Elangovan, K.; Gutierrez, L.; Tan, T.F.; Ting, D.S.W. Large language models in medicine. Nat.
Med. 2023, 29, 1930–1940. [CrossRef]
9. Kim, J.K.; Chua, M.; Rickard, M.; Lorenzo, A. ChatGPT and large language model (LLM) chatbots: The current state of acceptability
and a proposal for guidelines on utilization in academic medicine. J. Pediatr. Urol. 2023, 19, 598–604. [CrossRef]
10. Haupt, C.E.; Marks, M. AI-generated medical advice—GPT and beyond. JAMA 2023, 329, 1349–1350. [CrossRef] [PubMed]
11. Siddique, S.; Chow, J.C.L. Machine learning in healthcare communication. Encyclopedia 2021, 1, 220–239. [CrossRef]
12. Xu, L.; Sanders, L.; Li, K.; Chow, J.C.L. Chatbot for health care and oncology applications using artificial intelligence and machine
learning: Systematic review. JMIR Cancer 2021, 7, e27850. [CrossRef]
13. Chow, J.C.L.; Wong, V.; Sanders, L.; Li, K. Developing an AI-Assisted Educational Chatbot for Radiotherapy Using the IBM
Watson Assistant Platform. Healthcare 2023, 11, 2417. [CrossRef]
14. Kovacek, D.; Chow, J.C.L. An AI-assisted chatbot for radiation safety education in radiotherapy. IOP SciNotes 2021, 2, 034002.
[CrossRef]
BioMedInformatics 2024, 4 849

15. Lalwani, T.; Bhalotia, S.; Pal, A.; Rathod, V.; Bisen, S. Implementation of a Chatbot System using AI and NLP. Int. J. Innov. Res.
Comput. Sci. Technol. IJIRCST 2018, 6, 26–30. [CrossRef]
16. Wu, T.; He, S.; Liu, J.; Sun, S.; Liu, K.; Han, Q.-L.; Tang, Y. A brief overview of ChatGPT: The history, status quo and potential
future development. IEEE/CAA J. Autom. Sin. 2023, 10, 1122–1136. [CrossRef]
17. Li, J.; Dada, A.; Puladi, B.; Kleesiek, J.; Egger, J. ChatGPT in healthcare: A taxonomy and systematic review. Comput. Methods
Programs Biomed. 2024, 245, 108013. [CrossRef] [PubMed]
18. Chow, J.C.L.; Sanders, L.; Li, K. Impact of ChatGPT on medical chatbots as a disruptive technology. Front. Artif. Intell. 2023,
6, 1166014. [CrossRef] [PubMed]
19. Kao, H.-J.; Chien, T.-W.M.; Wang, W.-C.; Chou, W.; Chow, J.C. Assessing ChatGPT’s capacity for clinical decision support in
pediatrics: A comparative study with pediatricians using KIDMAP of Rasch analysis. Medicine 2023, 102, e34068. [CrossRef]
[PubMed]
20. Rawashdeh, B.; Kim, J.; AlRyalat, S.A.; Prasad, R.; Cooper, M. ChatGPT and artificial intelligence in transplantation research: Is it
always correct? Cureus 2023, 15, e42150. [CrossRef] [PubMed]
21. Chow, J.C.L.; Sanders, L.; Li, K. Design of an educational chatbot using artificial intelligence in radiotherapy. AI 2023, 4, 319–332.
[CrossRef]
22. Lubowitz, J.H. ChatGPT, an artificial intelligence chatbot, is impacting medical literature. Arthroscopy 2023, 39, 1121–1122.
[CrossRef] [PubMed]
23. Chin, H.; Lima, G.; Shin, M.; Zhunis, A.; Cha, C.; Choi, J.; Cha, M. User-Chatbot conversations during the COVID-19 pandemic:
Study based on topic modeling and sentiment analysis. J. Med. Internet Res. 2023, 25, e40922. [CrossRef]
24. Almalki, M.; Azeez, F. Health chatbots for fighting COVID-19: A scoping review. Acta Inform. Med. 2020, 28, 241–247. [CrossRef]
[PubMed]
25. Ayanouz, S.; Abdelhakim, B.A.; Benhmed, M. A smart chatbot architecture based NLP and machine learning for health care
assistance. In Proceedings of the 3rd International Conference on Networking, Information Systems & Security, Marrakech,
Morocco, 31 March–2 April 2020; pp. 1–6.
26. Olthof, A.W.; Shouche, P.; Fennema, E.M.; IJpma, F.F.; Koolstra, R.C.; Stirler, V.M.; van Ooijen, P.M.; Cornelissen, L.J. Machine
learning based natural language processing of radiology reports in orthopaedic trauma. Comput. Methods Programs Biomed. 2021,
208, 106304. [CrossRef] [PubMed]
27. Adamopoulou, E.; Moussiades, L. An overview of chatbot technology. In IFIP International Conference on Artificial Intelligence
Applications and Innovations; Springer: Cham, Switzerland, 2020; pp. 373–383.
28. Adamopoulou, E.; Moussiades, L. Chatbots: History, technology, and applications. Mach. Learn. Appl. 2020, 2, 100006. [CrossRef]
29. Chadha, N.; Gangwar, R.; Bedi, R. Current Challenges and Application of Speech Recognition Process using Natural Language
Processing: A Survey. Int. J. Comput. Appl. 2015, 131, 28–31. [CrossRef]
30. Malik, M.; Malik, M.K.; Mehmood, K.; Makhdoom, I. Automatic speech recognition: A survey. Multimed. Tools Appl. 2021,
80, 9411–9457. [CrossRef]
31. Zaib, M.; Zhang, W.E.; Sheng, Q.Z.; Mahmood, A.; Zhang, Y. Conversational question answering: A survey. Knowl. Inf. Syst. 2022,
64, 3151–3195. [CrossRef]
32. Reddy, S.; Chen, D.; Manning, C.D. Coqa: A conversational question answering challenge. Trans. Assoc. Comput. Linguist. 2019,
7, 249–266. [CrossRef]
33. Kocaleva, M.; Stojanov, D.; Stojanovik, I.; Zdravev, Z. Pattern recognition and natural language processing: State of the art. TEM J.
2016, 5, 236–240. [CrossRef]
34. Fu, T.; Gao, S.; Zhao, X.; Wen, J.-R.; Yan, R. Learning towards conversational AI: A survey. AI Open 2022, 3, 14–28. [CrossRef]
35. Sharma, D.; Paliwal, M.; Rai, J. NLP for Intelligent Conversational Assistance. Int. J. Innov. Res. Comput. Sci. Technol. 2021,
9, 179–184.
36. Locke, S.; Bashall, A.; Al-Adely, S.; Moore, J.; Wilson, A.; Kitchen, G.B. Natural language processing in medicine: A review. Trends
Anaesth. Crit. Care 2021, 38, 4–9. [CrossRef]
37. Lo Barco, T.; Kuchenbuch, M.; Garcelon, N.; Neuraz, A.; Nabbout, R. Improving early diagnosis of rare diseases using Natural
Language Processing in unstructured medical records: An illustration from Dravet syndrome. Orphanet J. Rare Dis. 2021, 16, 309.
[CrossRef] [PubMed]
38. Friedman, C.; Hripcsak, G. Natural language processing and its future in medicine. Acad. Med. 1999, 74, 890–895. [CrossRef]
[PubMed]
39. Khan, R.; Gupta, N.; Sinhababu, A.; Chakravarty, R. Impact of Conversational and Generative AI Systems on Libraries: A Use
Case Large Language Model (LLM). Sci. Technol. Libr. 2023, 42, 1–5. [CrossRef]
40. Alberts, I.L.; Mercolli, L.; Pyka, T.; Prenosil, G.; Shi, K.; Rominger, A.; Afshar-Oromieh, A. Large language models (LLM) and
ChatGPT: What will the impact on nuclear medicine be? Eur. J. Nucl. Med. 2023, 50, 1549–1552. [CrossRef] [PubMed]
41. Ethape, P.; Kane, R.; Gadekar, G.; Chimane, S. Smart Automation Using LLM. Int. Res. J. Innov. Eng. Technol. 2023, 7, 603.
42. El Saddik, A.; Ghaboura, S. The Integration of ChatGPT with the Metaverse for Medical Consultations. IEEE Consum. Electron.
Mag. 2024, 13, 6–15. [CrossRef]
43. Roumeliotis, K.I.; Tselikas, N.D. ChatGPT and Open-AI Models: A Preliminary Review. Future Internet 2023, 15, 192. [CrossRef]
BioMedInformatics 2024, 4 850

44. De Angelis, L.; Baglivo, F.; Arzilli, G.; Privitera, G.P.; Ferragina, P.; Tozzi, A.E.; Rizzo, C. ChatGPT and the rise of large language
models: The new AI-driven infodemic threat in public health. Front. Public Health 2023, 11, 1166120. [CrossRef]
45. Waisberg, E.; Ong, J.; Masalkhi, M.; Kamran, S.A.; Zaman, N.; Sarker, P.; Lee, A.G.; Tavakkoli, A. GPT-4: A new era of artificial
intelligence in medicine. Ir. J. Med Sci. 2023, 192, 3197–3200. [CrossRef]
46. Acheampong, F.A.; Nunoo-Mensah, H.; Chen, W. Transformer models for text-based emotion detection: A review of BERT-based
approaches. Artif. Intell. Rev. 2021, 54, 5789–5829. [CrossRef]
47. Sayeed, M.S.; Mohan, V.; Muthu, K.S. BERT: A Review of Applications in Sentiment Analysis. HighTech Innov. J. 2023, 4, 453–462.
[CrossRef]
48. Yang, Z.; Dai, Z.; Yang, Y.; Carbonell, J.; Salakhutdinov, R.R.; Le, Q.V. Xlnet: Generalized autoregressive pretraining for language
understanding. Adv. Neural Inf. Process. Syst. 2019, 32. Available online: https://api.semanticscholar.org/CorpusID:195069387
(accessed on 11 March 2024).
49. Raffel, C.; Shazeer, N.; Roberts, A.; Lee, K.; Narang, S.; Matena, M.; Zhou, Y.; Li, W.; Liu, P.J. Exploring the limits of transfer
learning with a unified text-to-text transformer. J. Mach. Learn. Res. 2020, 21, 5485–5551.
50. Hao, Y.; Dong, L.; Wei, F.; Xu, K. Visualizing and understanding the effectiveness of BERT. arXiv 2019, arXiv:1908.05620.
51. Catelli, R.; Pelosi, S.; Esposito, M. Lexicon-Based vs. Bert-Based Sentiment Analysis: A Comparative Study in Italian. Electronics
2022, 11, 374. [CrossRef]
52. Chow, J.C. Artificial intelligence in radiotherapy and patient care. In Artificial Intelligence in Medicine; Springer: Cham, Switzerland,
2021; pp. 1–13.
53. Yang, R.; Tan, T.F.; Lu, W.; Thirunavukarasu, A.J.; Ting, D.S.W.; Liu, N. Large language models in health care: Development,
applications, and challenges. Health Care Sci. 2023, 2, 255–263. [CrossRef]
54. Chakraborty, C.; Bhattacharya, M.; Lee, S.-S. Need an AI-enabled, next-generation, advanced ChatGPT or large language models
(LLMs) for error-free and accurate medical information. Ann. Biomed. Eng. 2023, 52, 134–135. [CrossRef] [PubMed]
55. Sanaei, M.-J.; Ravari, M.S.; Abolghasemi, H. ChatGPT in medicine: Opportunity and challenges. Iran. J. Blood Cancer 2023,
15, 60–67. [CrossRef]
56. Adhikari, K.; Naik, N.; Hameed, B.Z.; Raghunath, S.K.; Somani, B.K. Exploring the ethical, legal, and social implications of
ChatGPT in urology. Curr. Urol. Rep. 2023, 25, 1–8. [CrossRef] [PubMed]
57. Goodman, R.S.; Patrinely, J.R.; Stone, C.A.; Zimmerman, E.; Donald, R.R.; Chang, S.S.; Berkowitz, S.T.; Finn, A.P.; Jahangir, E.;
Scoville, E.A.; et al. Accuracy and reliability of chatbot responses to physician questions. JAMA Netw. Open 2023, 6, e2336483.
[CrossRef] [PubMed]
58. Walker, H.L.; Ghani, S.; Kuemmerli, C.; Nebiker, C.A.; Müller, B.P.; Raptis, D.A.; Staubli, S.M. Reliability of medical information
provided by ChatGPT: Assessment against clinical guidelines and patient information quality instrument. J. Med. Internet Res.
2023, 25, e47479. [CrossRef] [PubMed]
59. Fournier-Tombs, E.; McHardy, J. A medical ethics framework for conversational artificial intelligence. J. Med. Internet Res. 2023,
25, e43068. [CrossRef]
60. Chang, I.-C.; Shih, Y.-S.; Kuo, K.-M. Why would you use medical chatbots? interview and survey. Int. J. Med. Inform. 2022,
165, 104827. [CrossRef]
61. Chung, K.; Park, R.C. Chatbot-based heathcare service with a knowledge base for cloud computing. Clust. Comput. 2019,
22, 1925–1937. [CrossRef]
62. Kumar, Y.; Koul, A.; Singla, R.; Ijaz, M.F. Artificial intelligence in disease diagnosis: A systematic literature review, synthesizing
framework and future research agenda. J. Ambient. Intell. Humaniz. Comput. 2023, 14, 8459–8486. [CrossRef]
63. Lee, S.; Lee, J.; Park, J.; Park, J.; Kim, D.; Lee, J.; Oh, J. Deep learning-based natural language processing for detecting medical
symptoms and histories in emergency patient triage. Am. J. Emerg. Med. 2024, 77, 29–38. [CrossRef]
64. Wilkins, A. The robot doctor will see you soon. New Sci. 2023, 257, 28. [CrossRef]
65. DeSouza, D.D.; Robin, J.; Gumus, M.; Yeung, A. Natural language processing as an emerging tool to detect late-life depression.
Front. Psychiatry 2021, 12, 719125. [CrossRef]
66. Farhat, F. ChatGPT as a complementary mental health resource: A boon or a bane. Ann. Biomed. Eng. 2023, 51, 1–4. [CrossRef]
67. Cheng, S.W.; Chang, C.W.; Chang, W.J.; Wang, H.W.; Liang, C.S.; Kishimoto, T.; Chang, J.P.; Kuo, J.S.; Su, K.P. The now and future
of ChatGPT and GPT in psychiatry. Psychiatry Clin. Neurosci. 2023, 77, 592–596. [CrossRef]
68. Zhang, T.; Schoene, A.M.; Ji, S.; Ananiadou, S. Natural language processing applied to mental illness detection: A narrative
review. NPJ Digit. Med. 2022, 5, 46. [CrossRef] [PubMed]
69. Tanana, M.J.; Soma, C.S.; Kuo, P.B.; Bertagnolli, N.M.; Dembe, A.; Pace, B.T.; Srikumar, V.; Atkins, D.C.; Imel, Z.E. How do you
feel? Using natural language processing to automatically rate emotion in psychotherapy. Behav. Res. Methods 2021, 53, 2069–2082.
[CrossRef] [PubMed]
70. Madhuri, S. Detecting emotion from natural language text using hybrid and NLP pre-trained models. Turk. J. Comput. Math.
Educ. (TURCOMAT) 2021, 12, 4095–4103.
71. Pestian, J.; Nasrallah, H.; Matykiewicz, P.; Bennett, A.; Leenaars, A. Suicide note classification using natural language processing:
A content analysis. Biomed. Inform. Insights 2010, 3, BII.S4706. [CrossRef] [PubMed]
BioMedInformatics 2024, 4 851

72. Nijhawan, T.; Attigeri, G.; Ananthakrishna, T. Stress detection using natural language processing and machine learning over
social interactions. J. Big Data 2022, 9, 33. [CrossRef]
73. May, R.; Denecke, K. Security, privacy, and healthcare-related conversational agents: A scoping review. Inform. Health Soc. Care
2022, 47, 194–210. [CrossRef]
74. Li, J. Security Implications of AI Chatbots in Health Care. J. Med. Internet Res. 2023, 25, e47551. [CrossRef]
75. Hasal, M.; Nowaková, J.; Ahmed Saghair, K.; Abdulla, H.; Snášel, V.; Ogiela, L. Chatbots: Security, privacy, data protection, and
social aspects. Concurr. Comput. Pract. Exp. 2021, 33, e6426. [CrossRef]
76. Oca, M.C.; Meller, L.; Wilson, K.; Parikh, A.O.; McCoy, A.; Chang, J.; Sudharshan, R.; Gupta, S.; Zhang-Nunes, S. Bias and
inaccuracy in AI chatbot ophthalmologist recommendations. Cureus 2023, 15, e45911. [CrossRef]
77. Jin, E.; Eastin, M. Gender Bias in Virtual Doctor Interactions: Gender Matching Effects of Chatbots and Users on Communication
Satisfactions and Future Intentions to Use the Chatbot. Int. J. Hum.–Comput. Interact. 2023, 39, 1–13. [CrossRef]
78. Kim, J.; Cai, Z.R.; Chen, M.L.; Simard, J.F.; Linos, E. Assessing Biases in Medical Decisions via Clinician and AI Chatbot Responses
to Patient Vignettes. JAMA Netw. Open 2023, 6, e2338050. [CrossRef]
79. Pearman, S.; Young, E.; Cranor, L.F. User-friendly yet rarely read: A case study on the redesign of an online HIPAA authorization.
Proc. Priv. Enhancing Technol. 2022, 2022, 558–581. [CrossRef]
80. Ebers, M.; Hoch, V.R.S.; Rosenkranz, F.; Ruschemeier, H.; Steinrötter, B. The European Commission’s proposal for an artificial
intelligence act—A critical assessment by members of the robotics and AI law society (rails). J 2021, 4, 589–603. [CrossRef]
81. Schmidlen, T.; Schwartz, M.; DiLoreto, K.; Kirchner, H.L.; Sturm, A.C. Patient assessment of chatbots for the scalable delivery of
genetic counseling. J. Genet. Couns. 2019, 28, 1166–1177. [CrossRef] [PubMed]
82. Gräf, M.; Knitza, J.; Leipe, J.; Krusche, M.; Welcker, M.; Kuhn, S.; Mucke, J.; Hueber, A.J.; Hornig, J.; Klemm, P.; et al. Comparison
of physician and artificial intelligence-based symptom checker diagnostic accuracy. Rheumatol. Int. 2022, 42, 2167–2176. [CrossRef]
[PubMed]
83. Wiedermann, C.J.; Mahlknecht, A.; Piccoliori, G.; Engl, A. Redesigning Primary Care: The Emergence of Artificial-Intelligence-
Driven Symptom Diagnostic Tools. J. Pers. Med. 2023, 13, 1379. [CrossRef] [PubMed]
84. Jarrah, A.M.; Wardat, Y.; Fidalgo, P. Using ChatGPT in academic writing is (not) a form of plagiarism: What does the literature
say? Online J. Commun. Media Technol. 2023, 13, e202346. [CrossRef]
85. Meskó, B. The impact of multimodal large language models on health care’s future. J. Med. Internet Res. 2023, 25, e52865.
[CrossRef]
86. Meskó, B.; Topol, E.J. The imperative for regulatory oversight of large language models (or generative AI) in healthcare. NPJ
Digit. Med. 2023, 6, 120. [CrossRef] [PubMed]
87. Yu, P.; Xu, H.; Hu, X.; Deng, C. Leveraging Generative AI and Large Language Models: A Comprehensive Roadmap for
Healthcare Integration. Healthcare 2023, 11, 2776. [CrossRef]
88. Karabacak, M.; Margetis, K. Embracing Large Language Models for Medical Applications: Opportunities and Challenges. Cureus
2023, 15, e39305. [CrossRef]
89. Safranek, C.W.; Sidamon-Eristoff, A.E.; Gilson, A.; Chartash, D. The role of large language models in medical education:
Applications and implications. JMIR Med. Educ. 2023, 9, e50945. [CrossRef]
90. Valtolina, S.; Marchionna, M. Design of a Chatbot to Assist the Elderly. In International Symposium on End User Development;
Springer International Publishing: Cham, Switzerland, 2021; pp. 153–168.
91. Huq, S.M.; Maskeliūnas, R.; Damaševičius, R. Dialogue agents for artificial intelligence-based conversational systems for
cognitively disabled: A systematic review. Disabil. Rehabil. Assist. Technol. 2022, 17, 1–20. [CrossRef]
92. Guleria, A.; Krishan, K.; Sharma, V.; Kanchan, T. ChatGPT: Ethical concerns and challenges in academics and research. J. Infect.
Dev. Ctries. 2023, 17, 1292–1299. [CrossRef]
93. Iannantuono, G.M.; Bracken-Clarke, D.; Floudas, C.S.; Roselli, M.; Gulley, J.L.; Karzai, F. Applications of large language models in
cancer care: Current evidence and future perspectives. Front. Oncol. 2023, 13, 1268915. [CrossRef] [PubMed]
94. Abd-Alrazaq, A.; AlSaad, R.; Alhuwail, D.; Ahmed, A.; Healy, P.M.; Latifi, S.; Aziz, S.; Damseh, R.; Alrazak, S.A.; Sheikh, J. Large
Language Models in Medical Education: Opportunities, Challenges, and Future Directions. JMIR Med. Educ. 2023, 9, e48291.
[CrossRef]
95. Gao, C.A.; Howard, F.M.; Markov, N.S.; Dyer, E.C.; Ramesh, S.; Luo, Y.; Pearson, A.T. Comparing scientific abstracts generated by
ChatGPT to real abstracts with detectors and blinded human reviewers. NPJ Digit. Med. 2023, 6, 75. [CrossRef]
96. Hart, S.N.; Hoffman, N.G.; Gershkovich, P.; Christenson, C.; McClintock, D.S.; Miller, L.J.; Jackups, R.; Azimi, V.; Spies, N.;
Brodsky, V. Organizational preparedness for the use of large language models in pathology informatics. J. Pathol. Inform. 2023,
14, 100338. [CrossRef]
97. Chakraborty, C.; Pal, S.; Bhattacharya, M.; Dash, S.; Lee, S.-S. Overview of Chatbots with special emphasis on artificial intelligence-
enabled ChatGPT in medical science. Front. Artif. Intell. 2023, 6, 1237704. [CrossRef]
98. Caldarini, G.; Jaf, S.; McGarry, K. A literature survey of recent advances in chatbots. Information 2022, 13, 41. [CrossRef]
99. Prasad, G.; Ranjan, S.; Ankit, T.; Kumar, V. A personalized medical assistant chatbot: Medibot. Int. J. Sci. Technol. Eng. 2019,
5, 42–46.
BioMedInformatics 2024, 4 852

100. Følstad, A.; Araujo, T.; Law, E.L.-C.; Brandtzaeg, P.B.; Papadopoulos, S.; Reis, L.; Baez, M.; Laban, G.; McAllister, P.; Ischen, C.;
et al. Future directions for chatbot research: An interdisciplinary research agenda. Computing 2021, 103, 2915–2942. [CrossRef]
101. Kaur, A.; Singh, S.; Chandan, J.S.; Robbins, T.; Patel, V. Qualitative exploration of digital chatbot use in medical education: A pilot
study. Digit. Health 2021, 7, 1–11. [CrossRef] [PubMed]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.