ChatGPT in veterinary medicine: a practical guidance of generative artificial intelligence in clinics, education, and research

ChatGPT, the most accessible generative artificial intelligence (AI) tool, offers considerable potential for veterinary medicine, yet a dedicated review of its specific applications is lacking. This review concisely synthesizes the latest research and practical applications of ChatGPT within the clinical, educational, and research domains of veterinary medicine. It intends to provide specific guidance and actionable examples of how generative AI can be directly utilized by veterinary professionals without a programming background. For practitioners, ChatGPT can extract patient data, generate progress notes, and potentially assist in diagnosing complex cases. Veterinary educators can create custom GPTs for student support, while students can utilize ChatGPT for exam preparation. ChatGPT can aid in academic writing tasks in research, but veterinary publishers have set specific requirements for authors to follow. Despite its transformative potential, careful use is essential to avoid pitfalls like hallucination. This review addresses ethical considerations, provides learning resources, and offers tangible examples to guide responsible implementation. A table of key takeaways was provided to summarize this review. By highlighting potential benefits and limitations, this review equips veterinarians, educators, and researchers to harness the power of ChatGPT effectively.


Introduction
Artificial intelligence (AI) is a trending topic in veterinary medicine.Explorations in this field include applications such as dental radiograph (1), colic detection (2), and mitosis detection in digital pathology (3).Machine learning (ML), a subset of AI, enables systems to learn from data without being explicitly programmed (4).Generative AI, in turn, is a field within ML specializing in creating new content, leading to the development of large language models (LLMs) for their human-like text generation capabilities.Notable LLMs such as ChatGPT (OpenAI) (5), Llama 2 (Meta) (6), Gemini (Google) (7), and Claude 2.1 (Anthropic) (8), offer diverse functionalities-some are downloadable, others open source, or integrated into productivity software.ChatGPT stands as the most accessible for newcomers to this technology.
GPT, or Generative Pre-trained Transformer, excels in generating new text, images, and other content formats rather than solely analyzing existing data.It is pre-trained by exposure to vast datasets of text and code, enabling it to recognize patterns and generate human-like responses.It employs the transformer neural network architecture that is particularly adept at processing language, which enables coherent and contextually relevant outputs (9).The free version of ChatGPT 3.5 (GPT-3.5)provides the capability of answering questions, providing explanations, generating creative content, offering advice, conducting research, engaging in conversation, supporting technical tasks, aiding with education, and creating summaries.With the release of ChatGPT 4.0 (GPT-4) (https://chat.openai.com/), a subscription-based model, ChatGPT 4.0, the medical field has taken particular interest due to its expanded knowledge base and capabilities in text, image, and voice analysis and generation.Most recently, OpenAI introduced Sora, a model capable of generating ultra-real videos solely from text prompts (10).Despite ChatGPT's widespread use, a comprehensive review of its applications in veterinary medicine is lacking.
The breadth of ChatGPT in medicine covers a wide range of areas, ranging from answering patient and professional inquiries, promoting patient engagement (11), diagnosing complex clinical cases (12), and creating educational material (13).In veterinary medicine, ChatGPT has been integrated to enhance virtual assistance, diagnostic accuracy, communication with pet owners, and optimization of workflows (14)(15)(16)(17)(18)(19)(20).While examples of ChatGPT applications are prevalent on social media and in various publications (21)(22)(23), the best way to understand its impact is through direct engagement.This article aims to discuss the applications of ChatGPT in veterinary medicine, provide practical implementations, and examine its limitations and ethical considerations.Highlights of each section are listed in Table 1.

ChatGPT 101: Prompts and Prompt Engineering
Understanding prompts is crucial before engaging with ChatGPT or other generative AI tools.Prompts act as conversation starters, consisting of instructions or queries that elicit responses from the AI.Effective prompts for ChatGPT integrate relevant details and context, enabling the model to deliver precise responses (24).Prompt engineering is the practice of refining inputs to produce optimal outputs.For instance, researchers instructing ChatGPT to identify body condition scores from clinical records begin prompts by detailing the data structure and desired outcomes: "Each row of the dataset is a different veterinary consultation.In the column 'Narrative' there is clinical text.Your task is to extract Body Condition Score (BCS) of the animal at the moment of the consultation if recorded.BCS can be presented on a 9-point scale, example BCS 6/9, or on a 5-point scale, example BCS 3.5/5.Your output should be presented in a short-text version ONLY, following the rules below: . . .(omitted) (24)" Writing effective prompts involves providing contextual details in a clear and specific way and willingness to refine them as needed.
Moreover, incorporating 'cognitive strategy prompts' can direct ChatGPT's reasoning more effectively (refer to Supplementary Material for more details).For a comprehensive understanding of prompt engineering, readers are encouraged to refer to specialized literature and open-access online courses dedicated to this subject (25)(26)(27)(28).Proper prompt engineering is pivotal for shaping conversations and obtaining the intended results, as illustrated by various examples in this review.

Using ChatGPT in Clinical Care
ChatGPT has the potential to provide immediate assistance upon the client's arrival at the clinic.The pre-trained ChatGPT-4 model is adept at processing chief complaints, vital signs, and medical histories entered by emergency medicine physicians, subsequently making triage decisions that align closely with established standards (29).Given that healthcare professionals in the United States spend approximately 35% of their time documenting patient information (30) and that note redundancy is on the rise (31), ChatGPT 's ability to distill crucial information from extensive clinical histories and generate clinical documents are particularly valuable (32).A veterinary study utilizing ChatGPT 3.5 Turbo for text mining demonstrated the AI's capability to pinpoint all overweight Body Condition Score (BCS) instances within a dataset with high precision (24).However, some limitations were noted, such as the misclassification of lameness scoring as BCS, an issue that the researchers believe could be addressed through refined prompt engineering (24).
In the context of daily clinical documentation, veterinarians can input signalment, clinical history, and physical examination findings into ChatGPT to generate Subjective-Objective-Assessment-Plan (SOAP) notes (30).An illustrative case presented in Supplementary Material involved the generation of a SOAP note for a canine albuterol toxicosis incident (33), where ChatGPT efficiently identified the diagnostic tests executed in the case report, demonstrating that ChatGPT can be used as a promising tool to streamline the workflow for veterinarians.
Moreover, recent research has investigated ChatGPT's proficiency in addressing clinical challenges.One study found that GPT-4 could accurately diagnose 57% of complex medical cases, a success rate that outperformed 72% of human readers of medical journals in answering multiple-choice questions (12).Additionally, GPT-4's top diagnosis concurred with the final diagnosis in 39% of cases and included the final diagnosis within the top differential diagnoses in 64% of cases (34).With the updated image upload function, the capability of ChatGPT-4 extends to the interpretation of blood work images.The Supplementary Material illustrates an example of ChatGPT analyzing Case of the Month on eClinPath (35) and providing the correct top differential despite its limited ability to interpret the white blood cell dot plot.
ChatGPT-4 can interpret ECGs and outperformed other LLM tools in correctly interpreting 63% of ECG images (36).However, it's worth noting that in a specific example provided in the Supplementary Material, ChatGPT did not identify an atypical atrial flutter with intermittent Ashman phenomenon in a 9-year-old Pug despite the addition of asterisks in the ECG to indicate the wide and tall aberrant QRS complexes (35).This example emphasizes that while ChatGPT is a powerful tool, it cannot replace specialized AI algorithms approved by the Food and Drug Administration (FDA) for ECG interpretation.(37,38).Nevertheless, advances in veterinary-specific AI tools, such as a deep learning model for canine ECG classification, are on the horizon, with the potential to be available soon (39).

Using ChatGPT in Veterinary Education
Recent studies leveraging Large Language Models (LLMs) in medical examinations underscore their utility in educational support.ChatGPT 3's performance, evaluated using 350 questions from the United States Medical Licensing Exam (USMLE) Steps 1, 2CK, and 3, was commendable, achieving scores near or at the passing threshold across all three levels without specialized training (40).This evaluation involved modifying the exam questions into various formatsopen-ended or multiple-choice with or without a forced justification-to gauge ChatGPT's foundational medical knowledge.The AI-generated responses often included key insights, suggesting that ChatGPT's output could benefit medical students preparing for USMLE (40).
Another investigation benchmarked the efficacy of GPT-4, Claude 2, and various open-source LLMs using multiple-choice questions from the Nephrology Self-Assessment Program.Success rates varied widely, with open-source LLMs scoring between 17.1-30.6%,Claude 2 at 54.4%, and GPT-4 leading with 73.7% (41).A comparative analysis of ChatGPT-3.5 and ChatGPT-4 indicates the newer version substantially improved in the neonatal-perinatal medicine board examination (42).Veterinary researchers at the University of Georgia used GPT-3.5 and GPT-4.0 to answer faculty-generated 495 multiple-choice and true/false questions from 15 courses in the third-year veterinary curriculum (43).The result concurred with the previous study that GPT-4.0 (77% correct rate) performed substantially better than GPT-3.5 (55% correct rate); however, their performance is significantly lower than that of veterinary students (86%).These studies highlight the variances in LLM knowledge bases, which could affect the quality of medical and veterinary education.
Beyond exam preparation, the ChatGPT 4 Plus subscription enables users to craft tailored versions of ChatGPT, referred to as GPTs, for diverse uses (41).Veterinary educators, for instance, can harness these tools to develop AI tutors to boost veterinary students' learning.An example of a custom GPT is a specialized veterinary clinical pathology virtual tutor named Vet Clin Path Resident (44).This custom GPT draws from legally available open-access textbooks with Creative Commons licenses (45)(46)(47) and the eClinPath website (35), ensuring the information provided is sourced from credible references.Students are encouraged to pose any question pertinent to veterinary clinical pathology and can even request specific references or links to web pages.More information about this GPT is detailed in the Supplementary Material.

Using ChatGPT in Academic Writing
Leveraging editing services enhances clarity and minimizes grammatical errors in scientific manuscripts, which can improve their acceptance rate (48).While acknowledgments often thank editorial assistance, the use of spelling-checking software is rarely disclosed.Nowadays, AI-powered writing assistants have integrated advanced LLM capabilities to provide nuanced suggestions for tone and context (45), thus merging the line between original and AI-generated content.Generative AI, like ChatGPT, extends its utility by proposing titles, structuring papers, crafting abstracts, and summarizing research, raising questions about the AI's role in authorship as per the International Committee of Medical Journal Editors' guidelines (49) (Supplementary Material).Notably, traditional scientific journals are cautious with AI, yet NEJM AI stands out for its advocacy for LLM use (50).However, these journals still refrain from recognizing ChatGPT as a co-author due to accountability concerns over accuracy and ethical integrity (50)(51)(52).The academic community remains wary of ChatGPT's potential to overshadow faculty contributions (53).
Several veterinary journals have updated their guidelines in response to the emergence of generative AI.Among the top 20 veterinary medicine journals as per Google Scholar (54), 14 instruct on generative AI usage (Supplementary Material).They unanimously advise against listing AI as a co-author, mandating disclosure of AI involvement in Methods, Acknowledgments, or other designated sections.These recommendations typically do not apply to basic grammar and editing tools (Supplementary Material).Research shows ChatGPT can enhance writing efficiency, particularly benefiting less skillful writers, suggesting academia's broader acceptance could mitigate productivity inequality, fostering a more inclusive scholarly community (23).
A pertinent question is the detectability of AI-generated content without explicit disclosure.A study revealed that reviewers could identify 68% of ChatGPTproduced scientific abstracts; however, they also mistakenly tagged 14% of original works as AI-generated (55).In another study, veterinary neurologists only had a 31-54% success rate in distinguishing AI-crafted abstracts (56), highlighting the risk of misjudging authentic academic work for AI-generated content.This misconception led to an ecologist at Cornell University being falsely accused of writing fraud by a reviewer who deemed her work as "obviously ChatGPT," resulting in publication rejection (57).
To counteract this, a 'ChatGPT detector' has been suggested.An ML tool utilizes distinguishing features like paragraph complexity, sentence length variability, punctuation marks, and popular wordings, achieving over 99% effectiveness in identifying AI-authored texts (58).A subsequent refined model can further distinguish human writings from ChatGPT-3.5 and ChatGPT-4 writings in chemistry journals with 99% accuracy (56).While these tools are not publicly accessible, OpenAI is developing a classifier to flag AI-generated text (57), emphasizing the importance of academic integrity and responsible AI use.

ChatGPT's Ethical Issues and Limitations
A recent survey on AI in veterinary medicine by Digitail and the American Animal Hospital Association, involving veterinarians, veterinary technicians/assistants, and students, showed 83.8% of respondents were familiar with AI and its applications in veterinary medicine, with 69.5% using AI tools daily or weekly (59).Yet, 36.9% remain skeptical, citing concerns about the systems' reliability and accuracy (70.3%), data security and privacy (53.9%), and the lack of training (42.9%) (59).This section highlights critical concerns to equip veterinary professionals for the upcoming AI era.

Hallucination
Hallucination, or artificial hallucination, refers to the generation of implausible but confident responses by ChatGPT, which poses a significant issue (60).ChatGPT is known to create fabricated references with incoherent Pubmed ID (61), a problem somewhat mitigated in ChatGPT 4 (18% error rate) compared to ChatGPT 3.5 (55% error rate) (62).The Supplementary Material illustrated an example where ChatGPT could have provided more accurate references, including PMIDs, underscoring its limitations for literature searches.

Cybersecurity and Privacy
As an LLM, ChatGPT is trained using undisclosed but purportedly accessible online data and ongoing refinement through user interactions during conversations (62).It raises concerns about copyright infringement and privacy violations, as evidenced by ongoing lawsuits against OpenAI for allegedly using private or public information without their permission (63)(64)(65).Based on information from the OpenAI website, user-generated content is consistently gathered and used to enhance the service and for research purposes (66).Concerns about data security are amplified when analyzing clinical data, suggesting a preference for uploading de-identified datasets.Alternatively, considering local installations of open-source, free-for-research-use LLMs, like Llama 2 or Gemma (Google), for enhanced security is recommended (67)(68)(69)(70).

U.S. FDA Regulation
While the FDA has approved 692 AI and ML-enabled human medical devices, primarily in radiology (76.7%), followed by cardiology (10.3%) and neurology (2.9%) (71), veterinary medicine lacks specific premarket requirements for AI tools.The AI-and ML-enabled veterinary products currently span from dictation and notetaking apps (17,18), management and communication software (19,20), radiology service (14-16), and personalized chemotherapy (72), to name a few.These products may or may not have scientific validation (73)(74)(75)(76)(77)(78)(79)(80) and may be utilized by veterinarians despite the clients' lack of consent or complete understanding.The absence of regulatory oversight in veterinary medicine, especially in diagnostic imaging, calls for ethical and legal considerations to ensure patient safety in the United States and Canada (81,82).

Practical Learning Resources
Resources for learning about ChatGPT and generative AI are abundant, including OpenAI's documentation (83), online courses from Vanderbilt University via Coursera (25), Harvard University's tutorial for generative AI (84), and the University of Michigan's guides on using generative AI for scientific research (85).These resources are invaluable for veterinarians seeking to navigate the evolving landscape of AI in their practice.Last but not least, readers are advised to engage ChatGPT with well-structured prompts, such as: 'I'm a veterinarian with no background in programming.I'm interested in learning how to use generative AI tools like ChatGPT.Can you recommend some resources for beginners?' (Supplementary Material).

The Ongoing Dialogue
In the 2023 Responsible AI for Social and Ethical Healthcare (RAISE) Conference held by the Department of Biomedical Informatics at Harvard Medical School, the discussion on the judicious application of AI in human healthcare highlighted principles that could be effectively adapted to veterinary medicine (86).Integrating AI into veterinary practices should amplify the benefits to animal welfare, enhance clinical outcomes, broaden access to veterinary services, and enrich the patient and client experience.AI should support rather than replace veterinarians, preserving the essential human touch in animal care.
Transparent and ethical utilization of patient data is paramount, advocating for opt-out mechanisms in data collection processes while safeguarding client confidentiality.AI tools in the veterinary field ought to be envisioned as adjuncts to clinical expertise, with a potential for their role to develop progressively, subject to stringent oversight.The growing need for direct consumer access to AI in veterinary medicine promises advancements but necessitates meticulous regulation to assure pet owners about data provenance and the application of AI.
This review discussed the transformative potential of ChatGPT across clinical, educational, and research domains within veterinary medicine.Continuous dialogue, awareness of limitations, and regulatory oversight are crucial to ensure generative AI augments clinical care, educational standards, and academic ethics rather than compromising them.The examples provided in the Supplementary Material encourage innovative integration of AI tools into veterinary practice.By embracing responsible adoption, veterinary professionals can harness the full potential of ChatGPT to make the next paradigm shift in veterinary medicine.

Introduction
• Machine learning (ML) is a subset of artificial intelligence (AI) that enables systems to learn from data without being explicitly programmed.

Figure 1 .
Figure 1.Visual Abstract of the Review.
Google), and Claude 2.1(Anthropic).•GPTstands for Generative Pre-trained Transformer, indicating its characteristics of content generation, pre-trained by text and codes, and the use of transformer neural network.ChatGPT's Ethical Issues and Limitations • Most veterinary professionals are familiar with AI and its application in veterinary medicine, while some remain skeptical about its reliability and accuracy, data security and privacy, and a lack of training.Hallucination• Hallucination, or artificial hallucination, refers to the generation of implausible but confident responses by ChatGPT, which poses a significant issue.Cybersecurity and Privacy• ChatGPT is trained using undisclosed but purportedly accessible online data, and user-generated content is consistently gathered by OpenAI.• When analyzing clinical data, uploading de-identified datasets is recommended.• Alternatively, considering local installations of open-source, free-for-research-use LLMs, like Llama 2 or Gemma, for enhanced security.U.S. FDA Regulation • Most FDA-approved AI and ML-enabled human medical devices are in the field of radiology, followed by cardiovascular and neurology.• FDA has not set premarket requirements for AI tools in veterinary medicine.• The AI-and ML-enabled veterinary products include dictation and notetaking apps, management and communication software, and radiology services, which may or may not have scientific validation.Practical Learning Resources • Resources for learning about ChatGPT and generative AI are abundant, including OpenAI's documentation, online courses from Vanderbilt University via Coursera, Harvard University's tutorial for generative AI, and the University of Michigan's guides on using generative AI for scientific research.• Readers are encouraged to ask ChatGPT about learning resources: https://chat.openai.com