Man, or Machine? Artificial Intelligence Language Systems in Plastic Surgery

Abstract

Artificial intelligence (AI) language models are computer programs trained to understand and generate human-like text. The latest AI language models available to the public have impressive language generation capability with immediate applications in both academia and private practice. Plastic surgeons can immediately leverage this technology to more efficiently allocate valuable human capital to higher-yield tasks. This can ultimately translate to higher patient volume, higher research output, and improved patient communication. Commercially available models offer business solutions that should not be ignored by plastic surgeons hoping to establish, optimize, or grow their practices. In this paper, the authors review the current state of AI language systems, discuss potential applications, and explore the risks and limitations of this technology.

As the cost of running a plastic surgery practice continues to increase and the overall landscape becomes more competitive, plastic surgeons who are able to leverage emerging technologies will have an edge over those who delay their adoption. Recently, marked advances in artificial intelligence (AI) language models have sparked widespread interest by the general public (Figure 1). These models have been trained on massive amounts of text data and have demonstrated the ability to generate highly coherent and human-like text. The use of AI language models has grown in various industries, including healthcare, finance, and e-commerce, to gain insights, improve customer service, and make predictions. AI language models are also being used in creative fields such as writing, music, and art to generate new content. Applications for this technology are relevant to all industries including plastic surgery. AI language models are becoming more advanced, more accurate, and more widely used. However, there are still many challenges to be addressed such as reducing the bias that these models can acquire from the data they trained on, maintaining accuracy of information, and optimizing models for specific uses. This paper aims to explore the utility of the current AI language models to encourage individuals to try this technology for themselves. We hope to provide a cursory introduction to this technology and highlight how it can be integrated into new and established plastic surgery practices.

WHAT IS AN Artificial Intelligence (AI) LANGUAGE MODEL?

The first AI language model to attract significant public attention was GPT-1 (Generative Pre-trained Transformer 1). It was released by OpenAI (San Francisco, CA) in 2018 and had the ability to generate text that was indistinguishable from text written by a human. This model was a breakthrough in the field of AI language models and set the stage for the development of more advanced models such as GPT-2, which was released the following year in 2019, and GPT-3, which was released in June 2020. These later models were trained on even more data and have shown even more advanced capabilities and more human-like text generation. Although these AI language models have only recently caught the attention of the public, the field of natural language processing (NLP) that these models derive from has been around since the 1950s with the development of early computer programs that were capable of analyzing and understanding human language.¹ This subfield of AI is concerned with the interaction between computers and human language. It involves using computational techniques to analyze, understand, and generate human language.

AI language models are a specific type of NLP model that are trained to generate text. With the advent of machine learning in the 1980s and 1990s, AI language models began to incorporate neural networks and other machine learning techniques, which allowed for more sophisticated models.² In recent years, the field of AI language models has seen a significant boost due to the availability of large amounts of text data and the development of more powerful computational hardware. This has led to the development of large-scale models such as GPT-1, GPT-2, GPT-3, T5 (Alphabet Inc., Mountain View, CA), and RoBERTa (Meta Platforms Inc., Melo Park, CA), which have demonstrated the ability to generate highly coherent and human-like text. Publicly available, user-friendly versions of these models include paid options from JapserAI (Austin, TX), FridayAI (Alwar, Rajasthan, India), and CopyAI (Memphis, TN), while OpenAI is currently offering its free-to-use ChatGPT.³ Both free and paid versions function in a similar manner: the user types in a request in plain language, such as “write a 100-word caption for an Instagram post about rhinoplasty recovery,” and the model will write plain language content that can be copied and pasted to any platform. Paid versions are presented in more user-friendly formats with templates for different social media sites that can be directly linked to these sites for a streamlined posting.

CURRENT AND FUTURE APPLICATIONS

Although AI technology has been used in several contexts within plastic surgery, research involving AI often uses custom-made models that require in-depth computer science knowledge to create and train.^4–7 With the recent development of publicly available and user-friendly AI programs, however, those with limited computer science knowledge will no longer be left out of the loop. Companies marketing these user-friendly AI language models to businesses have greatly enhanced the accessibility of this technology. These companies optimize AI language models for business use and offer them on easy-to-use platforms. Business-savvy plastic surgeons can find immediate utility in these services to help grow both new and established practices (Figure 2).

Copywriting

A heavily advertised use of these language models is copywriting. With online presence becoming increasingly important, plastic surgeons are now making use of multiple social media sites in addition to traditional websites and online reviews.^8–10 This requires a commitment to posting content often multiple times per day because social media algorithms reward high content output. Similarly, website search engine optimization (SEO) relies on large content volumes such as 1000-plus-word blog posts. This demand for large content volume necessitates constant idea generation, rapid content production, and up-to-date insight into current media trends. In response, several companies have optimized and repackaged AI language models for copywriting. AI models can be trained to analyze large amounts of data and create content that is tailored to a specific target audience through their use of demographic and psychographic data. They can generate high-quality, engaging content for websites, social media posts, blog articles, email campaigns, and other marketing materials. Outsourcing these tasks to social media management agencies, freelancers, or SEO specialists can cost hundreds or even thousands of dollars per month with individual blog posts costing upwards of US$50. In a specialized field such as plastic surgery, this outsourced content can vary greatly in quality and accuracy. In contrast, AI language models can access vast amounts of information to support their writing, can generate text in seconds, and are highly cost effective. The cost to use these models ranges from free to around US$500 per month for high-end specialized models. This content can be used as is for SEO-oriented blog posts or generic, fact-based articles. Alternatively, this content can serve as a starting point to facilitate customized human-generated content.

Chatbots

Another form of AI language model that has been optimized for commercial use are chatbots. AI chatbots have been used for customer service and support since the 1990s. These early chatbots were typically rule-based and were able to understand and respond to simple, predefined commands. In recent years, there has been a significant advancement in AI chatbot technology, particularly in the field of NLP. This has led to the development of chatbots that are able to understand and respond to human speech and text with a high degree of accuracy. Nowadays, chatbots are widely used across multiple industries, from customer service, healthcare, entertainment, education, and more. With the advancement of AI and NLP, chatbots are now able to simulate human-like conversation and can respond to a wide range of questions and requests. A chatbot on a plastic surgeon's website can be an excellent way to provide round-the-clock assistance to patients. A chatbot can be programmed with information about a surgeon's services, procedures, and pricing, which can help to educate and inform patients who are considering a procedure. Additionally, the chatbot can be designed to answer frequently asked questions and provide personalized recommendations and guidance. It can also be used to schedule consultations and follow up with patients after their procedure. Furthermore, chatbots can be integrated with the clinic's electronic medical records system, which allows them to provide more accurate and personalized advice to patients based on their medical history. They can be used to assist in preoperative instructions, postoperative communications, and follow up with patients, thus reducing the workload on the clinic staff. These chatbots can be integrated into websites to create synergy with social media marketing and SEO. They can serve as a first layer of prospective patient support and screening to reduce the workload of office staff. This can be particularly useful for multiphysician groups that utilize a single landing page. A chatbot can provide clickable links to patient requests for information on specific physicians, procedures, or before-and-after pictures. Although practices may not feel comfortable with chatbot involvement in all aspects of patient communication, chatbots are highly customizable. This allows practices to designate specific situations in which a chatbot is allowed to provide an answer, and when it should default to a qualified human provider.

Research

AI language models are also likely to cause a paradigm shift in the way research is done. AI language models can be a powerful tool for plastic surgery research as they can help analyze large volumes of data and extract insights that are difficult or time-consuming for researchers to identify manually. These models can assist in literature review, study design, report generation, and even predictive modeling.¹¹ For example, an AI language model can be trained on a large dataset of research papers and articles on plastic surgery to identify key trends and emerging topics in the field. This can be used to inform the design of future studies. Additionally, AI language models can be used to summarize data such as patient demographics, outcomes, and complications. This can drastically shorten the timeline for completing studies and reviews. Manually conducting systematic reviews or meta-analyses requires a team of multiple people, specialized software, and often many weeks of work. An optimized AI model can conceivably analyze hundreds of papers and extract relevant data in minutes.¹² We expect publicly available AI language models to soon be able to identify trends, patterns, and relevant information from raw data and report these findings in narrative text. The capabilities of these models will continue evolving as new techniques and technologies are developed. We are likely very soon to see full research papers written entirely, or in part, by AI language models.

LIMITATIONS AND RISKS

Although this technology is arguably still in its infancy, conversations about the potential ethical and legal risks of utilizing AI language models have already begun.¹³ A prominent concern with these models is the risk of bias. AI language models are trained on large amounts of data, and if those data contain biases, the model can inadvertently perpetuate those biases in its predictions and decisions. This can have a significant impact on individuals and groups that are disproportionately affected by these biases. For example, a language model that is trained on a dataset that contains mostly white, male faces could have difficulty identifying and recognizing faces of individuals who are not white or male. This could lead to inaccuracies or errors in systems that rely on that model and could result in unfair treatment of certain groups of people.

Another risk is privacy and security, as AI language models often process sensitive personal information, such as medical records, financial data, and location data. If these models are not properly secured, these data could be accessed by unauthorized individuals, which could result in serious privacy breaches. Furthermore, there is a risk of misinformation and disinformation when it comes to AI language models, especially with the rise of misinformation and disinformation online. Developers must ensure that information used to train these models is accurate, which presents a logistical hurdle if the model is pulling publicly available information to stay up to date. Additionally, there are risks related to transparency and accountability, as AI language models can be difficult to understand and interpret, making it hard to understand how they arrived at a particular decision or prediction. This can make it difficult to hold the developers and users of these models accountable for any errors or negative impacts that they may cause.

Lastly, there is a risk of job displacement as AI language models can automate certain tasks and processes, which could lead to job losses for humans, particularly for trainees in plastic surgery who currently perform many of the tasks that will likely be the first to become automated. Trainees such as students and residents rely on these tasks to get their foot in the door and gain experience while providing value to attending physicians. As AI language models continue evolving and new risks arise, it is important for organizations to have robust ethical and legal frameworks in place, and to continuously monitor and assess the impact of these models on individuals and on society as a whole.

CONCLUSIONS

Although AI language systems are still in their infancy, they already have practical utility. Nearly half this paper (45% in fact) was written by the publicly accessible, free-to-use ChatGPT language model (Figure 3). ChatGPT was asked to make an outline for this paper; this was used to ask ChatGPT questions to produce the content in each section, and content was then added to increase the specificity of information and create smoother transitions between sections. Final edits were made to include information that was not yet available to the language system, and to better match the writing style that is expected in academic journals. This meta-application of AI language models highlights both the immediate utility of this technology and the need for a concerted effort to create uniform policies that will guide human-AI collaboration. Guidelines for appropriate disclosure of AI language model involvement in manuscript creation will need to be establish considering the technology's ability to generate a large percentage of manuscripts (Table 1). Overall, AI language models and plastic surgeons can create synergy to increase productivity without having to spend more or give up more of their valuable time; however, this collaboration must be done with caution to avoid legal and ethical pitfalls (Video).

Although we believe this overview provides a starting point for those who are new to the AI language model space, we must acknowledge the limitations of this article. First and foremost, this cross-sectional discussion of a rapidly growing field is inherently limited because updates are likely to be abundant and frequent. Additionally, the discussion of future applications of the technology is speculative, although based on natural advancements of current capabilities. We also acknowledge that there are profound differences between human thinking, reasoning, and language that AI language models may never be able to replicate; however, an in-depth exploration of these differences is outside the scope of this article.¹² On a practical level, AI language models are ultimately tools, and it is up to each individual user to determine if there is an application that they find valuable. We hope this article serves as a guide for plastic surgeons to leverage AI language models in their own practices.

Supplemental Material

This article contains supplemental material located online at www.aestheticsurgeryjournal.com.

Disclosures

The authors have no disclosures related to this work.

Funding

No funding or other support was received related to this work.

REFERENCES

Author notes