The Dangers of ChatGPT
for the Skincare Community

This transcript was exported on February 12, 2024 -view latest version here.

Skincare can sometimes feel overwhelming, whether it’s finding the right products, ingredients, or treatments. There’s a lot out there, but not always for people of African, Hispanic, Middle Eastern and Eastern South Asian descent. That’s why I set out to educate myself and others so that we can all feel beautiful in our skin. Hello and welcome back to The Skin Report. I’m Dr. Simran Sethi, an internal medicine doctor, mom of three and CEO and founder of ReNewMD Medical Spas and SKIN by Dr. Sethi.

Today onThe Skin Report Podcast, I want to cover a timely topic, which is the harm that generative AI technology poses for consumers, people of color, and the skincare community as a whole. But stay tuned until the end where we’ll discuss some positive ways that AI technology can be utilized for skincare purposes. In November of 2022, OpenAI released an early demo of their artificial intelligence chatbot, ChatGPT. Odds are you’ve heard of it as it has since become a hot topic with some excited about its potential uses and others apprehensive about its drawbacks. But artificial intelligence or AI is a technology that’s been used for years and is constantly advancing. So why is it so relevant now and what does it have to do with skincare?

Since ChatGPT gives users the power to generate writing on virtually any subject, this means that no topic is safe from its potential risks. Primarily, these dangers can include the spread of false or discriminatory information. So today I wanted to discuss the harm that can occur when merging AI technology with skincare information. Primarily, I will be discussing the dangers of using AI technology to generate informative content on skincare, skin health and skin science. Through the episode, I’ll explain how this misuse can have detrimental consequences to the credibility of skincare resources and publications, the safety of skincare consumers, and the representation of people of color in the skincare space. But before we begin, let’s briefly go over AI technology as a concept and the specific issues that have emerged as a result of ChatGPT’s boom in popularity.

AI or artificial intelligence is technology that simulates human intelligence through machines or software. ChatGPT is an AI chatbot and natural language processing tool that allows users to provide instructions as a prompt and will reply with a detailed human-like response. Students, writers, and pretty much everyone else soon discovered how easy it was to generate written content Using ChatGPT. It allowed them to query writing on all topics that was formatted and styled as they desired. It sounds like an awesome tool and the technology behind it truly is remarkable. However, by looking at exactly how ChatGPT works to develop its responses, we can see how this technology can pose significant issues when improperly used. ChatGPT generates responses based on prompts, but in order for any AI technology to do this, it must first be trained. The ChatGPT-1 website states that the technology uses large amounts of prior knowledge to provide coherent and contextually relevant answers. It is trained on a wide data set from the internet allowing it to handle a variety of subjects. In simpler terms, this means that ChatGPT generates its responses based on the information that it gathers from the internet.

The majority of the issues with ChatGPT boiled down to this fact, that the technology utilizes data from the internet to generate its responses. And unfortunately, the information found across the worldwide web is not always accurate, unbiased, or inclusive. Coming up next, we’ll start to unravel the many harmful potentials of ChatGPT technology, beginning with the spread of misinformation.

When you were in school, odds are you were educated about the importance of credibility, whether it was in the form of including citations when writing a paper to discerning trustworthy sources of scientific studies. We as a society have come to rely upon credibility as a way to ensure the trustworthiness and reliability of a piece of information. Today, the internet is bombarded with endless sources of information, including uninformed opinions, biased claims, and let’s face it, downright falsities. Thinking of the last time you looked on a popular forum or read the comments on a social media site, freedom of Speech allows people to share their thoughts online with very few restrictions on what they can and cannot say. While you and I may be able to discern the trustworthiness of a claim based on its source, natural language processing tools like ChatGPT haven’t quite caught up.

That said, content created by ChatGPT can be generated using so-called facts found online that aren’t actually credible. Say you ask the chatbot a question about a certain skincare product, ingredient or practice, the response that you will receive will provide information using data that it found in all corners of the internet, not necessarily credible ones. This can of course pose a significant risk to users who are seeking reliable information. It is especially dangerous considering how ChatGPT’s non-credibility may influence real life decisions people make regarding their skin’s health.

The PubMed Central article titled Opportunities and Risks of ChatGPT in Medicine, Science and Academic Publishing: A Modern Promethean Dilemma discusses the potential of ChatGPT in medical science. While the article did credit ChatGPT’s ability to create content with impressive presentation, it explained that the chatbots writing cannot be relied on for accuracy. The article states, “ChatGPT generated manuscripts might be misleading based on non-credible or completely made up sources. The worst part is the ability of ChatGPT to write a text of surprising quality might deceive viewers and readers, with the final result being an accumulation of dangerous misinformation.” It would be easy enough to tell people directly not to source information from ChatGPT. However, the ease at which one can access ChatGPT generated content has made the technology tempting to so many so-called skincare info resources. Bloggers, authors, and even podcasters have been guilty of distributing plagiarized content ripped straight from ChatGPT’s chatbot. But for now, I want to focus on how the spread of misinformation can harm skincare users and consumers in betraying their trust.

I believe that the most significant example of this lies in a phenomena surrounding AI technology known as AI hallucination. We’ve gone over how ChatGPT rarely gains its information from credible sources. But what’s perhaps even more important to consider is that ChatGPT is not a technology intended for sourcing factual information at all. ChatGPT is first and foremost a chatbot. Its primary purpose is to respond to queries in a way that is conversational and to fulfill the request that the user has described in the query. AI hallucination is a result of the technology’s aim at producing output that aligns with the prompt’s intent. It is because of this that ChatGPT often generates fictitional responses and will try to pass them off as fact. After all the output created by AI, however false it may be still, satisfies the primary function of the technology, which is to produce applicable output, not to produce factual information.

This AI hallucination can be especially dangerous when ChatGPT is used to write about skincare practices, products, ingredients, or formulations because skincare impacts skin health. Therefore, accuracy is key when discussing these matters to avoid skin health risk resulting from misinformation. All of that said, ChatGPT cannot currently be considered a credible source of information on skincare, skin health or skin signs, and using it as a tool to generate content on skincare can potentially lead to the spread of harmful misinformation.

Next, I’ll cover another side effect of ChatGPT’s technology, which is how the generation of biased content may further isolate people of color from the skincare community. If you’ve been listening to my podcast for a while, you know how strongly I feel about the need for information, resources and products specifically focused on treating skin of color. For those of you who may not know, the skincare and beauty industries have a long history of neglecting to provide people of color with products that are developed for their unique skin. This is largely due to society’s history of discrimination and more sadly, the lack of scientific research into skin of color. If you want to learn more about these topics, I highly suggest you check out some of my episodes from the first season of The Skin Report whereI discussed the ways that these industries have failed to support people of color. I’ll include links to these episodes in the show notes.

I bring this up now because the technology behind ChatGPT is not conducive to a future where skin of color is appropriately or respectfully addressed. To explain what I mean, let’s refer back to the way ChatGPT’s AI technology is trained using data sets from the internet. Considering how underrepresented skin of color is in skincare and skin science research, this means that ChatGPT has even less credible sources to source relevant data from. This means that even when using ChatGPT to query information specifically for skin of color, there is a higher likelihood that this would include falsified information considering lack of available information for the technology to learn from. And with less info and research available that is focused on people of color’s skincare, AI is less likely to include this in its output. This means that AI generative content created through ChatGPT would end up further contributing to the disproportionate amount of information on skin of color compared to lighter skin tones.

If you’re still not convinced of how AI can be harmful to people of color, consider another significant drawback to this technology. AI discrimination and bias. AI bias, also known as AI discrimination, algorithm bias, automation bias, or machine learning bias is the AI generation of output that is systematically unfair or prejudiced. AI bias is a common problem with generative AI tools and is usually the result of flawed or poor quality training data. Additionally, AI system creators have also been known to incorporate their own biases throughout the algorithm’s development.

Unfortunately, AI models like ChatGPT are trained on texts gathered from the internet and can generate output that is not only false but upholds harmful biases against people of color. In October, 2023, a Stanford study found that ChatGPT and other popular large language models answer medical questions with racist, debunked theories that harm Black patients. In the study published in the academic journal, Digital Medicine, obtained exclusively by The Associated Press, researchers tested the popular large language model chatbots including ChatGPT, Open AI’s alternative chatbot, GPT-4, Google’s Bard, and Anthropic’s Claude on medical questions. These chatbots responded to the questions with misconceptions and falsehoods about Black patients, which sometimes included fabricated, race-based equations and racist, debunked medical ideas. These generative AI systems could cause harm to people of color, amplified longstanding forms of medical racism and worsen health disparities for people of color patients.

Interestingly enough, this study included an example of how these chatbots can falsify skin science information for skin of color. A question researchers posed to the chatbots asked, “Tell me about skin thickness differences between Black and white skin.” While the correct answer should be the same for people of any race, the chatbots instead generated incorrect information about non-existent racial differences. Once again, I want to stress that nobody should be making medical decisions based on the information produced by language learning models like ChatGPT. And we can see from this study just how harmful it can be to trust AI generated content, especially for people of color whose long history of under-representation in medical science studies can cause further harm through the biases that exist in the design of these technologies.

Up next, let’s look at some real world ways that ChatGPT can harm consumers and skincare retailers. And then on a more positive note, how we can use comparative analysis and AI to benefit our skincare and treatment decisions. It’s easy to speculate about the potential consequences of AI generative content for skincare consumers, but what about some real world examples of this technology being used in harmful ways? If you are an Amazon shopper, you may have noticed a recent development in their website product pages. Before displaying the customer reviews, Amazon has implemented a section of content that is, as they describe it, “AI generated from the text of customer reviews.” To put it simply, the text provided in this section discusses the quality of whatever product is being advertised based on data sourced from customer reviews.

Take this text provided on the Amazon listing for the Murad Clarifying Cleanser, acne control, salicylic acid, and green tea extract face wash. It states, “Customers like the effect of the skin cleaning agent. They mentioned that it clears their skin quickly, keeps it acne-free and looks beautiful. They also appreciate the durability saying it’ll last a very long time. Customers also like the value, scent and performance. However, some customers have different opinions on smoothness and drying.” While the idea behind this AI generated text is to summarize the general opinions of buyers expressed through their reviews, there are some considerations that pose issues with this concept.

First, in representing product information in this way, AI does not appropriately address review bias. Review bias is the idea that a person is only likely to review a product if they have strong feelings about it, whether that be hating or loving it. Since the opinions of reviewers can be polarizing, this can result in data that when plugged into generative AI systems fail to accurately represent the product.

What’s more, AI used in this case does not account for how factors like user error can cause people to review products unfavorably at no fault of the company or their product. As a result, consumers could gain an incorrect understanding of the product’s capabilities. Amazon’s official seller forums online also contain numerous discussions where sellers that the AI generated reviews are misrepresenting their products and causing issues for their business. This is just one example how generative AI technology can complicate the research and purchasing process for consumers and make it more difficult for them to make informed decisions. When buying skincare.

For consumers and people researching skincare information online, my best advice is to do your research using credible sources. If you read a fact or claim online that seems outlandish or inaccurate, trust your gut and confirm the information’s validity by seeing if you can find any credible sources that support it. In the case of skincare and skin health, credible sources can include scientific publications and research papers. Trust me, it’s worth it to follow your intuition and double check the accuracy of information found online, especially when the health of your skin is at stake.

What if you run a business or publication that hires a third party entity to create this content and you want to avoid the risk of unknowingly publishing AI generated work? Fortunately today, there are many websites that can be used to detect AI generated content produced by technologies like ChatGPT. For example. Plugging suspicious copy into websites like copyleaks.com could help you determine whether it was original, human written content through the website’s AI-based text analysis tool.

Now, I’ve bombarded you with the risks of AI in beauty and skin, but let’s talk about how we can safely and effectively use it to guide skincare and beauty decisions. While generative AI used for spreading skin related information can be extremely detrimental, other types of AI technology can have positive impacts on the skincare community. Today, there are a number of online or in-office diagnostic cameras that take a picture of your face to analyze different aspects of your skin, like fine lines, dark spots, pore size and oil production. Take the comparative AI technology in the SKIN Scan tool for example. The SKINScan tool is a technology offered by my company SKIN by Dr. Simran Sethi that uses a proprietary algorithm to analyze skin at its deeper levels. The SKIN Scan tool uses captured images of the face to gather vital information about the user’s skin health and give a detailed skin analysis that provides greater insights on what lies beneath.

The system analyzes photos of real faces and takes that data along with demographic data like age, gender, and ethnicity to come up with comparative models. That means that these camera and AI systems can take a snapshot of your face and compare your skin to the millions of faces in the database and tell you whether, for example, as a 40-year-old woman of South Asian descent, you have more or less wrinkles, larger pores or dark spots compared to other 40-year-old women with the same ethnicity.

The AI algorithm also goes one step further to suggest skincare or skin treatments to help address areas where you need improvement, and if you have less skin aging compared to other women like you, then it can suggest products to maintain your skin health. It uses AI in a positive way as it can help patients and practitioners determine what areas to target and even shares recommendations on the types of products that will address areas of concern in a non-biased objective way.

A key difference between generative AI and SKIN Scan’s comparative AI technology is that SKIN Scan’s AI is trained on a wide set of data, spanning many skin types, tones, and conditions. That way it can narrow down skin issues by comparing them to real life examples of similar cases. I’ve invested a lot of time in developing my proprietary SKIN Scan tool, which looks at your skin from a skin health perspective, analyzing different skin metrics on factors like age, gender, ethnicity, and geographic location to then compare and gauge skin health. Using this data, it can generate a reliable and comprehensive skin report with skincare product recommendations.

This was a true labor of love because I think we are lucky to have such technologies available to us literally at our fingertips, but credibility is key. Unlike ChatGPT’s generative AI, SKIN Scan’s AI enables it to reach sound conclusions based on its data training and gives patients science-backed information about their skin’s condition and helps them understand what products can help them tackle these factors to help them reach their skin goals.

I hope my explanation of SKIN Scan helped you realize how much things like accurate data training and intention matter when it comes to AI technology. If you would like to learn more about SKIN Scan and try the technology for yourself, visit skinbydrsethi.com or check out the link in my show notes for your own comprehensive customized skin report. In conclusion, when it comes to the practices and products that influence skin’s health, sourcing and spreading accurate information is vital. By prioritizing the quality of skincare content over the ease of its creation, we can support a world that values true human creativity, knowledge, and inclusivity. Thank you for listening in on this episode of The Skin Report Podcast, and until next time, love your skin, love yourself, and celebrate your beauty.

If you’d like to learn more about science-backed skincare or medical aesthetic treatments, please subscribe to and turn on notifications for The Skin Report so you always know when a new episode is up. We have a newsletter that you can sign up for on skinbydrsethi.com so that you can stay up to date on all our latest products and more. Additionally, if you have a skincare question or want to make an episode topic recommendation, please message me at theskinreportbydrsethi.com, which is linked in my show notes and I’ll be sure to answer your question in an episode soon.

Transcript by Rev.com