The Promise and Peril of ChatGPT in Healthcare and Pharma Marketing

Man’s discovery of fire. Edison’s invention of the light bulb. OpenAI’s release of ChatGPT. Is one of these not like the others or did we just witness the introduction of something that will fundamentally change how humans operate? John Nosta, Founder of the digital health think tank NostaLab, very much believes it is the latter.

Since its release on November 30, 2022, this AI chatbot, which uses a Generative Pretrained Transformer (GPT) language model and deep learning techniques to deliver human-like responses in a conversational manner, has generated buzz across all corners of the internet and within every industry.

“This is a technological inflection point we didn’t see coming,” Nosta explains. “We thought the way COVID forced digital acceleration would be a game changer. But it came down to something that is a fundamental tool for everyone around the planet—search. ChatGPT is achieving functional, practical advances in search that are widely available with broad consumer utility. This advancement will be a pivotal point in the technological revolution.”

Perhaps you have tried ChatGPT—or heard from others who have—and came away impressed or dismissive. It can be easy to understand either opinion, as this generative AI can pull together content in seconds from any query you provide it, but it has also been known to generate false information or even make up references.

“ChatGPT is a Large Language Model (LLM), which means it does very well with predicting the next best word, phrase, and sentence, but it may not always be factual,” explains Abid Rahman, Vice President, Innovation at EVERSANA. “But we are only just scratching the surface of what AI can do for us. Advanced AI models and algorithms such as GPT4 will enhance ChatGPT’s capabilities and make it more valuable. So while we shouldn’t rely only on ChatGPT to provide information where accuracy is critical, such as medical information, these issues may be fixed in the next version of the technology.”

That is one reason why Nosta is quick to dismiss the naysayers of ChatGPT who point to its current flaws.

“It’s a mistake to define ChatGPT by its current iteration; we have to look at this technology as not a point in time, but a trajectory,” Nosta says. “And when you look at it that way, it’s almost inconceivable not to see the tremendous transformative power that is emerging today. In fact, with all technological innovations there is the duality of wonder and fear. Consider fire, which was our first technology and remains the leading cause of property damage in the U.S. We have to put it into that kind of a perspective.”

Martin Samples, Head of Digital, Senior Vice President at Precision Value and Health, offers one perspective through which to view ChatGPT—a prototype that is simply proving what this technology is capable of.

“What ChatGPT has done is show that a large language model, trained on vast swaths of data on the internet can accomplish complex tasks and respond to multifaceted queries using generative AI,” Samples explains. “The makers of ChatGBT wanted to create interest in generative AI and give people both practical and creative ways to use this smart technology. By doing so they have millions of contributors testing and training the large language model daily. I have to believe that with the right datasets and the right training this AI engine will create value across a number of verticals including life sciences.”

So, given that this technology is likely to have in impact in healthcare, the life sciences, and marketing, what are some of its potential use cases in both the short- and long-term?

ChatGPT Uses in Healthcare and the Life Sciences

In a recent poll on Sermo’s physician platform, 34% of physicians report being excited about ChatGPT, 24% nervous, and 39% a little bit of both. However, 76% said they see the potential that ChatGPT could offer to physicians and other healthcare workers with 39% saying they would use it for research, diagnostics, and treatment; 27% open to using it for patent care and delivery; and another 27% willing to use it for clinical and non-clinical workflow.

“ChatGPT and AI are going to change the way we educate, learn, and care for patients,” says Dr. Zach Horne, Radiation Oncologist at Allegheny Health Network and Sermo Medical Advisory Board Member. “With AI assistance, the activity of searching for information whether to help a patient in the clinic, for a research project, or in general medical education, will be drastically different. Currently, the clinician or researcher is still left with all the hits of a query and has to sort through the information. With AI, all the information being sought should be produced front and center with a simple query. This could be a huge time saver. But the biggest caveat is the accuracy of the information delivered from queries. There will need to be some rigorous testing done before AI can be trusted to deliver accurate medical information.”

However, Dr. Ryan Chuang, Emergency Physician & Medical Toxicologist and Sermo Advisory Board Member, thinks we are still years away from this technology being used for medical purposes.

“I think the big issues of using AI in healthcare are validating outcomes to make sure it is safe to use AI to interact with patients and then the liability issue,” Dr. Chuang says. “Who is liable if something goes wrong—the person employing the AI, the AI manufacturer, the patient, or nobody? These are tricky issues to consider when thinking about using ChatGPT.”

Still, Samples can see generative AI being used as a digital front door of healthcare, serving as a virtual assistant that is there to collect information on the actual physician’s behalf.

“A virtual HCP would be able to answer a lot of questions without a real person initially, and then it can make a warm handoff to an HCP at the moment where the next step can only be done by a physician,” Samples offers. “Basically, it can serve as an adjunct helper to the doctor by gathering answers to questions about symptoms, lifestyle choices, and more while also providing educational information to the patient.”

ChatGPT, or similar technology, could also help to enhance the dialogue between real-life patients and doctors.

“We can bring this tool to support the physician-patient conversation, whether it’s guiding people to support options or breaking down high-science diagnoses to relatable terms and answering questions,” explains Brian Lefkowitz, Chief Creative Officer at Digitas Health. “These tools can help patients take better control of their care and support physicians to have more productive and personalized conversations. Generative AI tools will eventually help us design hyper-relevant experiences for people who don’t follow the ‘average’ patient journey.”

Even though ChatGPT may be new to the scene, Jim Lefevere, International Business Leader, Digital Partnering Solutions at Roche Diabetes Care, is quick to point out that healthcare has been using AI for years to analyze big data within digital therapeutics, diagnostic tools, and as an aid for clinical planning.

“There was a recent JAMA study that showed AI was 84% accurate in cardiovascular disease prevention advice and that will only improve,” Lefevere says. “I think the potential is endless and much like the early days of the internet many of the use cases have to be created.”

ChatGPT Uses in Marketing

For now, Alfred Whitehead, EVP at Klick Applied Sciences, says they have yet to see any pharma companies using ChatGPT to communicate directly with their patients. However, early adoption is happening in less regulated spaces, such as BuzzFeed which is planning on using GPT to assist in generating online quizzes. Even if marketers within the life sciences aren’t fully comfortable using ChatGPT externally, Whitehead says immediate benefits can be applied internally.

“What we do see happening is the use of generative AI technologies in the ‘behind-the-scenes’ work of marketing such as a ‘brainstorming buddy’ to help copywriters focus more on shaping messaging to match brands’ strategies than on the creation of raw material,” Whitehead explains. “Right now, this class of AI is capable of text creation, summarization, insight discovery from text, style transfer, question answering, and more. Other models dive into the generation of sounds, music, images, video, and transcription.”

The use of ChatGPT as a tool to aid creatives instead of a potential content-creating replacement is an important distinction to help ease the worries of anyone who fears this technology may ultimately cost people their jobs. As Lefevere says, “I am not in the camp that it is a doomsday job killer. In reality, every technological shift, which AI is, has taken years, if not a decade or more, to reach a mainstream state.”

Nosta adds that even though ChatGPT is shifting the cognitive heavy lifting to technology, that doesn’t suppress creativity—it elevates creativity. And he points to Norman Rockwell as an example.

“Rockwell did two things most people don’t know about when painting ‘Freedom from Want’ or what is better known as the Thanksgiving painting,” Nosta says. “Number one, he hired a set designer and a photographer to create that actual shot. Then that picture was traced onto a canvas using a Lucy device. In other words, Rockwell used technological augmentation tantamount to GPT to create his finished artistic work. And when asked about it, Rockwell said, ‘the Lucy is a horrible machine, and I’d be lost without it.’”

So, yes, generative AI may be used to help write social media posts, craft a marketing plan, improve customer engagement, but marketers will still have an important role to play.

“Writing, design, strategy, and analytics in healthcare are about building connections, and creating meaningful connections will always require a human touch to infuse emotion or lived experience,” Lefkowitz says. “Especially, when it comes to making content that fits into and impacts the lives of people so personally.”

With that said, this technology could help to deliver more personalized experiences to customers, which is a goal that marketers are constantly striving towards. According to Samples, you could develop predictive engines that will tell you what kind of content a particular user might be interested in and what action they might take based on that content.

“This type of hyper personalization could be used to enable a more effective educational experience by learning over time what specific questions a patient with a particular type of symptom or diagnosis might have,” Samples says. “As the generative AI better understands the intent of a question and the nuances of why a question is asked this tool could provide more helpful information.”

Beware the Potential Pitfalls

Besides any privacy issues that marketers always need to pay attention to, the use of generative AI also comes with other potential issues that marketers should know about.

“There is a Cambrian explosion of use cases right now, and many of the apps are harvesting the data they are fed,” Whitehead says. “Marketers and agencies need to be careful not to expose confidential information that may be used in training future versions of GPT. Directly using the outputs of these models in campaigns also carries some legal risk. There are intellectual property questions about AI-generated derivative works that are not settled in statute or case law yet.”

Life sciences and healthcare companies should also avoid inputting any clinical trial data or patient data into the free version of ChatGPT, says Matt Lewis, Global Chief Medical Analytics and Innovation Officer at Inizio Medical.

“ChatGPT will take all of this input data and use it to improve its performance, so if you ingest data on an ongoing clinical trial on breast cancer before those results have been analyzed, reported, and the like, the raw information will be in the cloud and could be either directly or indirectly shared with others in the world without your knowledge,” Lewis explains. “The newer paid model or the models incorporated into existing engines may prove safer, and safer yet are closed platform and SaaS applications where vetted, security-checked developers are integrating these large language models like GPT-3 and 3.5 as well as Google’s Sparrow into their technology stacks.”

Other issues to be aware of include the fact that many of these current AI tools have diversity, equity, and inclusion (DE&I) blind spots due to incomplete or biased datasets, explains Will Reese, Chief Innovation Officer at Evoke, an Inizio company. Additionally, given that ChatGPT can be prone to error, Chris Seminowicz, SVP, Data and Analytics at Digitas Health, advises marketers to wait until we see real oversight, transparency, and traceability back to original and trusted data sources. And Whitehead adds that marketers must consider how ChatGPT-like interfaces will impact search and product discovery.

“Right now, ChatGPT has knowledge up until the end of 2021,” Whitehead explains. “That means, if you launch a new product today, ChatGPT will have no knowledge of it until it’s retrained. We expect this sort of interface to become popular, so we’ll need to understand how that impacts product discovery—and to scare regulatory teams—how updates to labeling are communicated to AIs.”

The Future of Generative AI

Microsoft, which is a backer of OpenAI, has already announced it is incorporating ChatGPT technology into its Bing search engine and Edge browser. However, the company recently had to places limits on its new Bing AI about after a few unsettling interactions with users, but quickly restored longer chats just a few days later with the goal of making them more responsible. Meanwhile, Google was quick to announce the release of Bard, which is powered by its Language Model for Dialogue Applications (LaMDA) technology, to keep pace in the chatbot race. It also made a rocky debut when it responded with a factual error in its first demo. Despite that, Samples says not to count out Google.

“The big thing that separates Google from the rest of the world is that they have a lot of data,” he says. “They’re going to start with a little bit of an advantage just because of that volume of data and the fact they have a natural place to integrate generative AI given their search engine dominance.”

As these larger platforms grow, Rahman says to expect to see additional implementations on top of them.

“There are already many startups building products on top of these large generative AI platforms,” Rahman explains. “This is reminiscent of the early days of the internet and the dot com boom. Some of these smaller products/companies will likely become bigger and some will cease to exist, however the AI will continue to grow and evolve. It is important to note that Microsoft, Google, and other large technology companies are already planning on implementing additional AI services within all their products.”

As an example, Microsoft recently released BioGPT in collaboration with Peking University, which Whitehead says is something that should be on the radar of everyone in life sciences.

“It’s a model similar to GPT, which underpins ChatGPT, but it’s trained specially for use in biomedical research,” he explains. “It has set a new record for question and answering accuracy using questions curated from PubMed. This sort of model has plenty of potential, both for research and for accelerating some types of medical writing.”

To prepare for this future where generative AI becomes prevalent in our lives and marketing, Reese encourages organizations to begin leveraging AI tools internally to build organizational familiarity, including productivity tools such as Moveworks (enterprise services conversational AI), Timely (AI time tracking), Grammarly (AI writing support), or Fireflies (meeting note automation). Meanwhile, Seminowicz mentions keeping Adobe Sensei (real-time marketing intelligence) and Jasper AI (long-form writing) on your radar because they fit well into a marketer’s current process flow.

“A few years from now, we’ll likely look back on ChatGPT specifically as one moment of hype,” Whitehead says. “It’s one of several generative AI models out there and there’s no guarantee it will be the winner in future competition. If you think of the early internet, there were several waves of search engines (Yahoo, Webcrawler, Altavista, etc.) before Google finally achieved a dominant position. But it’s also becoming clear that these AI technologies are so fundamental that they will be part of our lives in ways that will become invisible.”