AI in Clinical Research: Opportunities, Limitations, and What Comes Next

Artificial intelligence (AI)—particularly generative AI that can summarize, analyze, and generate responses to inputs—is the most talked-about innovation in health care right now. But with it comes intense debate: What are the ethical responsibilities researchers have, especially those who work with patients? What are acceptable uses for AI, and when is the human mind irreplaceable? Where can it help make tasks easier without causing downstream issues?

AI will transform health care: it’s transforming the industry as you read this, in fact. “I can’t even predict AI’s capabilities by the time this article is published—it changes every day,” explains Jamie M. Robertson, PhD, MPH, senior research scientist at Brigham and Women’s Hospital and assistant professor of surgery at Harvard Medical School. 

Robertson and her colleagues are currently conducting research to explore the full capabilities of tools like generative AI and large language models on clinical trials. AI is here to stay, though, and experts are constantly developing better models. In other words, researchers will need to familiarize themselves with the improvements that AI can make in their work, but they also need to know the limitations. 

The Power of AI in Clinical Contexts 

Even if we exclude the ethically gray area of using AI to analyze data and even write papers, these tools can still help significantly with workflow. Robertson explains that clinical research, clinical practice, and clinical education are three potential areas for increased speed and efficiency—recruiting and training the workforce, for example, or replacing high-cost systems with tools that work more quickly and at a lower cost. AI can also be helpful in the process of iterating trials and research. 

“I think that AI can help speed up many of the processes that are tedious and challenging,” Robertson adds. “It can help us come up with code to do data analysis and even suggest scenarios. But it’s critical for people who are interacting with AI as part of clinical studies to be knowledgeable about the right and wrong applications, and in the correct context.”

The Limitations of AI on Clinical Trials 

With all of its manifold benefits, AI lacks some of the higher-level functioning at which humans excel—skills that often go under appreciated in clinical research, notes Robertson. That includes creativity and ideation: exploring the unknown and thinking up new solutions to existing problems. 

“Creativity is an important part of developing research questions, generating hypotheses, and coming up with novel methods of research. AI is not as good at this because it relies on information that it already has. It’s not able to generate some of these new ideas or take into account the personal experience that people have in their field,” she says.

Similarly, the ability to assess and analyze—to synthesize data or results—is a distinctly human trait. “AI is not searching your mind. It’s not necessarily pulling the things that you want. We know that sometimes generative AI makes up information. It hallucinates; it over-generalizes,” Robertson explains. 

Thus, instead of taking away effort and responsibility, AI forces users to engage in higher-level thinking. If it suggests a methodology, for example, a subject-matter expert needs to draw from personal understanding and knowledge to assess the pros and cons. The same goes for problem-solving—AI  doesn’t know why it has the wrong answer, so it requires human intervention to be corrected. 

The Future of AI in Clinical Settings 

With all of this in mind, it is even more important for researchers to develop their AI proficiency—to become familiar with how it operates and when it falls short of the required rigor. “We, as people, have to learn how to interact with AI,” says Robertson. “It forces us to be knowledgeable specialists who can take information and transform it.”

Robertson has seen how experts are significantly excited to develop a working knowledge of AI, and she notes that there’s great potential for it to make researchers’ lives easier. But she adds that the best AI won’t replace the most nuanced aspects of clinical work—most importantly, the interpersonal interactions with patients. 

“I go to my physician for the training, expertise, and pattern recognition that they’ve gained from years of practice to decide the right choice for me. That level of expertise applies across fields,” she adds. “Keeping people in trials long term requires relationship-building.”

AI is in its heyday. But much like the internet of things and extended reality before it, once the novelty is gone, industry experts will have a better understanding of the best ways to apply AI. Unlike some of those former innovations, though, “we will be using AI regularly, so the question is how it will be implemented. The answer is: always carefully, thoughtfully, and with the oversight of people who know what they’re doing,” says Robertson.