Let’s talk about sex: AI-driven chatbot addresses taboo subjects
Kandyce Brennan, DNP
When Kandyce Brennan, DNP, was training as a nurse midwifefirst at Duke University Hospitals and then at the Durham County Department of Public Healthshe saw firsthand the profound disparities in reproductive health care that still exist in the U.S. today.
Access is a big issue, particularly for those who must contend with transportation barriers, financial constraints, or a lack of nearby clinics. Women of color often face bias or discrimination, leading to less attention to their specific needs and wants when it comes to contraception. And rampant misinformation can affect how well women understand their contraceptive options, their side effects, and how to use them correctly.
For Brennan, who is now a clinical assistant professor of nursing at UNC, the need for better contraception counseling was all too clear. The question was how to provide that counseling in a more equitable way.
"When OpenAI became a thing, it's just like a light bulb went off for me," said Brennan. "I knew this would be a really great toolif we could make it reliable and accurateto get sexual and reproductive health information to people in rural and underserved communities where we know we have provider and care deserts."
More than 80% of U.S. counties lack adequate access to health care. As a result, 19 million women live in contraception "deserts," meaning they do not have reasonable access to a health center that offers the full range of contraceptive methods. A projected shortage of more than 100,000 primary care physicians by 2034 will only make things worse.
Many experts, including Brennan, believe that technology can help solve the problem. With support from NC TraCS, Brennan has been developing an AI-driven tool known as the Sexual And Reproductive Health Assistant or SARHAchat (pronounced "Sara-Chat").
SARHAchat is a next-generation chatbot designed to encourage candid conversations about reproduction and safe sex, informed by the latest scientific evidence.
Tianlong Chen, PhD
"When we were building the chatbot, I really wanted it to feel like you were talking with your best friend's older sister, someone with life experience who you could trust," said Brennan. She collaborated with Tianlong Chen, PhD, an assistant professor of computer science, to tailor the tone or "temperature" of SARHAchat to be warm and empathetic.
"There is a fine balance, though," she said. In the beginning, the chatbot was a little too effusive, framing every response with emojis and thank-you's. So they dialed it back.
The researchers also restricted which sources of information the digital assistant could pull its responses from (CDC recommendations and Google) and which were off limits (Reddit and other social media). Brennan and her team created an extensive reference document with evidence-based information on various forms of birth control that SARHAchat could check the accuracy of its responses against. They designed prompts that follow the stages that Brennan typically goes through with a client around family planning and trained the model using examples from real user conversations.
...if you're in a rural area and you don't have access to ask someone a very personal question, you ask the internet.
SARHAchat uses natural language processing and machine learning to anticipate what users are going to ask as they are typing and to respond quickly, even when the conversation goes a little off track. That use of AI makes responses feel less robotic and more tailored to individual needs than traditional chatbots.
"We've had people ask questions that you could tell they wouldn't necessarily feel comfortable asking a provider," said Brennan. "And I think part of that, too, is if you're in a rural area and you don't have access to ask someone a very personal question, you ask the internet."
Brennan said she has been happy to see people seeking not just information but also advice on how to navigate difficult conversations, such as how to talk to their partner about using protection or getting tested for sexually transmitted infections.
But while the team is eager to foster these conversations, it has set some boundaries.
"If someone starts talking about politics or the election with our chatbot, it will remind them that it is here to help them with sexual and reproductive health, or to answer questions about contraception, but that it can't talk about politics because it's not an expert in that area," said Brennan. "That keeps it from getting volatile or dipping into dangerous territory." It also has safety features built in to guide users to seek immediate help if the user provides concerning information like thoughts of self-harm or unsafe situations like intimate partner violence or abuse.
Brennan presented the latest version of SARHAchat at the FemTech Demo Day, hosted by FastTraCS, the medtech incubator at the NC TraCS Institute. FastTraCS also facilitated an advisory session with the FemTech Advocacy Council, where Brennan and her team received critical feedback, encouragement, and resources to translate her idea. "Connecting with other people who have done this kind of work has really helped me to refine my approach," said Brennan.
To further support strategic planning, FastTraCS organized a Business Model Canvas session to explore the commercialization and long-term sustainability of SARHAchat.
Bianca Allison, MD, MPH
"They really helped me navigate what path I want to go down," she said. "Do I want to go down the commercial route or do I want to go down the research route? What are the next steps? It's really helped me lift my head up and look forward, rather than just keeping my head down focused on the research."
Right now, Brennan is working with Bianca Allison, MD, MPH, a health services researcher and assistant professor of pediatrics, to expand SARHAchat to adolescent populations. Allison, who is a member of the FemTech Advisory Council, has talked with youth to find out what gaps exist in their knowledge of sex and contraception so they can add information to the chatbot to address those questions. Brennan and Chen are also pursuing ways to make SARHAchat more user-friendly for rural and underserved populations where internet access may be less stable, such as by creating a "broadband light" version of the chatbot.
"My long-term goal, though, is to use this tool to counter the concerning amount of misinformation and disinformation that's out there," Brennan said. She admits that it is a challenging task, as research shows that engaging with this content can draw more attention to it and make inaccuracies and falsehoods stick in people's minds. "We need to flood these spaces with accurate information and encourage people to engage with it instead."
This project was also supported by a seed grant from the UNC School of Data Science & Society.
NC TraCS is the integrated hub of the NIH Clinical and Translational Science Awards (CTSA) Program at the University of North Carolina at Chapel Hill that combines the research strengths, resources, and opportunities of the UNC-Chapel Hill campus with partner institutions North Carolina State University in Raleigh and North Carolina Agricultural and Technical State University in Greensboro.