Meet Sam, an AI-powered bot for health sciences

Arizona State University (ASU) graduate students enrolled this summer in IBC 633: Motivational Interviewing for the Behavioral Care Provider experienced a first for the course.

Shortly into the semester, students were introduced to their patient Sam, who is described as a middle-aged person with a long history of smoking. Sam has tried to quit several times but finds it immensely difficult. 

Sam is a chatbot that uses generative artificial intelligence (AI) — the first of its kind for the course.

The course teaches motivational interviewing, an evidence-based approach that is used to support patients towards behavior change. Clinical Professor Colleen Cordes in the College of Health Solutions offers the course as an elective as part of ASU’s Doctor of Behavioral Health program. 

“We started to build Sam in the spring semester when we earned licenses to ChatGPT Enterprise,” shared Cordes

Two assignments in the course involve students practicing their motivational interviewing skills through a simulated patient experience. 

Generative AI as a tool to pave the way for innovation 

In January 2024, ASU announced its collaboration with OpenAI, the AI research and deployment company behind ChatGPT. Soon after, the university launched the AI Innovation Challenge, which invited faculty and staff to submit proposal ideas to leverage the AI tool across academic, research and work environments. 

Sam is a result of the Challenge. The AI-powered chatbot simulates a patient-provider interaction for students, who take on the role of the provider to support a patient through a behavior change. In Sam’s case, he hopes to quit smoking. 

“Simulation is an important component of the course for students to practice their motivational interviewing skills,” Cordes continued. “When generative AI came on the scene, we were immediately interested in what new pathways this technology could offer our students.”

A key component of motivational interviewing is practice. 

Prior to Sam, the simulation practice for students was built around a decision tree that uses branching logic — select answer A and you move through a specific sequence; select B and you move through another sequence. While effective, this type of assignment is capped at the number of sequences available — in this case, three.

Using generative AI to drive the sequence resulted in many more pathways for discussion —- where a single word change could result in a completely different discussion outcome, much like real life. Cordes noted that this was reflected across the students’ discussion transcripts. 

As noted, Sam is a custom GPT. This means a set of instructions provided to the bot offers directions for how the bot interacts. Take, for example, that part of the instructions developed for Sam is that he has family support. Students who honed in on this positive change aspect resulted in discussions that drove towards: you want to be healthy so you can spend time with your family. This positive change aspect provides a different depth of desire and motivation for Sam to quit smoking, versus those who might have honed in on other aspects, like saving money, as their motivation. 

“Students who picked up on different things resulted in very different outcomes,” Cordes said. 

The generative AI chatbot could also provide feedback immediately following an interaction, offering students insights into what they did well and areas for improvement. 

Cordes added that the AI’s feedback was not ready to be used for grading; however, it did provide helpful insights. “The bot, while not perfect, allows for students to authentically practice their motivational skills, even in the absence of an instructor,” Cordes continued. 

Cordes also noted that the transcript of the interaction provided from ChatGPT Enterprise was also “immensely easier” for her to use to provide qualitative feedback as part of her grading. “If I wanted to cue in on something that the student said and how they could have said it differently so that it was more aligned with [motivational interviewing], I could cut and paste from the transcript.”

Cordes went on to describe what this feedback might look like. Building upon the example above of the positive change aspect, she might provide feedback to a student that goes like this: 

When providing your reflection to Sam, the final part of your sentence stated: “you want to change for your family but it’s ‘really hard’.” By ending the statement with “it’s hard”, you are invoking sustain talk, in which the statement supports maintaining a health-risk behavior, in this case smoking. However, if you alter it to “it's hard but you’re not giving up because it’s important to your family” that ends using positive change talk to offer greater motivation to change.

This is one of many examples where Cordes would use the transcript provided by generative AI to dive into specific feedback. 

Related: ASU students debate with famous philosophers using generative AI

Designing and introducing Sam to students

The chatbot was designed and built in spring 2024, with input from a variety of interprofessionals that included Principal Instructional Designer Tamara Mitchell and Learning Technology Specialist Wesley Fleming from EdPlus.

“We took a human-centered approach when designing the bot,” shared Mitchell.

Mitchell noted things like equity — ensuring all students had access to ChatGPT Enterprise licenses — and prior testing by students and faculty were key to their success. 

“We also worked a lot on defining trust in AI intelligently,” Mitchell noted. Students are taught to be critical of the responses provided by generative AI.

The summer course, offered through ASU Online, had 50 students enrolled, most of whom noted a positive experience working with Sam, the AI-powered chatbot. 

“Survey results from the students showcased 94% agreed or strongly agreed the chatbot felt human and 89% said that the chatbot strengthened their learning experience,” Cordes concluded.  

Cordes, who has since taken on a new role as Assistant Vice Provost for Career-Track Faculty, teaches the course every other year. However, she notes that the success of the chatbot has applications far beyond her course: “There are other courses in our program that use simulation as part of the curriculum, and this chatbot could be adjusted to work inside those courses.” 

ASU is gaining global recognition for its commitment to chart the future of generative AI in higher education. And with over 200 projects activated that explore the use of ChatGPT Edu through the AI Innovation Challenge, the university is sure to see more examples of generative AI to support academic, research and work environments now and continuing into the near future.