Virtually instantaneous help in writing essays, planning trips, making dinner reservations, creating equations, and much more is drawing everyone to ChatGPT. School of Medicine faculty are using its capabilities to improve the learning experiences of their students and assist with their own research.
ChatGPT is a chatbot developed by Open AI that uses natural language processing (NLP) technology to answer questions in a conversational tone.
The program has exploded in popularity in the last year. “It’s really just one more advance in a long series of advances in AI, but this advance happened almost overnight, and gave much more impressive results,” said David Page, Ph.D., chair of the Department of Biostatistics and Bioinformatics. He’s been studying AI and NLP for 35 years.
“It’s rare to see such a big jump in improvement in the field,” Page said. The improvements seen with ChatGPT are the result of upgrades in the transformer architecture. “It allows researchers to address different natural language problems the same way. It is also now able to predict the next word of a sentence or phrase without losing the meaning or context behind it,” Page continued.
Many people have concerns about the potential for misuse, but Duke faculty find ChatGPT’s energy-saving capabilities to be its best asset. “It can help you overcome the initial hesitation to get started on a new project,” said Matthew Engelhard, MD, Ph.D., assistant professor in the Duke Department of Biostatistics and Bioinformatics.
Users enter the information they’re looking for in the textbox, and ChatGPT instantly generates a response. “Let's say you’ve outlined a new model or approach you want to develop,” Engelhard said. “You can have a dialogue with these systems and say, ‘I’d like you to write code that has these characteristics, and these objectives.’ The result might not be exactly what you need, but it’s often a helpful starting point.”
Page found this to be true for him as well. He’s exploring using ChatGPT to help with scientific papers and grant proposals. “For someone who's an expert and just wants to save the time writing, it's very good because it can give you a great first draft and you can correct it and maybe go back and forth with it several times,” Page said.
David Carlson, Ph.D., assistant professor in Duke’s Department of Civil and Environmental Engineering and the Department of Bioinformatics and Biostatistics, has his software engineering students use it the same way. Carlson said the instantaneous feedback and constructive criticism that ChatGPT provides has facilitated student growth.
“It’s been transformative in how much better the quality of their work is,” Carlson said. “It's not so much that it's doing the work for the student, it's that it's working with the student to help them come up and improve their results or improve their scientific documents.”
Carlson said the system’s timely responses are one of the main educational benefits. “If you get feedback quickly, you learn a lot better,” he said. “Before, a student would write something and send it to me. But sometimes it could be days before somebody has time to look at it. With the instantaneous feedback, it allows you to really be in the mindset of writing and improve it rapidly.”
He’s gotten mixed reactions from students. Carlson said some have been thoroughly enamored with the system, while others have been more hesitant.
But none of them are ignoring the obvious dangers of misusing ChatGPT. To quell concerns about cheating, Carlson is adjusting the way he gives homework. He plans to increase the number of coding problems, but also add conceptual questions about the results.
Engelhard teaches graduate students and plans to use ChatGPT in his fall classes. He believes their drive and internal motivation to learn the material will sway them away from using the system to try to cut corners on assignments. But its use will need to be monitored, and students must have a foundational understanding of the subject matter.
“It's not so different from using a calculator in math class,” he said. “There are cases where you need to put ChatGPT to the side to make sure that you are understanding the relevant concepts yourself, as opposed to using it as a crutch.”
However, more broadly, concerns about ChatGPT’s accuracy are growing. Page said ChatGPT doesn’t yet fact-check itself and is still very unreliable when used as a search engine. “It will sound very authoritative and give you an answer, and it may even cite references and say, ‘Here's where I'm getting this information.’ But the references may be all made up. It's just trying to look realistic and not really be realistic,” Page explained.
While many have expressed their concerns about how ChatGPT and other systems will influence society, many are still looking forward to seeing what’s next. “The systems are far from flawless, but I'm also extremely excited about the opportunities and how we can improve student outcomes by taking advantage of these technologies,” Carlson said.