University of Colorado Boulder Professor Michael Klymkowsky created two AI bots to help students learn: Dewey and Rita.
In his classroom at CU Boulder, Dewey and Rita can act like tutors, analyze answers and even provide feedback on how Klymkowsky can improve how he teaches material.
“It seems likely that (AI) policing schemes are ultimately going to be futile,” Klymkowsky said. “What we would like to do is incentivize students to use AI tools to enhance learning and to achieve the grades and goals they want and need.”
Artificial intelligence, or AI, is a technology that allows machines and computers to perform complex tasks and mimic human intelligence and behavior. CU Boulder held a virtual showcase of how the university uses AI on Wednesday.
Klymkowsky, a professor of molecular cellular and developmental biology, created Dewey and Rita using a type of AI called retrieval-augmented generation.
This type of AI is more accurate, Klymkowsky said, because it’s provided with curriculum information, textbooks and research papers that create a knowledge base. The AI draws from this base of information and knows what it can’t answer. So, it won’t speculate or hallucinate. AI hallucinations are when a bot generates incorrect or misleading information.
“This has the advantage that it won’t answer a question it hasn’t been trained on, where a standard ChatGPT will answer any question,” he said.
Dewey can evaluate student answers to questions and identify trends of what students are missing or misunderstanding. It can also determine whether course learning goals are being met.
“We can get a summary of the ideas students are struggling with and we can even get suggestions for instructional improvement,” Klymkowsky said.
Rita is an AI bot that acts as a Socratic tutor. Socratic teaching or tutoring is a strategy to promote critical thinking by giving students questions instead of answers.
Rita is a more challenging bot, Klymkowsky said, because she has to be engaging for students. He said she can usually determine whether a question is answered correctly and pose questions to students about what’s missing or wrong in their answers. She’s improving at engaging in Socratic conversations with students.
“It’s not meant to be a version of Google, it’s meant to be a tutor provoking a student to engage with the material,” he said.
The bots tend to be extremely accurate. Klymkowsky said they are no-code bots and creating one can be done in half an hour if the materials for its knowledge base are ready to go.
AI is also being used to analyze large data sets on the research side of CU Boulder. Computer Science Professor Esther Rolf explained how her team uses geospatial machine learning, a type of AI that analyzes geographical data.
“Connecting those sources of data to make some sort of actionable insight remains really challenging,” Rolf said.
One example she provided is detecting artisanal mining, a type of mining for underground minerals, and its connection to health outcomes.
“Are these small metal mining operations happening near streams, diluting streams with different sediment and different chemicals, and is that impacting people living downstream from these mining operations?” Rolf said.
While there’s a lot of data available, there are no maps of where artisanal mining is happening across the world.
“And so how are we possibly going to connect that with people’s health outcomes?” Rolf said.
To try and solve this problem, she and her team created a public website that takes satellite images and AI machine learning and transforms it into structural information that can be used. The goal is for the site, mosaiks.org, to be fast, easy and accessible.
Their work has been recognized by the United Nations.
“It’s an exciting model because .. these models are actually used to generate insights that help advance science and understanding of our world,” Rolf said.
On the administrative side, CU Boulder is starting to incorporate Microsoft’s AI platform, Copilot. CU Boulder Office of Information Technology staff member Tim Crean explained how Microsoft is implementing AI and how users on campus might use it.
“This version of Copilot can actually answer questions related to meetings, chats, emails and documents,” Crean said, adding, “Copilot can only access data that you have access to in Microsoft 365.”
Copilot can organize data sets in Microsoft Excel, for example. If someone joins a meeting on Microsoft Teams 15 minutes late, they can ask Copilot to recap what happened.
“It will go through the transcript and tell you what’s been said so far,” Crean said.
The Office of Information Technology completed a six-month pilot program for Copilot where 30 users participated and provided feedback. The office collected data about where and how often it was used to help employees get their work done. Now, the license request process is being developed.
Crean addressed concerns about security, privacy and data protection. More information on security is available at oit.colorado.edu/services/it-security/artificial-intelligence-ai-data-security-guidelines.
The virtual showcase was hosted by CU Boulder’s AI Community of Practice, a group of staff and faculty that have an interest or expertise in AI. The goal of the AI Community of Practice is to identify campus resources and initiatives, contribute to AI best practices and establish campus guidelines and policies.
The campus has also formed an AI steering committee that will help with resources and guidance on AI. The AI Literacy Ambassadors program is a cohort of instructors who encourage and support AI innovation.
For more information, visit colorado.edu/information-technology/ai-cu-boulder/ai-efforts-campus.
Get more business news by signing up for our Economy Now newsletter.
