Chatbots Go to Work on the IT Help Desk at George Washington University

User Profiles and Massive Query Histories Guide AI Assistants

MARTHA’s engine is the BMC Helix Digital Workplace platform, which leverages IBM Watson Assistant conversational interface functionality — the brains behind the system, Fozard says. 

GWU launched the chatbot with some of the most common requests. When a student keys a question into the system, MARTHA uses cognitive automation capabilities to reach out to a database, find the answer and reply. The authentication-based system identifies users after they sign in, drawing insights from their user profiles.

“We’ve created a service broker that can handle decisions on where to go to look for information,” Fozard says. “As we educated it and users tested it, the Watson component was learning alongside of us. If someone types in a question about 3D printing, we know that’s most likely a student who has access to 3D printers in the engineering classroom or a medical enterprise.”

The tricky part is getting the details right. While MARTHA can distinguish between faculty and staff users, it still needs to know, for example, the location of a door that’s not working, since a person could work in one building but encounter a problem in another.

Data is key to developing chatbots that can successfully perform the jobs for which they’re designed. 

At the Georgia Institute of Technology’s School of Interactive Computing, professor Ashok Goel may have been among the first instructors to put a virtual teaching assistant to use, back in 2016.

He and a team of graduate students created the first generation of the Jill Watson assistant using IBM’s Watson technology (they have since moved to using proprietary and open-source software). 

To give Jill the tools she needed to respond to students’ queries, Goel’s team trained the system using a database of more than 40,000 questions.

“That databank was critical for building the first version of Jill Watson,” Goel says, “because you could train the engine to answer questions, including new questions that it had not previously seen.”

Nova Southeastern University in Florida has provided a chatbot, Julie, since 2018. 

When a student asks a question about a course, facility or other administrative topic, application programing interfaces direct the system to the appropriate database for an answer.

Director of Software and Integration Eduardo Scheinkman notes that “student records, building locations — all that information is available in different systems. The chatbot is integrated with those systems and will pull information as it’s used.”

NSU created Julie with Microsoft’s Azure Cognitive Services (APIs, software development kits and services to help developers build intelligent applications) and Bot Framework solution.

“We wanted to be able to customize and brand the experience to feel like a cohesive part of the other products we have in place for students,” says James Drew, NSU’s director of innovation and information architecture.

MORE FROM EDTECH: See how universities take advantage of AI to boost graduation rates.

Data-Driven Audits Provide Insights to Students, Advisers and Staff

In Pittsburgh, a small group of Carnegie Mellon University students created a multipurpose system that’s now in use to improve CMU’s degree audit and course planning functions.

The cloud-based Stellic Degree Audit Application solution uses AI to deliver data-driven, customized support that helps students and advisers monitor academic progress. 

For example, students can identify interests in the system so AI can suggest related courses within their degree requirements.

Moving to Stellic in early 2018 also meant the university no longer had to store related data on campus. Advisers can access the application anywhere.

Previously, academic audit requirements had be programmed in separate files, and the system could generate an audit for only one program at time.

“With the former tool, users spent a considerable amount of time programming the application,” says Jamie Brandon, senior systems analyst in CMU’s registrar’s office. “Some advisers opted to audit degree programs manually, using spreadsheets and checklists. Advisers can now focus their time on advising students.”

IT Staff Reviews Help to Refine Help Desk Chatbot Performance

Chatbots do require fine-tuning. To improve MARTHA’s ability to provide help desk support, IT staffers review how the system processes terms and requests.

GWU’s system will flag terms, says Fozard, particularly if it determines that a certain percentage of references are leaving the chatbot stumped.

“Keywords are where AI struggles sometimes,” he says, noting that one word can have very different meanings depending on context. “It’s important to monitor what’s happening and how AI is making decisions so you can refine how it’s learning keywords over time.”

While chatbots aren’t a fit for every scenario, early adopters see benefits.

Goel saw engagement increase when students had Jill Watson as a resource. GWU’s chatbot has significantly freed up IT staffers, according to Fozard, so they have more time for in-person work.

“Sometimes you need to literally get a cable, or if there’s a keyboard issue, you need to have a physical presence to fix it,” he says. “We’re seeing a lot of those quick questions get taken off our plates, so teams can do more of the hands-on training and support aspects that AI simply can’t do.”