Higher Education's Increasingly Nuanced, AI-Powered Chatbots

Years ago, any questions students had about issues such as enrollment, academics or housing could be asked and answered only during designated hours, whether in person or by phone.

Today, however, a number of schools are turning to artificial intelligence–powered chatbots that crank out detailed responses 24 hours a day, using data pulled from a knowledge base that acts as a central repository of information, says Patricia Velazquez, director of education and research industry strategy at Oracle, which offers a digital assistant that is used in higher education environments.

“COVID definitely highlighted the need for digital assistants,” Velazquez says. “Phone lines were overloaded in the beginning of the pandemic. Users were unsure where they’d get information. A digital assistant really helps in an instance like this when it can provide accurate and to up-to-date information across the institution by pulling it from a single source.”

DIVE DEEPER: As colleges adopt AI, they’ll need ethical frameworks.

Can a Chatbot Have Personality?

In recent years, as chatbot technology advanced, a growing number of institutions moved from offering a single-solution chatbot — with a generally reactive, user-initiated format — to something more complex. The new iterations can plug into additional systems and support multiple streams of information across an entire enterprise, Velazquez says.

“Institutions have felt the need to really help students through their tenure at the university, and students today are digital natives,” she explains. “They expect 24/7 support and immediate answers, whether they’re on or off campus. Chatbot
technology is one way that institutions can provide that type of experience.”

Enhanced communication capabilities can help tailor the content and tone of chatbot responses. “Leveraging AI, machine learning and natural language processing, you can have a digital assistant that’s much more robust, multipurpose and proactive and that provides predictive engagement based on a user’s profile and role,” Velazquez says.

“You can also just process more natural and complex sentences,” she continues. “If students ask a question that has a slang term, the system can provide answers, even if the algorithm wasn’t trained specifically for how that question was configured.”

The University of Oklahoma’s SoonerBot supplies responses that have been custom-coded to imbue character, says Bryce Kunkel, assistant director of enrollment management communications in the Office of Admissions and Recruitment. If, for instance, a student types, “I love you,” SoonerBot will jokingly respond with, “I don’t know that computers are meant for love. Maybe that’s more of a human thing.”

“We wanted to give it a personality,” Kunkel says. “The bot has given us insight into what we should have on the site or could take off. If no one is asking about a topic, maybe that’s not good information to have on the website anymore.”

MORE ON EDTECH: In AI, colleges see cost-effective ways to boost enrollment.

Creating Responsive Bots With an Extensive Knowledge Base

When creating a bot, some schools opt for a vendor-provided natural language understanding platform, such as Google’s Dialogflow, IBM Watson or Amazon Lex. This technology attempts to decipher the intent of a sender’s question to formulate a response.

Other chatbots involve proprietary technology. CSUN’s relies on a standard SMS text format, making it compatible with Android phones and iPhones, which more than 50 percent of the school’s students use, according to a campus survey.

At CSUNny, the bot’s knowledge base currently contains more than 3,000 understandings. Students typically receive a response to their question within 10 to 15 seconds, Adams says. In 2020, they sent more than 42,000 messages to the bot.

“The largest chunk of those questions were about registration, classes and advising, but we also had a bunch about the coronavirus, applications status and financial aid,” Adams says. “It’s kind of amazing how much students are willing to engage with it.”

A number have talked to CSUNny about their stress levels, Adams says. The system is also set up to flag certain keywords and notify the appropriate entities, such as a university counselor or the campus police.

“There have been some students who have indicated a desire to self-harm. That immediately goes to the counseling center, and then we can reach out and get connected to them right away,” Adams says. “That’s a piece we didn’t have before, and it doesn’t happen a lot, but when it happens, you think, ‘I’m really glad we had that available.’”

LEARN MORE: Automation adds agility to campus service delivery.

Advanced Bots Still Require Human Training

While automation is a key component of the structure of AI-enabled bots, the technology can require some human involvement.

“There is a prebuilt stack of questions they can pull from,” Velazquez says. “However, like anything with AI, you have to train it. The more queries it receives, the more it learns, and the more accurate answers it can give in the future to better serve the educational community.”

In some cases, Velazquez says, organizations work with a managed service provider whose business analysts can identify which questions, issues or tasks could be best supported with chatbot technology and to design any subsequent workflows.

Since launching CSUNny, the number of questions that the bot can accurately comprehend and answer has risen from between 50 and 60 percent to about 80 percent. One dedicated employee works with the bot regularly: a writer, Adams says, not a tech person, because the university wanted to ensure it had a consistent voice.

“She is in there every day because we also need to monitor it,” Adams says. “If it’s getting 80 percent of the questions right, that means it’s not getting 20 percent right, and we want to be able to answer those.”

As the functionality has progressed, the school has created campaigns focused on specific challenges, such as an alert system that lets students know early in the term if they’re in danger of failing a notoriously difficult class, while also providing information about helpful resources.

OU’s Office of Admissions and Recruitment primarily uses its bot for user-initiated interactions, but Kunkel says the department also used it last summer to remind students about a class registration day.

“The beautiful thing about texting through the platform is when students respond,” Kunkel says. “They can have a lengthy Q&A session with the chatbot about where to park, what time they should be there, different things about the orientation.”

The Bots Are Here to Stay

Day to day, OU’s chatbot autonomously answers questions about admissions, enrollment and other topics. Certain contact solutions, such as Cisco’s Unified Intelligence Center, can also integrate with chatbot technologies, enabling the chatbot to connect students with a call center agent during business hours, if necessary.

“The bot may say, ‘It looks like I don’t know the answer. Would you like to be connected to a live human?’” Kunkel says. “Ultimately, this helps with call wait time and allows admissions counselors to spend more quality time with prospective students.”

Schools thinking of adding a chatbot should first try to determine where the technology would be most useful, Velazquez says. “Start by looking at the lowest-hanging fruit,” she says. “Where would it make the most impact?”

While chatbots have yet to reach ubiquity in higher education, Velazquez anticipates they’ll find their voice, given the mounting interest.

“Digital assistants are here to stay,” she says. “The technology will not just be a tool for student engagement. It will be a tool for supporting faculty and staff, to simplify their workloads and to truly focus their time on more complex tasks. It will definitely be something that will continue to evolve, and the use cases will continue to grow.”