AI's Impact on Ed Tech

Artificial intelligence is making its way into a variety of education technologies. Here, vendors talk about their current and future work with AI in the higher education space.

outstretched hand holding digital artificial intelligence icon

In 2015, when Georgia Institute of Technology professor Ashok Goel experimented with using an artificial intelligence-based teaching assistant called "Jill Watson" to answer students' questions in online forums, it opened a lot of eyes to the potential of AI on campus. But there remained a lot of well-founded skepticism about how algorithms would be deployed. For instance, in 2016 AdmitHub CEO Andrew Magliozzi contacted universities to ask if they would like to incorporate an AI chatbot into their recruitment and retention strategy. The reception was unenthusiastic: "They didn't exactly say, 'Hell, no!" he recalled. "But most of them did not respond to our e-mails."

Still, machine learning technologies are indeed making their way into many ed tech products — focused on both administrative tasks and teaching and learning.

Despite the initial cool reception, Magliozzi said things changed for AdmitHub as word got out about a successful pilot at Georgia State University. The Boston-based company also benefitted from a gradual change in thinking about AI, he noted. It now has 50 campus customers for its custom virtual assistant, including Arizona State University and California State University, Northridge. Institutions deploy it to engage with students with text messages, make recruitment easier, increase enrollment, ease orientation and improve retention. Campuses are particularly eager to reduce their rate of "summer melt," the term for when incoming freshmen fail to show up in the fall.

"Even administrators who are nervous about it will do a focus group before launching, and without fail the students will say this is cool and that they want it on campus. And then the administrators see it gets results. It changes student behavior in a positive way," said Magliozzi, whose experience includes running his own tutoring company and working at a natural language processing startup.

Magliozzi describes AdmitHub as a student success company powered by AI, but he stresses that the technology is secondary. "We help schools devise a conversational strategy and deploy it. It just happens to involve a chatbot. We have to understand the main challenges and available resources at a school, so we can help them achieve their goals by directing students to the resources already on campus that happen to be underutilized," he said. "We are not the ones getting an increase in enrollment or decrease in dropout rates, but we can nudge students to the resources that already exist on campus that we know contribute to improved student outcomes, such as a tutoring center or financial aid office. We help streamline processes that already exist and help students in the process."

A Nudge from the LMS

Instructure, the company behind the Canvas learning management system, says there are several ways campuses are using artificial intelligence to take advantage of data from the LMS.

"One of the things we have prided ourselves on is building Canvas as an open platform, which means institutions have access to their own data," said Jared Stein, vice president of higher education product strategy for Canvas. Some institutions are warehousing that data and applying their own data science teams to gain insights from the data. Others turn to third-party providers, such as learning analytics vendor Civitas Learning, that take advantage of the Canvas application programming interface to access data for common customers. "When we have a Canvas customer who is also a customer of Civitas, they will give Civitas permission to receive their Canvas data, and Civitas leverages some machine learning techniques to deliver their products and services," Stein explained.

The company has a similar relationship with a vendor called Zoomi, which uses algorithms and analytics to predict learning outcomes and guide the creation of personalized learning programs.

In addition to making data available, Instructure is engaging in its own research and development activities that leverage machine learning. "We believe that data analytics is going to best serve teachers and students when it is sort of invisible — when it informs features or the ways the system interacts with you instead of being a dashboard or visualization," Stein said. To that end, Instructure is piloting a tool called "Nudge." The idea is that some students would benefit from nudges to engage with their course materials or classes to become better students. "In the current iteration, Nudge uses machine learning techniques to prompt students to log into Canvas or turn in assignments if we think they are likely to not do those things," he said, adding that it is crucial that the message be personalized. "We don't want to send nudges to the A-plus student who is already going to turn in the assignment. That is just going to be annoying."

Supporting Recruitment and Retention

Andrew King is a machine learning developer on the applied research team at Ellucian, which develops enterprise resource management, student information system and customer relationship management solutions for higher education. He says Ellucian is working on proofs of concept of machine learning approaches to retention and recruitment.

"We are right at the beginning of understanding of what machine learning can accomplish. It is going to find its way in just about every process," King said. "On the recruitment side, it can be difficult to find candidates right for a company or in this case right for an institution. We will start to see a lot of work being done in that area to support admissions teams to make determinations about their applicants. Ultimately, we want these decisions to be made by humans, but supported and enhanced with AI."

"Machine learning can be used to do identify patterns in the data," he noted. "For instance, perhaps students who get some financial aid are more likely to stay at an institution. Another use is to try to identify students at risk of leaving the institution and flag them for contact by advisers." Still another use might be to empower students with their own analytics, he added.

One of the complicating factors, King said, is that institutions are collecting data but it may reside in many different places, so being able to aggregate it in one place so a data scientist can do these analyses can be a burdensome process.

Rate of Change Increasing Exponentially

Ed tech giant Pearson deploys machine learning in several solutions. One product called Revel can provide students with automated scoring and feedback on essays. "The nice part about that is it helps students get some feedback early in a course when the instructor might not have the resources to give all the feedback they would like to," said John Behrens, vice president of advanced computing at the company's Data Science Lab. "It is a way to provide feedback in complex performances, which is one of the main ways we are going to see machine learning go in helping people through the learning process."

Pearson's MyLab and Mastering adaptive learning solutions have features based on machine learning that recommend different types of practices to optimize learning, as well as help instructors identify struggling students, he added.

Behrens is enthusiastic about the pace of change in the field. "I cannot tell you how exciting it is to be in this space at this time," he said. "With the revolution that is happening in deep learning and other advances in machine learning, the speed of change and innovation is increasing exponentially. We have a lot more data than we ever had, and we have the ability to manipulate that data through networks and high-speed, scalable computing through the new cloud environment. Combine that with the new inferential techniques that are largely open sourced from the big technology companies like Google and Facebook, and the rate of change for technological improvement is just remarkable."

When asked about creating virtual teaching assistants like Jill Watson, Behrens said Pearson is working on similar solutions, but he cautioned that it is not easy. Creating a chatbot for a specific course requires a certain set of tools and data, he explained, "but to scale that across disciplines, where each discipline has its own way of talking or thinking and its own professional standards, that takes another level of sophistication in machine learning, but also in understanding the educational and social ecosystem."

AI in the Classroom

Most commercial software vendors have focused their initial machine learning work on recruiting and student success, because the predictive models address problems where the return on investment can be measured. But what about applying machine learning to pedagogy? "There is not a lot of work being done around use of AI and machine learning in teaching and learning," said Kyle Bowen, director of education technology services at Penn State University. "We describe it as a moonshot. If we can figure this out, we could have a dramatic impact about how people think about open educational resources [OER], active learning and the design and development of courses. The goal is to support the faculty so they have time to be more creative."

PSU has developed machine learning tools to help faculty choose appropriate OER materials and identify the prerequisite knowledge a person would need in order to understand a particular body of text. (See "How Machine Learning Is Easing OER Pain Points.") The university also is working on a prototype algorithm that, given an OER chapter or a textbook, can suggest multiple-choice assessments and distractor questions.

One of the newest areas PSU is working on is something Bowen dubs a "Fitbit for Teaching." His team has set up microphones in an experimental teaching classroom to capture audio levels and conversations like a lecture capture tool might do. "We do an analysis using machine learning tools of the interactions in the classroom to identify what types of activities are happening," he explained. For instance, the system might measure the amount of time spent on direct instruction vs. other types of activities. The intent is to provide feedback to instructors on what actually happened in class, so they can compare that with their intentions and fine-tune their instruction.

Featured