AI-Powered Adaptive Learning: A Conversation with the Inventor of Jill Watson
By now, Jill Watson has been run in about 17 classes, including graduate, undergraduate, online, and residential … By offloading their mundane and routine work, we amplify a teacher’s reach, their scale, and allow them to engage with students in deeper ways.
Ashok K. Goel, Inventor of AI Teaching Assistant Jill Watson, Professor of Computer Science and Human Centered Computing at the Georgia Institute of Technology
The structures and traditions of education are often slow to change. While it may adopt some of the innovations of mainstream society (laptops in the classroom, online delivery of services, etc.), the core system of education is largely retained. The last truly disruptive shift to the way we teach and learn came during the advent of the Industrial Revolution, when standardized curriculums that focused on math and reading were introduced to better prepare students for factory-level jobs.
Artificial intelligence may bring about yet another revolution to the way we live and work. While AI won’t be replacing the jobs of teachers just yet, it may be joining them as a colleague quite soon. The intervention is likely to be welcome. According to a recent McKinsey survey, teachers are working an average of 50 hours a week, and further research estimates that 20 to 40 percent of those hours are spent on activities that could be automated using existing technology.
By taking over a teacher’s mundane, burnout-inducing tasks, it can give educators more time to focus on the student experience. At the same time, AI can drive innovations within education itself by making it smarter, faster, and more personalized. The concept of adaptive learning, which seeks to personalize education to each individual learner, has gotten a jolt of new energy thanks to the power of AI. For a historically stubborn field, education appears to be growing up.
The way we teach and learn may be due for another tectonic shift. In some pockets of the world, the change has already begun. To get a look at three ways AI and adaptive learning are being used to help teachers and students, read on.
Meet the Expert: Ashok K. Goel, PhD
Ashok K. Goel is a professor of computer science and human-centered computing in the School of Interactive Computing at Georgia Institute of Technology. He teaches classes in knowledge-based artificial intelligence, computational creativity, and cognitive science. He served as co-chair of the 41st Annual Conference of the Cognitive Science Society and is the editor of AAAI’s AI Magazine. Professor Goel is also the director of Georgia Tech’s Design & Intelligence Laboratory and the chief scientist for Georgia Tech’s Center for 21st Century Universities, where he pioneered the development of Jill Watson, an AI-powered teaching assistant.
Meet Your New AI-Driven Teaching Assistant: Jill Watson
Jill Watson has the potential to be every teacher’s new best friend. A cousin of IBM’s Jeopardy-winning Watson, Jill is an AI-enabled teaching assistant who can answer student questions about a particular class and curriculum.
Developed at Georgia Tech in 2016, Jill was first deployed onto the online discussion forum of a graduate-level computer science class in knowledge-based artificial intelligence. Alongside a team of human teaching assistants, Jill answered student questions as they came in. At the end of her semester-long debut, students were not able to distinguish which of the TAs was the AI. Four years later, she’s still going strong.
“We continue to build more powerful versions of Jill Watson every semester,” says Ashok Goel, a professor of computer science and cognitive science at Georgia Tech, and the creator of Jill Watson. “By now, Jill Watson has been run in about 17 classes, including graduate, undergraduate, online, and residential. Next, we want to take her outside Georgia Tech.”
The first iteration of Jill Watson took between 1,000 and 1,500 person hours to complete. While that’s understandable for a groundbreaking research project, it’s not a feasible time investment for a middle school teacher. So Goel and his team set about reducing the time it took to create a customized version of Jill Watson.
“Now we can build a Jill Watson in less than ten hours,” Goel says.
That reduction in build time is thanks to Agent Smith, a new creation by Goel and his team. All the Agent Smith system needs to create a personalized Jill Watson is a course syllabus and a one-on-one Q&A session with the person teaching it. Named after the self-replicating character in The Matrix, the Agent Smith program clones Jill Watson, but makes her a specialist in the area of need. Teachers from any grade level or subject domain can have a deployable, AI-powered teaching assistant for their class with minimal set-up.
“In a sense, it’s using AI to create AI,” Goel says, “which is what you want in the long term, because if humans keep on creating AI, it’s going to take a long time.”
Ten hours is a small investment for overburdened teachers to make. Especially when the result is a time-saving teaching assistant well-versed in the class curriculum.
“No one ever complained that we have too many teachers,” Goel says. “By offloading their mundane and routine work, we amplify a teacher’s reach, their scale, and allow them to engage with students in deeper ways.”
A Curriculum for Every Student: AI-Driven Adaptive Learning
AI can save the student time, too. The concept of adaptive learning uses AI to figure out more precisely what students know and what they don’t know. It then dishes out lessons in a manner that’s the most logical for a particular student. In doing so, it can shift a classroom, digital or otherwise, from a passive, instructor-driven experience to an interactive, learner-driven experience.
SquirrelAI is an adaptive learning system that’s fueled over 2,000 learning centers in over 300 cities across China. A unicorn ed tech startup valued at over $1 billion, it has registered more students than all of New York City’s public schools. Their Intelligent Adaptive Learning System (IALS) breaks down a subject into tens of thousands of knowledge components, each of which is matched with learning content.
As students navigate through a subject, SquirrelAI’s system tracks their progress, and determines when they’re ready to push ahead or step down. So far, it’s not only been effective at helping struggling kids get back on pace, but also at helping those who excel push ahead even further: the twin sons of founder Derek Li are currently studying eighth grade physics on the Squirrel AI system, despite only being in second grade.
A multiyear partnership with Carnegie Mellon University has yielded the CMU-Squirrel AI Research Lab on Education at Scale. The lab is working on improving adaptive learning experiences for students through AI, machine learning, and human computer interface (HCI) technologies.
Just like other forms of AI, adaptive learning can’t replace teachers. It won’t be able to discern if a student learns better visually or actively. It won’t be able to give them personalized comparisons or teachable anecdotes. But it can be enormously helpful in making rote learning, skill acquisition, and knowledge retrieval more efficient. In doing so, it can lend a helping hand to teachers and students at the same time.
Smile Like You Mean It: Surveillance in the Classroom
Sometimes AI’s helping hand reaches a little too far. Consider the Class Care System (CCS) from Hanwang Education.
Launched in China in 2017, it centers around the installation of a “mandarin-orange sized” surveillance camera in the roof of an otherwise typical classroom. Once per second, the camera takes a photo of the entire class, accounting for up to 50 students, and then zips the data over to Hanwang’s servers. Using facial recognition technology and deep learning algorithms, CCS classifies each student’s behavior into one of five categories: listening, answering questions, writing, interacting with other students, or sleeping.
Based on their facial expressions, each student is given a weekly score, accessible through mobile app. Teachers, parents, and school leadership can access the scores of each of their students, and learn how much time they spent on each of Hanwang’s five categories. Hanwang stresses that this information can be critical in identifying when students need help, and in what area. The system improves itself over time, aided by the lack of any requirement to obtain consent from parents or students. Most students aren’t even informed that the system is in place.
Freaked out yet? You’re not the only one. The Class Care System went viral in 2018 on China’s microblogging site, Weibo, with the hashtag #ThankGodIGraduatedAlready.
Overall, AI-enabled teaching tools show incredible potential for a range of applications, especially during an era where much of the world is learning exclusively from home. From teaching assistance to surveillance, the uses will ultimately depend on one’s teaching philosophy.