Robot Teachers
Transform Education


There was always something a little off about Jill Watson.

In early 2016 Watson had been assigned as the teaching assistant for Dr. Ashok Goel’s graduate course in artificial intelligence (AI). Offered online through Georgia Institute of Technology, the course taught hundreds of students from all over the world. The workload, Goel remembers, quickly became overwhelming for him. There were far more students than in a typical class. They lived across a dozen different time zones and they all had questions for their professor at all hours of the day and night.

“It was a large class and the questions were constant,” Goel says. “It takes a lot of time to pursue all of those messages and answer the questions. So we started off looking for a way to make the work easier on ourselves.”

So he called Jill, an artificial intelligence-backed “robot TA” that was developed at Georgia Tech and based on IBM’s Watson system.

The students, at least in the beginning, didn’t even notice they were dealing with a robot TA.

Jill was an incredibly effective TA. She answered student questions within minutes, no matter when they contacted her. She offered in-depth answers to a wide range of complex queries. She was generally more accessible, more personal, and more upbeat than any human could ever be. The class rolled through half a semester before Goel gave up Jill’s real identity. Since then, he’s used the AI system in a few other classes and has noticed that, beyond helping with his workload, Jill also improves the overall student experience, making for better, more effective, and more engaged learning.

Professor Ashok Goel in the classroom. Image: Georgia Tech


“What we realized is that, when a student asks a question, a human TA may not get to that question for six hours, 12 hours, 24 hours,” Goel says. “But Jill answers your question within a few minutes. That does tend to be really important because, if you're a student say, somewhere in Japan, and you ask a question and you don't even get a response for 24 hours, by the time the response comes you have moved on. You're doing something else. But when Jill responds within five minutes then you're still on the topic and can stay focused.”

Getting Personal

AI is another step on the path to personalized learning, Goel says. Teachers play many roles in the classroom. They are coaches and mentors. They provide assessments and offer cognitive challenges. The long-term goal for systems like Jill, is to combine all of those functions into one solution that can do it all. Eventually, the hope is that students will be able to ask an AI system a question and, based on data about a student’s learning trajectory or demographic or cognitive profile, the answers and guidance that come back will be tailored specifically to them.

“My understanding is that 400 years back, 500 years back, much of education was personalized, except it was accessible only to the elite.” Goel says, “Then mass education started and all of us started getting learning, but it became impersonal. And now I think we are swinging back toward personalized learning, where we will have systems that will build very excellent models of students, and then use those student models to personalize all the tutoring and teaching.”

To get there, technology needs to become more human. These systems need to understand human beings. They need to understand what we want and how we think, and then use that information to interact with us. In many cases, AI needs to emerge from a technological vacuum and function effectively in the real, human world.

“Being able to have AI that understands the learner is incredibly important because it gets a deeper focus on that learner and understands what they're doing,” says futurist Brian David Johnson, a professor of practice at Arizona State University’s School for the Future of Innovation in Society. “But, for me, we need to certainly get the core skills and we need to teach, but we also have to remember that we're going to have this AI and this technology doing more and more of our work for us.”

It’s the things that machines can’t do in the classroom—empathize, read emotions, adjust guidance—Johnson says that will really matter as AI systems become more common.

“We can’t lose sight of the fact that education is not just about skills,” he says. “It's also about developing human beings and developing whole learners.”

Growth and Scale

The potential applications for robot teachers—and AI-based learning systems in general— are broad and ambitious. One limitation for educators today is reach. A teacher standing in front of a classroom can teach up to about 50 students at a time. A college lecturer can reach maybe 200 or so. Digitally, though, the scope is almost limitless. Digital assistance allows educators to scale their services and reach communities and individuals who didn’t have access to different types of education.

“If you look at the global picture, we're short somewhere around 18 million teachers in the world today,” says Thomas Frey, author and futurist with the DaVinci Institute. “Teachers don’t want to go to Africa, they don’t want to go to Siberia, they don’t want to go to Afghanistan. There are lots of places in the world teachers don’t want to go to, and that’s what’s created this shortage of teachers.”

As a result, he says, nearly 25 percent of children worldwide are growing up today without any schooling whatsoever. This hinders their potential and consigns them to unskilled labor and subsistence jobs. Digital forms of education can help reverse that and actually raise the IQ of the entire planet.

“We have the Googles and the Facebooks of the world trying to bring Wi-Fi to the other three billion people on the planet, so educating those people just changes the equation almost instantly,” Frey says “Even small things, like teaching people how to take care of themselves or how to use a water filtration system, minor improvements in understanding of how the world works would be a huge change.”

Tim Sprinkle is an independent writer.

Being able to have AI that understands the learner is incredibly important because it gets a deeper focus on that learner and understands what they're doing.

Prof. Brian David Johnson, Arizona State University


November 2017

by Tim Sprinkle,