A student in Ashok Goel’s class last semester had a question: How long could the computer programs, or “agents,” they were building take to solve problems?
Since it was an online course, the student posted the question to the group discussion board. One teaching assistant replied, pointing to a portion of the assignment that set a 15 minute limit.
The student clarified that their agent was running a little slow, and could take a bit longer. Another TA, Jill Watson, offered some leeway:
“It’s fine if your agent takes a few minutes to run,” she wrote. “If it’s going to take more than 15 minutes to run, please leave notes in the submission about how long we should expect it to take. We can’t have all the projects taking a long time because we have to run them in a reasonable period of time.”
Here’s the weird bit: Jill, the gentler TA, isn’t actually a human being. “Jill” is an unfeeling artificial intelligence, guiding a student who had no idea this was the case.
Goel has taught this online, masters-level class in artificial intelligence since 2014. But this past January-April was the first semester in which he let the AI, secretly, help teach.
Here’s how he pulled it off: While the core of an online course is usually a video series with the professor, most of the actual class — including interactions with teaching assistants — happens through text posts.
In Goel’s case his students, 300 enrollees in Georgia Tech’s online masters in computer science program, had a class-wide forum where they could ask questions about the class, its subject matter, and the assignments. Goel introduced nine teaching assistants who would answer them, slipping in “Jill Watson” among them.
Jill Watson takes her (its?) name from IBM’s Watson artificial intelligence service, which Goel used to build his digital TA.
As students asked questions, Goel and his TA fed them to Jill. At first, her answers were not so good. So Goel’s human TAs answered themselves.
“But over time Jill’s answers kept on becoming better,” Goel told Tech Insider, “in which case the human TAs would take an answer from Jill and post it to the real discussion forum. And even further along Jill’s answers became almost perfect, so that we let her loose on the discussion forum with no human intervention at all.”
The class began on January 11 and ran until a final examination on April 25. On April 26, he told his students the truth about their TA.
“It’s a graduate-level class in AI,” Goel pointed out, “And the 300 students in this class typically are professionals. So the average age is about 40, it’s not the usual 22-year-old. We have 40-year-old professionals, very educated, very motivated, very engaged. And they are, you know, very bright.”
But none of them really figured Jill out until it was all over (though a few speculated, mostly because of her last name).
Goel sees Jill Watson as the first step toward a world where more people have access to personalised, high-quality education — and the things he learned training her should help him build bigger and better AI tools in future. Goel wants to build a world where AIs make online education better, more accessible, and more effective in teaching individual students who might otherwise drop out for lack of individualized attention.
Jill is just his first step.
This is the first article in a quick, three-part series on Jill Watson, her training, and her implications for the future of education. Check back soon for more.
NOW WATCH: Microsoft’s artificial intelligence is so smart it can tell blind people what’s going on around them
Business Insider Emails & Alerts
Site highlights each day to your inbox.