There’s been an explosion in online courses recently, with many top colleges and universities offering students all over the world the opportunity to study with the, no matter where they might live in the world and what time of day they have time to attend lectures.
These courses are fantastic at disseminating knowledge to a much wider audience than a course limited by physical location but one problem of online courses is the low retention rate. Ashoek Goel, a professor at the Georgia Institute of Technology suggests that the lack of teaching support is one reason for this low retention.
Even if there are online forums accessible to the students, sometimes the sheer number of attendees makes answering all these questions nearly impossible, even with a team of dedicated staff. As an example, on his online course in Knowledge Based Artificial Intelligence, the 300 enrolled students generally post 10,000 questions each time the course is run. That’s equivalent to 100 emails a day, all requiring a quick response. Of course, some of these questions will be very similar, but each require a separate answer. To achieve this, Goel employs eight Teaching Assistants but even they can’t keep up with the number of questions posted during the course.
To combat this, he came up with an exciting way of solving the problem. Goel teaches computer science, so it’s no surprise that he decided that the perfect way to combat the lack of interaction with online students was to employ modern technology. His plan was to create a new Teaching Assistant using Artificial Intelligence.
So last year, Goel employed his extra TA, Jill Watson. Jill was developed using technology from IBM’s Watson platform, hence her surname. Goel tasked some of his graduate students with the job of creating Jill. They fed over 40,000 discussion posts from the course forum to Jill, giving her the tools to deal with many common questions. The team wrote a special code that gives Jill the ability to answer the routine questions that pop up each semester. Students routinely ask when particular assignments are due and where they can find access to books and journal articles.
Initially, Jill’s answers weren’t up to scratch and she gave irrelevant and strangely worded answers. The solution to this was to post her responses in a forum that wasn’t visible to the students, whilst the team tweaked the code. One of the problems to start with was that Jill would get stuck on particular keywords and her answer would focus on those keywords, but in the wrong context. For example, if a student posted about meeting up to go over a video lesson, Jill would give the students a name of a book that contained information to supplement that in the video lesson. Slowly though, the team adapted the code until Jill was answering questions with 97% certainty.
When Jill got to this point, the human Teaching Assistants would upload her answers to the forums manually. After two months, it was decided that Jill could post the answer independently as long as she was 97% positive her answer was correct. Jill would also remind students of due dates and post questions on the forum in the middle of the week as a way of engaging students and encouraging conversation.
Goel left it four months before he revealed to his students that they had been unwittingly interacting with AI. Only one student guessed that she might be a computer. The plan is to keep Jill on, with a different name. Goel aims to have her answering 40% of all questions.
Goel does not anticipate that Jill and her AI successors will replace humans TAs, and this has never been his intention. He wants to use her to complement his TAs, and give them more time to focus on mentoring, coaching and motivating students, rather than answering mundane emails. As he says “where humans cannot go, Jill can go. And what humans do not want to do, Jill can automate.”
The use of AI to answer routine questions on the internet is more common that you might think, so it seems that having an AI Teaching Assistant is a natural next step, particularly for a course focusing on computer science. And if it means that more online students feel engaged and encouraged to finish their course then Jill and her team of human TAs have done their job.
Top image: Artificial intelligence illustration (Public Domain)