AI Episode 2: What Does An AI Teaching Assistant Look Like?
Jill Watson, an AI teaching assistant developed at Georgia Tech, turned 5 this year. Jill has been steadily improving since her 2016 rollout in Ashok Goel’s computer science course. The course has more than 350 students each semester. Students participate in online discussion groups with teaching assistants and the professor each week. Jill was designed to respond to questions emerging during the course. To do this, Dr. Goel and his assistants entered multiple semesters of question and answer data into Jill’s AI. They categorized questions and taught Jill to formulate responses with 97% accuracy. Five years later, the technology has been adapted for other courses. Jill now provides students with answers to their questions in Biology. Click here to see Jill’s development.
Dr. Goel presented a summary of his research on Jill Watson in 2016 at TED San Francisco. The video below is 20 minutes.
Dr. Ida Comacho is another researcher at Georgia Tech who is investigating how AI can improve social interactions in online environments. She is examining factors (e.g., location, hobbies, pets, etc.) that lead to successful course engagement and small group formation. Students can opt-in at the beginning of a semester to be connected with others in the course. The AI pairs students based on weighted factors. You can learn more about Dr. Comacho by listening to this 22-minute podcast.
These examples show that AI teaching assistants can respond to student questions with 97% accuracy. They are on at all hours of the day, providing students with feedback to their questions in seconds. In addition, Dr. Comacho’s research is demonstrating how AI can identify personal attributes that lead to positive social networks. Here is a word of caution before we engage in a new AI education system. Dr. Chris Piech, a lead researcher into AI tutors at Stanford states,
Current algorithms can’t read motivation, and are far from engendering long-term learning gains, instead of focusing on engaging students for the short term. The technical challenges are enormous…
The potential benefits for students seem to outweigh the technical challenges. An AI tutor can provide students with answers to their questions at any time of day, 365 days a year. The technology has the potential to decrease student inequalities resulting from a geographic location or social status. The only requirement is internet access. But what are the unintended consequences associated with AI?
Jordan Harrod gives us a reality check about how AI is used in colleges today. She notes the primary purpose is grading assignments with a secondary focus on tracking student progress. Check out her 8 & ½ minute video summary here.
In the next episode, we dive into some of the things that can go wrong. Preview it by watching the following video if you’re interested.