12 Days of Christmas Wishes — Part 2 — AI for Instruction
As an educator, I am highly skeptical of a general purpose teaching AI that takes the job of a teacher any time soon. Teaching is one of the most nuanced human occupations, relying on equal parts empathy and academic rigor. The place I think AI can make huge headway for the next ten years is augmenting the capabilities of teachers and reducing the time it takes them to do their job. I think a lot of these capabilities apply to all levels of education from early childhood, k12, higher ed, and workforce.
At my last company Zeal, we attempted to build machine learning to automate assessment. Our system was a lot more rigorous than anything you would see a teacher use normally. The biggest problem with it was that assessment creates more work, because it usually makes it clear how much your students have to learn. That caused us to pivot into tutoring, where we used human tutors instead of AI. I can say that assessment really lends itself to machine learning, because what questions you ask, how many you ask, where you send a student when they answer correctly or incorrectly, should be based on lots of previous student experience. This is precisely the kind of job that machine learning can do well and humans can not.
Another important area for AI is homework. Homework is a monumental time suck for teachers. It’s really important for students to practice what they have learned, but grading that work after the fact is not very useful. What would be far better is an AI that helps grade that work as a student goes along. I have an investment in a company called Kunduz that does this for test prep. 17Zuoye in China does this for homework there.
Another area that students need a ton of directed practice is writing, including spelling, grammar, and composition. Once again, this takes a ton of teacher time to correct and is slow and tedious because you write — submit-get corrections — rewrite in very slow loops. It would be much better to have the AI looking over your shoulder and making suggestions as you go. Companies like Writelab have worked on this problem.
The most far-out idea I’ve experienced is the idea that an AI could monitor students behavior to help the teacher understand motivation and understanding. This starts to verge on what we would call ‘creepy’ in the US, but in some contexts (online tutoring), might not feel as weird. Magic Mirror, is a new technology created by the largest tutoring company in China, TAL. It watches the faces of the millions of students being tutored every day in hopes of understanding their emotion and relaying that to the tutor.
If you are building something in the AI field, send me an email -john@danners.org. I would love to help!
