Opinion | A.I. Could Actually Be a Boon to Education

Khan and his team used GPT-4 as the engine behind software called Khanmigo (“Khan” plus “amigo” — a little goofy).

Khanmigo isn’t supposed to give away answers. Like a good flesh-and-blood tutor, it engages students in a Socratic dialogue to guide them. It’s good at figuring out what they aren’t getting. Plus, Khan said, “it lets students do things you could never do before, like talk to literary or historical characters.” He chuckled about one girl who had a conversation with an A.I. incarnation of Jay Gatsby. The experience was so real to her that when she finished she apologized for taking so much of the character’s time. For safety, all interactions with students are recorded and shared with parents and teachers.

Khanmigo is being tested in the public schools of Newark, N.J., as well as in School City of Hobart in Indiana, a charter school in Phoenix, the Mountain View, Calif., campus of Khan Lab School and Khan World School, which offers classes online.

There have been some growing pains. For example, GPT-4 doesn’t understand what it spits out and is not naturally good at math. It can generate wrong answers, which it gives with supreme confidence. Khan said that he and Kristen DiCerbo, the chief learning officer of Khan Academy, and others spent two weeks giving Khanmigo guidance on the ideal way to respond as a math tutor. He said the tool is much better at math now.

Another concern: When I tried Khanmigo on finance, I found that when I played dumb and asked for hints, it pretty much gave me the answers. “That’s good feedback,” Khan told me. “We’re trying to thread the needle between giving people too much and not enough.”

By highlighting the potential for artificial intelligence to do good, as I have in a few newsletters, I don’t mean to dismiss the risks. I’m as scared as anyone when I read warnings from people like Geoffrey Hinton, who was an A.I. pioneer at Google, or Yuval Noah Harari, who wrote in The Economist last week: “We have just encountered an alien intelligence, here on Earth. We don’t know much about it, except that it might destroy our civilization.”

But I hope it’s possible to get the good from A.I. while seeking to throttle back the bad. Khan, for one, has been using Khanmigo as his own intellectual amigo. Recently, he told me, he used it to satisfy his curiosity about an aspect of supernovas. He said it’s better at helping his kids with their programming homework than he is, even though he has a master’s degree in computer science from the Massachusetts Institute of Technology. “It is eerily good,” he said.