Khan Academy Founder and CEO Sal Khan joins Yahoo Finance Live to discuss how AI and ChatGPT can be used in schools to help with tutoring and teaching assistance and to dispel views that it’s primary use for students is cheating. Source; Yahoo Finance.
Many good points here, and the video transcript below.
RACHELLE AKUFFO: Non-profit organization Khan Academy has wide eyes for the future of artificial intelligence. It’s testing a new virtual tutor and classroom assistant using OpenAI’s latest education technology. Now, this comes as schools across the country voice their concerns on AI in the classroom. Cities like New York, Los Angeles, and Seattle have moved to ban ChatGPT in the classroom. And universities in India and France took the same measures.
Let’s bring in Khan Academy founder and CEO Sal Khan to discuss this. Good to see you here today, Sal. So you seem to be taking a very different approach from a lot of the critics that we’ve seen out there about having ChatGPT and its use in schools. Talk about your role in this and how you’re really viewing how this could be used in your classrooms.
SAL KHAN: Yeah. A lot of these school districts and systems banded on the premise that students will be able to cheat and cheat in an uncrackable way. And to some degree, that’s not false. We’ve actually been working with OpenAI for about six months now. It feels great to not be in NDA anymore, so I can talk about what we’ve been up to.
So well before ChatGPT came out, and we immediately saw with the GPT-4 technology, that it’s more nuanced. It’s better at math. That we could actually make it, so it’s not helping students to cheat, but it’s actually just helping students to learn, to be as a tutor, to do a novel activities.
And so when we launched Khanmigo, which we’re rolling out– there’s a waiting list for it right now– as part of the GPT-4 launch last week, we’ve actually been doing demos with some of the very same school districts that you just listed that banned ChatGPT. And their response has been this is exactly what we wanted, because we’re showing them that, hey, instead of writing the paper for the student, it can write the paper with the student.
And the teachers, and the student, and the parents can have oversight over what’s actually going. They can actually see the transcripts, that it can act as a tutor. This has been a bit of the holy grail in edtech for a very, very long time. And even after the pandemic, a lot of folks have talked about, how do you get tutoring to help kids with all of the pandemic learning loss?
And so when we showed this to school districts, when I’m doing three or four demos to district leaders almost every day, they’re saying, oh, this actually threads the needle for us, because we realized there’s a lot of benefit. But this mitigates the risks.
RACHELLE AKUFFO: And it’s interesting, because a lot of people think of this as support for students, but also for teachers as well. We saw how strained a lot of public schools were, especially amid the pandemic. Talk about that aspect.
SAL KHAN: Yeah. At Khan Academy, we’ve always positioned ourselves as we want to be every student on the planet’s tutor accessible. And for every teacher, we don’t view this as a replacement for the teacher, we view this as an assistant for the teacher. I think every teacher would love to personalize things for their students more, to give every student more attention, or to have essentially a teaching assistant or many teaching assistants in their classroom.
And this is what that does. The average teacher spends almost half of their time, 30%, 40%, 50% of their time doing things like lesson plans, grading papers, et cetera. This technology, you know, there’s older technology that could do some of this. But this technology really starts to be able to help teachers develop lesson plans, not in hours, but in seconds, minutes, allows teachers to get assessment in ways and get grade things that weren’t tradable before, like free response and essays.
So a lot of what we’re launching with Khanmigo actually teacher-facing tools. A lot of times, teachers are the secondary thought. We definitely want to put the teachers first in this, because we think it’s important. One, if they save more time, it’s more time that they can spend with the student. And also, the more that the trust that they have in these types of technologies, the more likely that kids are going to be able to leverage them.
RACHELLE AKUFFO: And obviously, with every advance that comes with AI, a lot of people sort of look ahead and think what’s next, not just for where AI can take us, but also some of the dangers. We know Bill Gates recently spoke about that in an op ed. And we know that he has donated, the Gates Foundation has donated to Khan Academy. Talk about that dichotomy of sort of wanting to push the boundaries with AI and what we can do and sort of really managing some of the risks that also come with it.
SAL KHAN: Yeah. The primary risks are it can make up facts. The hallucination is literally the technical term for when these large language models do that. Historically, large language models have not been good at mathematics. There’s always been a fear that maybe it could introduce some form of bias. And what– these are all real risks. And what we’re doing is trying to minimize them as much as possible.
GPT-4 does a lot less of let’s call it the hallucination behavior than what you saw in previous generations. On the math side, it’s a lot better. And we’re doing a lot of things on the Khan Academy side, where we are able to improve the math accuracy and be able to act– have it act as a nuanced tutor in ways that you couldn’t do previously with large language models.
In terms of bias, in terms of students maybe doing– trying to do things that are inappropriate, this is where we have a very strong moderation filter. Teachers, parents can monitor what students are doing. And if the students get into a particularly unproductive or even potentially dangerous conversation, it immediately flags the teachers and parents.
So these are the types of safeguards that, once again, when we’re showing these school districts, we’re showing teachers, they’re saying, wow, this is exactly what we need, because 40% of our teachers are already using ChatGPT. So it’s not like, despite the bans, folks aren’t using it. These are the teachers who are using it. But they just wanted safeguards. They want it to be more pedagogically positive, not just something that just gives answers.
And, you know, we’re able to introduce novel activities that didn’t exist before, like students being able to interview historical characters or literary characters, or being able to get into a debate and fine tune their arguments on things that you traditionally couldn’t have done with that deck.
RACHELLE AKUFFO: So then when you think about the classrooms of the future, and where we are now, just as we’re looking at GPT-4, which is already leaps and bounds of what we saw with GPT-3.5, how do you see the future of classrooms looking?
SAL KHAN: It’s really an extension of what we’ve always been preaching here at Khan Academy. The future would be kids able to learn at their own time and pace, real students that are learning. I always say the best education for most of human history was to have a tutor, who caters their education to what you need. If you haven’t mastered a concept, they’ll slow down. If you find something easy, they’re going to speed up.
Industrial Revolution, we had mass public education. It was a great thing. But we had to make compromises. We batched students together. We moved them lockstep. I think the classroom of the future is going to be what Alexander the Great had with Aristotle. But you’re going to have that for all 30 students in the classroom, where they’re going to leverage Khanmigo, they’re going to leverage other tools that Khan Academy has to learn at their own time and pace.
But then the teacher is still in control. But the teacher isn’t lecturing anymore. The teacher is able to use the information. They’re able to talk to the artificial intelligence. What are my students up to? What do you suggest I could work with them on? Just what a good teaching assistant would do. And then they could do a Socratic dialogue, a game, a simulation. They could go sit next to five students and do a more focused intervention or maybe motivate them, while the other 20 or 25 students are able to continue to work for the next 10 or 15 minutes with the artificial intelligence with Khan Academy.
You know, things that I thought weren’t going to happen for 20 or 30 years, I think, are going to happen in the next five years. We’re already working on ways. The artificial intelligence can have a more persistent memory. So it can start to understand what’s going to help the students more over a longer periods of time. We’re making it ways for the teacher to be able to have conversations with the artificial intelligence about what the students are up to, similarly, for the parents to have conversations with them– ways for the artificial intelligence to actually moderate conversations amongst the human beings.
You can imagine a teacher, and I hope to have this in place in the next year, where a teacher says, hey, artificial intelligence, hey, Khanmigo, I want you to break up my class of 30 into groups of two or three. And I want you to moderate a conversation with every group of two or three, have them debate an issue, have them brainstorm something, have them come to their own conclusions.
And then you’re just going to report back to me, the teacher, you, Khanmigo, the AI are going to report back to me, the teacher, what are the students been up to? What are some of the stronger arguments? What students have been struggling? And I think this will make every teacher’s job that much more fun.
RACHELLE AKUFFO: I can imagine that also being translated to news anchors. Moderating panels are doing interviews. The future is looking very interesting indeed. A big thank you there. Khan Academy founder and CEO, Sal Khan, thank you for your time this morning.
SAL KHAN: Thank you.