These sees Bloomberg Business Week with Carol Messer and Tim Stenebec on Bloomberg Radio. We did catch up with sal Cohn in late January. We got his thoughts on learning in our post pandemic world, what things are sticking, what things are not. We talked online education, something he and his team have been doing for about fifteen years or so, and now educators of all types, as you know, have a new tool maybe to use. So back to talk with us once again about the role of AI and
education and their specific moves in that regard. So com back with us. Founder and CEO Khan Academy with us via zoom from Mountain View, California. Sow, good to have you back with us. How are you. I'm doing great, Thanks for having me. Well, it's great to have you here once again. Late January. We talked about a lot of things, including AI, which it feels like all of a sudden everybody was talking about it once again. You're
launching your own AI platform. Can you tell us about it? Yeah, And when we talked in January, I was under n I was under NDA, so I couldn't tell you what. Okay. Back in back in August, open AI reached out to us. They were working on GPT four. As many folks know Chat GPT when it came out in November, it was
based on GPT three or three point five. But when we saw what was possible with GPT four, we said, hey, this could actually allow us to address some of the holy grails in education, making it truly interactive, giving every student a socratic tutor, being able to do things that might have only looked like science fiction a few months ago. And Open AI wanted to work with us because they said, look, GPT four is going to be exciting but also scary
in certain ways. We want to be able to launch with some social positive use cases, especially in education, especially from folks that with folks that people trust. So we've been working pretty feverishly on it for many months under NDA. So it was really hard in January when you were asking me questions about Chat GPT for me to not spill the beads. But a couple of weeks ago we launched and what we're calling con Migo, which is really our incarnation of AI on Kin Academy using GPT four.
It does essentially two major strands of things. One is now when a student is doing something on kin Academy, it acts as their tutor, and it really does as opposed to just giving you the answer. If a student says, you know, telling me the interests, I'm your tutor. I can help you and if you know, what do you think is the next step. So it really tries to push them socratically. It's not just in math, it's in
every subject. We have history, science, reading, comprehension. There's now a tutor and this is a beta now where we've only launched it to a few thousand people. There's a waiting lists that's forming for folks who want to join. It's about twenty thousand people right now, so we're trying to take people off as quickly as possible. But the other thing we realized is the things that would have been impossible without AI. We have activities now where students
can have conversations with historical characters or literary characters. There was a student recently, a high school student. She was trying to understand some of the symbolism and the Great Gatsby and she says, wait, I have conmigo, I can I can now talk to Jay Gatsby himself? And she has why are you staring at the green light? And he's like, well, it symbolizes the things that I'm trying to attain but I just can't. And she actually apologized to him for taking up this time. The AI simulation.
We're creating teacher tools for them to create lesson plans for them to refresh their own knowledge. We have a kind of a built in guidance counselor now as an AI I see a quick quick take piece. Yeah, a conversation between teachers and students. We're going to be talking and con migo. Yeah. Yeah, But it's just so it's it's fascinating because of the amount of use cases that
you mentioned. Is it available? Talk to me about the availability, because I don't think I'm clear on how widespread access is at this moment and where you're at in that and where it's going. Yeah. So when we launched two weeks ago, and we're still kind of in the process of launch launching, there's a couple of things we wanted to figure out. If one, we just wanted to make sure it's working well for in the last weeks we've started to get some really positive feedback on that one.
The other thing we're trying to figure out is how do we cost this out and how do we resource this the computation costs for leveraging these these next generation models like GPT four, it is not free even if you're a Chat GPT pro customer. And these are folks
paying twenty dollars a month to open AI. They only get GPT four for I think at twenty five interactions per day and you have to wait three hours, so they've so in a lot of ways, Conmigo, what we're doing is in many ways the most unfettered access to GPT four. So we're trying to figure out ways that we can we can resource it. So right now there's a waiting list, when people to get off the waiting list. Right now we're prioritizing people who are donating. It's a
twenty dollars donation. It helps us resource the computation costs and also do the R and D. We are not for profits, so that is a donation, a true donation. But we're trying to figure out how we can bring that that that price down over time so that we can make it more and more accessible. Because rather I think about you and kon Academy as this great leveler when it comes to education. Right, your videos are out there,
anybody can access them. So how do we make sure that once again, we don't have a technology potentially that widens the gap rather than reduces it. Yeah, this is front of mine for us, and I think there's two dimensions. So that one is just the cost issue that we've already talked about. Obviously, everything that we stand for is our mission is free world class education for anyone anywhere. We're going to try to make this as accessible as possible.
Keeping in mind there is a higher cost, but if you think about if we can bring this down to let's say a few dollars per student per month, this
is dramatically lower cost than conventional tutoring. And a lot of the school districts have been spending a lot of money on conventional tutoring lately with money that was released after the pandemic, and they haven't seen results because it hasn't been connected to what students are doing in class, and it isn't happening during class, So the kids who are engaging or are usually the kids who don't need it.
The kids who need it aren't engaging after school, etc. So we think we're gonna be able to get this to a much broader set of students. We're starting to partner with some school districts, large urban school districts, so we can bring it to their students for free, and then hopefully we make it more and more accessible over time.
The other dimension to your question, the way we've been thinking about it is a lot of school districts announce that they're banning chat GPT, and there's some good reason why they're doing that. Obviously, students can use it to cheat, They can do shady things with it because it's a very open tool on Conmigo, all of the conversations are viewable by the teachers and the parents. There's a strong moderation. All of the activities are designed to be very pedagogically positive.
There's hopefully no cheating. More. It will write a paper with you, it won't write a paper for you. It will tutor you, but it will not tell you the answer. So some of those same school districts who banned it, and I was afraid when the band because I'm like, this is going to create another digital divide because the well resource kids are still going to use it. Those same school districts who banned it. Now we're saying, hey, this is what we wanted. We don't want to stick
our head in the sand. We want to move forward with the times. We just want to do it in a safe way that's actually pedagogically positive. So we're pretty excited that hopefully we can bring get the cost down. Some of that is open AI and Azure and Microsoft and all of them getting the efficiencies, and then on our side, we're trying to bring the efficiencies in so that we can make it as accessible as possible and
make sure that all kids have access to this. But access is such a multi layered thing, Sal, I wonder how you're thinking about partnerships when it comes to that, because we're still trying to wire up the country. Well, right, we've got the internet access thing. I'm on your website. It looks like it's like an iPhone is being used or samsuent or whatever it is. Like, what about people who don't have access to a smartphone? Right, we have about a minute and a half left, But I wonder
how you're thinking about that. And then we'll come back and talk some more. Yeah, it's a tough question. Simple answer is digital access in the US at least has gotten dramatically better. Cell Phone access in the rest of the world also has gotten better. We are trying to skate to where the puck is going on that front. In the US, we think it's all about getting it to students in the school setting because at home, as
you know, there is more inconsistent access. And so that's why we're working so closely with school districts, which for the most part now do have one to one laptops or at least close to that inside of their classrooms. And then in other countries, in places like India, we're partnering with state governments and that is multi multi layered. We need to translate the content, we need to align it to their standards, and we also have to figure
out how we get the vice access right. This is why we always like talking to you because it's the global perspective. It's not just about teachers and kids and here in the US, but it's really about teachers and kids everywhere in the world. So so one of the things you're talking about, you know, your new AI platform that you guys have created, and I came across, are we and prepping for our conversation with you. Stanford put out a report earlier this month warning that decisions made
by AI could lead to nuclear level catastrophe. Why is it not too early to start playing around with the next level AI when it comes to education? Well, I think, like any technology, there could be a lot of positives and there could also be a lot of possible negatives. And I think as a society, it's all about mitigating the negatives and then using it for the positives where possible.
The genie is to some degree out of the bottle a little bit, and so I know, I tell this the team at Kon Academy, and tell this to our donors. I view it as our responsibility as a nonprofit in education to show how you can use these technologies and a very positive and a very safe and a very trusting way, Because if we do that, then we're going to be able to reach all these students that we've been talking about, and we'll hopefully be able to accelerate
their learning pretty dramatically. Now, there are other concerns about
what happens if you have super intelligent AIS. You know, who know who knows what GPT five or six or seven is going to be able to do planning military strategy or figuring out what targets are, etc. But even there, I personally think that I would rather have the actors in the US being on the cutting edge than US pausing and slowing things down and that and letting folks who are maybe not as altruistic continue to move forward.
You know, no matter what we do, I guarantee you the Chinese or the Russians are not going to be stopping their generative large language model research. So I actually think it's a national security issue to keep moving, not to stop. Yeah, that's that's such an incredible point. And also that you were mentioning that this is just a story we've heard before, Like every single innovation has this big question, and you can't just like let the United States be the only one to not have social media,
for example, because we're worried about the potential downfalls. Right, that's right, And you know social media is not I don't. Well, it can be a national security issues, as we've sometimes seen. It can lead to polarization, it can lead to you know, people believing things that aren't true, but it doesn't quite have the same strategic consequence as something like artificial intelligence.
So an artificial intelligtion is going to intersect with social media, It's going to be influencing how people think, et cetera. So I feel much better if relatively positive actors in the States and aligned or countries aren't thinking about how to use this responsibly and not falling behind folks who are who have less less issues with using it irresponsibly. Are you thinking about or concerned about any legislation on AI.
I know we talked about rulings at the school district level, but are you concerned about any federal legislation that might impact usage. I'm always worried when people want to pass legislation, should a regulation without having a clear picture of what
they're trying to protect against. I think right now there's a lot of knee jerk fears based on dystopian science fiction novels about AIS taking over, etc. You know, a lot of the folks calling for regulation can't exactly pinpoint to what tangibly they think is going to happen, and they can't point to actual examples of that happening. You know, the Internet is actually much more of a wild West than large language models are right now, and I think
for the most part, it's probably a good thing. It's been left unregulated. I'm not antiregulation. What I would say is, let's see where there are issues that pop up, and then determine is regulation the best mechanism to do that, and then and then that should be a you know,
chev robust conversation about that. For teachers, what would something like this, this kind of functionality and using generative AI specifically like you guys are doing, What does it mean for teachers specifically who We recently had an author on an incredible book and just you know, she embedded herself for a year with three different teachers and the stresses, the strains not paid well. It's it's a really tough profession right now, and to be fair, not really well
respected by many people out there. And that's unfortunate because this is talking about educating the next generation. So what might something like this do for teachers? Well, first and foremost, you know, your average teacher has thirty kids in the classroom. They wish they could replicate themselves, give more personalized attention to every student. This is something we've always worked on kind academy, but having this AI tutor in the room
with you. It is like replicating you, we've already seen this. When we've been piloting it at some classrooms, the teacher's like, wow, it just answered a question that I would have had to try to answer, but I wouldn't been able to connect with this other student. So it's already doing that. But I think even more importantly, teachers spend about thirty forty fifty percent of their time planning less than refreshing knowledge.
We have activities four teachers. We've made them a first year user here where it helps them develop lesson plans in seconds that would have otherwise have taken hours. It helps them prepare their knowledge, It helps them think about like almost like an instructional coach, what they how they can intervene with their students. So we're kind of calling Kanmigo the r AI a tutor for every student and a teaching assistant or an army of teaching assistance for
every teacher. And so I think it'll hopefully lower their burden and be able to free up more time for the teachers to spend with the students. Can it help with grading at all the simpleasures? Yes, I think it is. We didn't launched two weeks ago. We have been piloting some of this, but we want to make sure that
it's really robust before we launch it. But I think within the next year you're going to see things like a teacher saying, hey, Kanmigo, let's develop an activity for our students, and Kanmigo says, all right, let's have them all right, you know, whether the Civil War could have been avoided and how, and then the teachers like that sounds great, And then every student sees that assignment and then they work with Conmigo to kind of talk about
whether the Civil War would have happened otherwise. And then Kanmigo assesses it and tells the teacher, Hey, these students had some really exemplary responses. These students struggled. You know, I would give these students an a. I would have these students continue to work on it. Right, that's not science fiction anymore. I think we're months away from that, not years, Especially when we talk about individualized learning, like you think about this, it sounds conceptually like it could
really take us to that level. Hey, Saal just got about twenty five thirty seconds. Well we look back five years from now, ten years, and is there a superlative to describe what generative AI may mean to education. And I'm not trying to look for that crazy headline, but I'm just curious how you're looking at it. Just quickly. I think this is the biggest inflection point of our lifetimes, and I think this is going to give a tutor to every student on the planet and a teaching assistant
to every teacher. So it's a big deal. Well, listen, thank you so much, and so appreciate you coming back when you couldn't talk about it, now you could. We really appreciate you sharing that with us in our Bloomberg listeners and viewers. So appreciate it. All right. SoCon, be well, Take care. SoCon, Founder, chief executive officer of Khan Academy.