If you're seeing this message, it means we're having trouble loading external resources on our website.

If you're behind a web filter, please make sure that the domains *.kastatic.org and *.kasandbox.org are unblocked.

Main content

Part 1: Fireside chat with Sal Khan and Hadi Partovi

Join Sal Khan (Khan Academy) and Hadi Partovi (Code.org) for a fireside chat on embracing artificial intelligence in the classroom. Teachers can expect a lively discussion including some of the most controversial topics surrounding AI in education. Presented by: Code.org, ETS, ISTE, and Khan Academy.  Created by Code.org and Khan Academy.

Want to join the conversation?

Video transcript

Welcome to AI 101 for teachers. My name is Jess. I'm a former teacher and now specialize in adult learning. I will be your guide for this professional learning series. Artificial intelligence has become a hot topic in education, leaving many educators wondering about things like what is AI or my students going to use AI to cheat? Is AI going to replace me as a teacher? What should I be teaching my students about AI Throughout this professional learning series, our goal is to answer these questions and more, while also equipping you with the knowledge and skills you will need to navigate this new landscape. In this session, we will hear from Hadi Partovi, the creator of code.org, and Sal Khan, the creator of Khan Academy. Let's hear what Hadi and Sal have to say about some of the biggest questions related to AI in education. AI has been around for many years, but what's happened very recently with the launch of ChatGPT and similar tools is that AI can actually generate information it can generate creative writing, images, even videos. And so what used to be a tool for pattern matching can actually create new ideas, new thoughts. And the reason this is very important in education is there's so many aspects of education work that either teachers do or that students do that can now be done automatically using generative AI. So I know like both of us, I'm sure as just tech nerds of sorts have been paying attention to generative AI for many years, and you had GPT2, which could write and Elon Musk famously said, you know, this has to be controlled in some ways because it could write very convincing articles, but it was nonsense what it was writing, so it could be good for generating fake news. GPT3 did similar things, although it was just better at it, which in some ways made folks even more worried. Then things like Dall-E came out where it showed that you could create images, that you could have, you know, the Mona Lisa and you could tell the computer or the generative AI to say, well, what's what's the back of her head look like? Or what is the rest of the scene actually look like? When I was monitoring all of that over the years, I was just like, Oh, this is cool. But it's really just a quirky thing. It's not going to be really useful any time soon, especially in education. And I think what really changed was that transition from GPT3 to either you could say GPT 3.5, which is Chat GPT or more importantly GPT4. And that's when it really hit me that this thing that looked like a novelty for the last several years now could start to have a seemingly understanding, seemingly understand information and make sense of it, and not just answer the question, but explain why it chose that answer. Explain why the other choices are not the answer. Generate generative. It could generate other questions like it and it could you could prompt it so that it could act as a tutor. It could act as a teaching assistant. And so that's when we started to appreciate, okay, the world is now different. And this is not only going to be relevant to education, but it's probably going to transform it in some pretty significant ways. When you use a tool like ChatGPT or any tool that's generating text or code as an AI, these are based on an underlying technology called a large language model. And a large language model is basically a giant neural network that is trained on all sorts of information, all the books or articles or code that it can find on the Internet, and then can then actually generate new information, new essays, stories, even writing new code and it's worth understanding that this technology is not truly intelligent. It is just using statistics and probabilities to generate new information at Code.org, we're actually creating a video series explaining exactly how large language models work. And in fact, the chief technology officer of OpenAI Mira Murati, is one of the stars of this video to talk about basically an explanation of large language models. But one thing I will say, because people have said, oh, is this statistically figuring out the next best word to say in a lot of ways is starting to teach us a little bit about ourselves. like even when I'm talking right now, you know, when I'm speaking in oral language, I'm just expressing a series of thoughts that are coming. And I'm not consciously deciding each next word. I just know that the next word is going to be whatever it is going to be. So large language models to your memory is a neural. It's neural net. Neural nets are computer essentially computer simulations of neurons and synapses. And they're starting to approach the complexity of at least large chunks of the human brain. We're not at a point where A.I. is conscious self or has its own motivations and desires other than what we ask it to do. But yeah, it's worth noting that the probabilities and statistics, the numbers inside of an AI in generative AI or large language models are basically literally the computer equivalent of the synapses and the connections in the nerves in your brain, and you have trillions of nerves in your brain. And then the strength of all those synapses is what forms your memories, creates your thoughts, and gives you all of your intelligence. In a large language model that's being simulated across literally millions of computers. And is that intelligence? It's at least a representation of how your brain works. And it's interesting that they this works on math. Your brain works on neurons, but the outputs of what a large language model provides are quite similar to what humans view themselves. I think pretty much anybody working in education is surprised. was surprised to be able to ask AI to write an essay about the Civil War or explain photosynthesis or create a short story about a princess living on Mars, or write the code to, you know, to reverse, to shuffle a deck of cards, whatever it is. These are things that you might think of as homework or as exams for teachers. You can just ask them to make a lesson plan or to make a presentation about some topic in history. And it pretty much does all of it pretty well and is really causing a need to rethink how we do education and what's even 100% agree. What's even more amazing than that is it can answer these questions and do these tasks, but then you could even ask it to take on roles and personas. You can say be Robin Williams and Dead Poets Society and inspire me about literature. And it did it. And that's what that's what it's like. This. This is getting weird. it's hard for schools to say what tools you do or don't use when you do your homework. So the reality of the way homework is taught or assigned in most schools today, lots of it can be done automatically with a AI when ChatGPT first came out, I showed it on the first day to my son he’s 16 and his first question is can I use this for homework? He was like excited and then was about to answer. He said, actually, what's the point of school if this can even do all these things? And those two questions are sort of the two bounds of what I think educators need to think about. Because if we just continue assigning homework the way we have been, yes, you can use AI to do that. And you could argue that that's cheating. But I actually think we need to evolve what we think about as being the work that students need to do at home. Your point about homework. You have very little control about what happens. But I think in some ways it's the signs, a healthy spotlight. Because before Chat GPT you had chat big sister or or some people had parents or tutors and a lot of poorer kids did not have any of those things. The biggest inequity that forms in our school system is when the kids aren't in school. It happens when you assign homework, you or I can help our kids, or we might help get you get a family member to do it, or we might be able to hire a tutor. A lot of families can't afford to do that. That's where the inequity develops. And if you can do more in school, the better. When you think about equity and AI, there's two almost opposite things going on. On the one hand, a AI closes an equity gap in that every student who has access to it now can get basically a low cost, personalized tutor. They can get a lot of things done that in the past you needed basically more money to get done. AI basically provides a superpower to anybody who has access. On the flip side, the student who doesn't have Internet access doesn't have a device, whether it's a phone or a computer, is actually now farther behind, because they may see the gap of not having computer became more costly than it was before. Because here's this yet another superpower that you don't have. A.I. technology as a whole can have strengths and benefits, as we've talked about, but it also has downsides and risks. It makes mistakes that could create misinformation it can have biases. It can have either a political bias or a racial or gender bias. It might have it might get tricked into creating dangerous content. There's all sorts of potential pitfalls. And this is part of why schools and educators are have in many cases, hit on the brakes to say, let's not jump into this too quickly. But it's worth noting that there isn't a single tool that is all of AI. There's multiple tools coming out, some of which are general purpose, like ChatGPT, which is not even licensed for use by students under the age of 18. But there's tons of tools that are being created specifically for the purpose of education. Khan Academy is Khanmigo being a great example of that, where these tools have done specific things to protect against these risks and these these downsides to protect student privacy, to prevent the creation of misinformation or hallucination or biases or or or dangerous content. So the most important jobs educators or at least education I.T. departments need to think about is not whether you ban AI or allow AI. It's really choosing which tools are the ones you which AI tools are the ones that are ready for education that address these downsides. Yeah, I couldn't agree more and even on those very real downsides, like bias and misinformation and making mistakes, hallucinations. I also always point out that, you know, compared to what is already there and what also has bias and misinformation, the Internet has all of those issues. Other human beings have all of those issues. And there's often not the transparency where you can audit what's what, what was said or what what bias might be introduced. So in some ways it's just an extension of a lot of that. We were working on generative AI and large language models from about mid 2022. And then we all remember end of November ChatGPT comes out But that took the world by storm because when you chat with it, it just seemed to feel a little bit more magical as we all remember. But immediately the education community and teachers, rightfully so. Wait, this thing can write essays and I'm sure a lot of students realize that this thing could write essays and it was a cheating tool. And I was really worried because I was like, Yeah, yes, it can be a cheating tool and charge you, but it was not designed for education. So in March when we were able to to launch what we call Khanmigo, we were able to show that not only can you mitigate some of that, you can create generative AIs that have guardrails, everything that a student does, it's monitored by a teacher. You can have a second AI monitor what the student's doing with it, so they do anything shady, whether it's unproductive, unhealthy behavior all the way to cheating, it can actively notify the parents or actively notify the teacher. And then on top of that, create ways. So the AI isn't doing the work for you, but doing it with you. So if you ask Khanmigo to write your paper, it won't. But to say I can do your writing coach. And so we think the future is actually going to be there's going to be certain task where I think teachers will say, yes, use whatever tool you want, because that's happening in the workforce where sometimes you do want to use these tools to maybe write some parts or give you some revision or craft some parts of what you're doing. But then there's going to be other places where teachers will say, I need to make sure you're writing this yourself. So what we're doing on our side, we're trying to create those tools where teachers can assign through the AI, the AI can work with the student and the AI can report back to the teacher. Yeah, I worked with them. We did some outlining, we did some brainstorming. I give them some feedback and this is where they are now. And this is even my first pass at how they did according to the rubric When educators think about generative AI, the first thing they hear about or think about often is ChatGPT because of its popularity and because it was the first popular sort of generative AI tool that took the world by storm. But it's worth noting that almost every tool in tech is going to have an element of generative AI built into it. And this is actually most relevant as some schools early on band ChatGPT just to sort of say like, can we stop this and figure it out? we should recognize that Microsoft Word is going to have generative AI it will just create documents for you, PowerPoint or Google slides or Google Spreadsheets or Excel. They will generate formulas or generate slides. There's even an app called Gamma that will generate an entire presentation start to bottom. You just give it a topic, it'll give you an outline, you edit the outline, it generates the entire presentation for you. There's coding tools that automatically write code, and many of these are built on the same underlying technology as ChatGPT. You know, large language models. Yeah, but it's not. There's not just one tool. It'll be impossible to ban this in education because we won't be banning Microsoft Word and Google Docs use it in schools. So schools need to figure out how to embrace that this technology is going to be available. There's going to be tools like Khanmigo that are designed specifically for education, and then there's going to be tools that are designed for general purpose work. But schools will need to figure out that students will have access to these, whether they like it or not. There's a lot of sort of questions being raised about is AI going to replace jobs and there may be some jobs that are completely replaced. But when it comes to teaching, I'm quite sure we're not going to have less of a need for teachers globally. The world has a massive teacher shortage. If I can simply help make the job of teaching easier, reduce the sort of hardship of teaching and enable more personalized learning, we might at least get to a point where we no longer have teacher shortages. We're far from a worry of teachers losing jobs or we have such a dramatic shortage. And what I believe AI is going to do is enable more personalized learning and less busy work on the job of the teacher, more coaching and facilitating and working with that kid who has a something is blocking them or they're struggling with something or they need a motivation boost or whatnot where so much of a teacher's day is busy doing things that aren't directly with that student. AI should help us get us to a world where teachers spend more one on one time with students. And that is not about job losses in education. It's about really empowering teachers to do the work that they really got into the field to do. I couldn't agree with you more. When I think of the jobs that are safest in this generative AI world, anything that is really about that human connection and about kind of elevating and almost being the conductor and of other human beings I think are going to be a very good place to be. And generative AI is going to make that job a lot more sustainable. To your point, the reason why we have the shortage is because so much teacher burnout. If you can take the administrative things off of a teacher's plate and also give them ways to support their students better in a more differentiated way, they're going to be enjoying their work. More of what I've told a lot of teachers is imagine if all of a sudden the Department of Education or your district said, hey, good news everyone, we somehow found budget to give every one of you five teaching assistants who are just going to be there, do what you need, help you with lesson plans, help you grade papers, help you work with your students, answer questions when you're not, but they're going to report back to you and they're going to do exactly what you want. I think every teacher would celebrate. That's what's about to happen The center of AI is going to touch every industry at every possible part of it. And I do want to make clear a lot of people have started to differentiate, which I think is an important thing between jobs and tasks. I think it's going to allow a lot of folks to do more tasks more productively, but it's not going to replace, in many cases, whole whole jobs. So pretty much any job where you have to do any type of writing, I think you're going to have a productivity improvement at least 2x, probably more. Any job where you have to create presentations, you have to create any type of artifact, probably any job that involves working with a computer in any way is going to become in some way not replaced by AI but easier by AI, which means whether you're writing a code, whether you're reviewing contracts, if you're writing a patent, if you're reviewing a patent, if you're creating marketing, if you're posting to social media, if you're creating a spreadsheet or analyzing data with a spreadsheet every year, from marketing to sales to accounting, you name AI isn't going to replace your job. It'll make your job easier because certain tasks that are repetitive will become completely automated. Certain tasks where you might have writer's block AI might make the first draft for you or help you out or give you corrections to basically reduce the time and expense to get the job done. One of the things that’s important when you think about how much all digital work is going to be impacted by AI is recognizing how critical it is that our education system teaches students what this is, how to use it, how it works, and what are its risks, what are its strengths and weaknesses. Because right now most adults just treat it as a sort of magical tool oh it just came out and it's magic, but it's important to actually understand the underpinnings of what is a large language model, realizing that it's using statistics to generate its thoughts, to realize that that statistical generation is why it can make mistakes and recognizing that it's been trained on data. So the data that it's used to figure out what it's creating could have biases, could have, you know, it's missing gaps in the data, things like that. It's important for students to, as they get prepared for this workforce of the future, to know how the technology they're using in these future jobs is actually working. So one thing that a lot of non teachers don't realize, but every teacher realizes is that almost half of their time is grading papers, writing progress reports, generating lesson plans, and not in the class, not it might be happening in a classroom, but it's not facing the students. And that's above and beyond all of the issues around. In a typical classroom, you might have three or four or five grade levels of students, and so you're trying to differentiate and address all that. But generative AI, we think is going to be a huge gift to teachers. Almost everything else in edtech, even if it's very valuable. A lot of stuff we've historically done, it kind of can we think it can really enhance what's happening in the classroom. But teachers like, oh, it's just one more thing for me to learn and I'm already I'm already burnt out. But what's exciting about generative AI that let's call it ten, 15, 20 hours a week where teachers are writing lesson plans, writing progress reports, rubrics, etc., etc., grading papers that might over the next few years come from 15 hours, 20 hours a week down to a few hours a week. So that's going to save a lot of time and energy for teachers one to recharge themselves because they're getting spread too thin and also have more time and energy for their students. And that, of course, is going to help them create more differentiated instruction. It's going to be able to support them as teaching assistants. It's going to be able support the students as tutors or teaching assistants in the classroom. So they'll also be able to to personalize more for their students. Yeah, some of this is stuff that is already available and a lot of it is going to be worked into the tools that teachers already use to make grading easier or creating rubrics or even testing work against the rubrics is something that AI can now do for you. So lots of tools are going to be incorporating this into them to make teachers lives easier. Even before ChatGPT and generative AI, people are like, you know, the world is changing so fast, what the kids need to learn. And they're oftentimes surprised that I said, well, you know, it's actually the same traditional skills. If you learn to write really well, if you have a solid knowledge base, say, of civics, financial literacy, if you have a you know how to use tools well, and increasingly, these are going to be technological tools. You have strong critical thinking skills, as you know, probably at least as an algebra early algebra type type of level, some statistics, knowledge, you're going to be quite potent in any point in history. And I think that's going to be even more so in this generative AI revolution. A lot of people say, well, well, the generative AI can write papers now, the generative AI is going to be able to solve the math problem. But one way to think about it is students of the future are going to have to manage these generative AI. And if I told you that somebody is going to be an editor of a newspaper, they wouldn't have to manage writers. Does that person need to know how to write? And you're like, not only do they need to know how to write, they don't know how to write better now because they're going to be managing other people. They're going to act as an editor. So generative AI doesn't give us a free pass to say, Oh, no one has to learn any of this stuff anymore. For the folks who want to operate in the knowledge economy and want to have really well-positioned careers, they need to elevate above the generative AI and know how to manage those generative AI. So that means not just understanding the tools, but actually it means understanding the underlying skills of the generative. AI is doing better than the AI itself. if I think about the skills that are going to be most relevant for the future, it's those things that you mentioned critical thinking, problem solving, reading, writing. I also think digital skills, both knowing how to use digital tools but also actually understanding how they work, are going to be more and more important in the workplace. People also ask me all the time because I run Code.org whether coding jobs are going to disappear because AI can now write code and it's very similar to writing. AI can write essays. Do we stop teaching writing anymore? I don't think anybody is questioning whether we should stop teaching people to write because writing is a form of communication and for the exact same reason we shouldn't stop teaching coding. as you mentioned, basically the person who knows how to code just got a superpower because they can manage a team of coders that are AI just like a person who knows how to write gained a new superpower because AI could do They now have a team of writers. Yeah. No one's going to be able to. Maybe people are going to be able to create prototypes of apps without being able to code, but you're not going to be able to create a real app that you could put out into the world because you're not going to know all of the edge cases, the security holes, the performance issues that it might face. You need to know how to code. We’re in this moment right now where because of ChatGPT and large language models we're all imagining that AI is going to be the thing that we for the most part chat with. I think a lot of people are thinking about, oh well you could put text to speech and speech to text and you could start to, to talk with it the way that you might do it, Siri or Alexa. But what we're starting to see is technologies where the large language models can, even though their language models can also understand writing, and they can also generate images. So these different dimensions of generative AI are starting to converge. So it's going to be really interesting over the coming years, probably in a 3 to 5 year time horizon where you're going to, a generative AI or an AI is going to be able to make sense of students writing. It's going to be able to talk with them, it's going to be able to listen in, in conversations, and it's going to be able to make make insights about them. So it's a really interesting time where the chat, the traditional chat interface that we all know from ChatGPT, I think might just be the stepping stone to something much richer. How long how far away do you think we are from a point where an AI tutor is actually something that you can see and speak to and hear its responses? I think 2025, 2026 you'll be able to video conference with an AI tutor and it'll be pretty thoughtful. It'll be able to remember you. It'll have insights about you. I think the other trend that AI has almost leapfrogged in certain ways, but Apple now is introducing its virtual reality headset. But I have a feeling that it can take a little bit longer for that to get mainstream. That's probably more in the 5 to 10 year horizon. But yeah, in ten years I could imagine that it almost feels like you're in the room with a with an AI if you if you, if you have your goggles on. But you're going to be able to zoom with an AI and in the next three years, I don’t know about the virtual reality part just because of the weight on the head for video conferencing with an AI having a personal human like tutor I agree is on the order of five years away. It's worth it for anybody in education to think about what is the world going to be like when every student with a computer, with an Internet access actually can get a one on one tutor? It is a great story from the standpoint of equity, because today only the wealthiest families can get one on one tutoring for their kids. And now it's going to be something that not just with chatting, but even talking is going to be available. so so if I'm a teacher who's new to this world of generative AI, first I would tell them, breathe, relax. Even though it feels like the world is moving very fast. Just the fact that you are thinking about it right now means that you're still on an early, you're on the cutting edge, you know, no need, no need to worry. So especially if you have a few cycles, if it's over the summer or if it's over a holiday, start playing with these tools. You could go to something like ChatGPT and you could probably make it immediately useful to yourself, say, Hey, help me plan a lesson and help me create a rubric. Use Khanmigo, Khanmigo has special purpose, prompts and and layers in there that have the best practices of what does a good rubric look like? What is a good lesson plan look like? We have activities for teachers where they can refresh their own knowledge before they go into a lesson. It can also help create exit tickets, lesson plans, rubrics, learning objectives. And then there's a whole bunch of activities for students. Some of them are just classic tutor me and in STEM or tutor me in the humanities, Let's have a conversation with George Washington or we're about to read or we've just read the first three chapters of The Great Gatsby. We have a special guest today, kids, Jay Gatsby himself. And then you could this isn't science fiction that sounds like science fiction, but literally this exists on on Khanmigo today. And teachers can start playing with it literally right now. You can also ask AI to give you a creative way to teach a certain topic. You know, if you've had the same way, you've taught something, but you want to mix it up a little, what's a creative way to teach and then enter the name of the topic and it'll come up with ideas that might not be ones that the teachers thought of already. So it's a great tool for brainstorming with. On that point. Our most one of our most used activities by teachers already is create a lesson hook which is exactly that a lesson a lesson hook like how do I get the kids excited? And we've been impressed with how creative the AI can get. And once again, it's not about deferring the creativity to the AI. What happens is if you have two creative people in the room, it doesn't make one person less creative it makes them both more creative. So we've seen teachers riff with the and they're like, Well, what about this? And this is like, Oh yeah, and what about this? And then that gets it gets better and better is to a place that neither party would have done on their own. People talk about the risks of aid education. I actually think the biggest risk is doing nothing is just saying we're going to teach everything the exact same way. We always have them in sort of the head stand approach of just like, let's let's let's hope this technology goes away or prevent it from being used in our classrooms. The biggest risk is teaching the exact same subjects the exact same way. The real opportunity with AI is the job of teaching can become easier. The job of learning can become more engaging and more personalized and just more creative. And also some of the old stuff we may not need to teach anymore. And there's new skills that students can learn. They can learn creating with AI and develop superpowers that kids from last century didn't have. Wow. There sure is a lot of potential in AI for education. In the next session of this Professional Learning series, we will dig deeper into AI as we explore what AI is and how teachers can use generative AI for tasks such as lesson planning, creating classroom resources and saving time on administrative tasks. If you are a teacher who constantly feels like there just isn't enough time in the day, you definitely don't want to miss Session 2 Demystifying AI for Educators. In the meantime, be sure to check out our two new instructional videos. One video is all about chat bots and large language models. The other video addresses questions such as Does AI really have creativity and imagination? These videos are a great way for teachers and students to learn more about generative AI and how it works, visit the AI 101 for Teachers website at code.org/ai101. to view these videos and additional resources from Code.org, ETS, ISTE and Khan Academy. Thanks for joining us. See you again In Session 2.