Sal Khan on AI in the Classroom
The author and founder of Khan Academy chimes in on the debate surrounding LLMs and computer science education.
After our previous post on computer science education in the era of large language models (LLMs) elicited some good-spirited debate within our community, we decided to ask our friend Sal Khan for his thoughts on the matter (as a reminder, Sal was Somak’s college roommate at MIT, and we had the pleasure of interviewing him at our annual meeting in the early days of the pandemic). Is a degree in computer science still a safe bet when it comes to the job market? How might AI rebalance (or further unbalance) our STEM-dominant education system? What aptitudes should today’s students focus on cultivating?
As a pioneer in the virtual learning space in the early 2000s, Sal is well-positioned to prognosticate on this topic. Decades after his groundbreaking instructional videos placed him well ahead of the curve of virtual pedagogy, he is now the author of Brave New Words: How AI Will Revolutionize Education (and Why That’s a Good Thing), forthcoming from Viking this May. Here’s what he had to tell us:
ASV: How will AI transform computer science education?
Sal Khan: No one can predict for sure. I won't claim that I have the perfect crystal ball, but I'll cite a trend that I experienced in the early part of my career. In the late ’90s, early 2000s, there was a lot of offshoring, especially to places like India. And a lot of what people were telling young software engineering grads like me at the time was, Go find something else to do, because all of this coding work is going to go — probably to India or to another lower-cost place. And salaries are going to go down or maybe the jobs are going to disappear entirely from the United States.
We know that what actually happened was the exact opposite.
Some of the integration work, data processing work, let's call it the “grunt work,” did go offshore, but in no way did that shrink the pie for what was happening here in the United States. If anything, over that same time period the average salary — even adjusted for inflation — for people with high-caliber software engineering skills (especially those who could combine it with creativity and design and leadership) has gone through the roof. They're even more valued than before.
I see a similar trend happening because of AI. Yes, it will be able to take out the grunt work (or some of the grunt work), and every subsequent generation of AI tools is going to be able to do more and more of that. But it's still going to take thoughtful, creative people to be able to put the pieces together, to be able to put it out into the world and make it relevant. It will drive value toward those who can play multiple roles, those who can think technically and understand how a system works, but also those who can lead, who can design, who can figure out how to go to market, who can sell...
I think that the likely trend is more one of concentration, with multiple functions being concentrated in highly skilled individuals rather than needing armies of software engineers like you might have needed in the past to do certain tasks. So with respect to computer science education: students should learn to use these tools, and there's no shame in that. I remember when I was a computer science major in the ’90s, even though no one was actually coding in machine language or assembly, we still learned those things because our teachers said, Well, you still have to know how that kind of stuff works. Some of the old-school professors viewed coding these higher-level, abstract languages as almost cheating compared to what they had to do with punch cards and whatever else in the ’60s and ’70s. I think you're going to see something similar when it comes to using one of these AIs as a copilot. It's going to be a huge productivity tool for anyone, but someone who can think structured thoughts and think through how a system works is going to do really well.
Maybe I'm one of these old fogies who thinks it's still useful to learn how to code, but we never want to get to a Matrix situation where people have forgotten and the machines have taken over. We still need to know exactly what the machines are writing and understand it.
In your view, is our STEM-dominant educational system due for a revamp?
SK: Yes and no. Even before this AI trend, there's ways that I would have modified the education system. Now, I don't know if I would make it less STEM-dominant or more STEM-dominant, but I think there are a lot of practical life skills that are very empowering that are absent from most students’ educations unless they go to some kind of professional school. And this isn't necessarily STEM-related — but actually it is STEM-related, because if you're starting a business, if you're starting a startup, if you're trying to get a patent, all of these things matter tremendously.
Things like business skills, accounting, finance, capital markets, legal skills, understanding how the system works — it's amazing how little of that is in the school system today. My best guess is that you're going to see more and more money, more and more careers, almost every career is going to be quasi-STEM, and a lot of the remunerative careers in the future (even if you go work for a law firm) you’re going to be asking: How can you use these tools to really streamline workflows?
It's going to happen in every industry. I don't see the tech “gold rush” ending, in any way, shape, or form. If anything, I think it's going to accelerate because of AI.
I would say one trend that we've already been seeing over the last decade or two decades, which I think will also accelerate, is that it's not so much about the specific coursework. Whether it's reading Shakespeare or learning how to code, really, it's a proxy for your ability to think critically, show up, and do your assignments on time. What's going to be valued is experience using these tools, especially when these tools are developing so much faster than any of these traditional curricula are going to have time to adapt to.
So I think the young people who are going out there and playing with frontier tools and are putting them to good use, regardless of what they major in or whether they go to college at all, are going to be the ones who really thrive in this environment.
Do you see the line between STEM and non-STEM majors blurring?
SK: I still see, even in my own workplace, that it's not necessarily someone's major that tells you how strong they are at structured thinking. It's not like, Oh, if you major in STEM, you're going to be really good at thinking through our product spec, and if you majored in humanities, you're not. I don't think that's the case.
Sure, there is probably a correlation between structured thinking and people who have had to go through the discipline of engineering something. When you engineer something, whether it's industrial, mechanical engineering, or software engineering, you’re trying to put an idea into practice and having to think through all of these edge cases and all of these tweaks that, let's call it a “normal person,” when they use a product, software or hardware, doesn’t realize how many edge cases have to be considered for that product. I think it's that capability — being able to not only create a system but to account for all of the edge cases that it needs to support and then tweak it and debug it — that is going to remain a really valuable skill.
At Khan Academy, we're doing a lot of prompt engineering, and prompt engineering in its current incarnation is primarily just writing stuff. But we are definitely seeing a difference between people who have, say, a software engineering background and who can think about how the large language models work and those who don’t and can’t. At least to our observation, if they have that engineering background and they have some type of a content depth in whatever we're trying to prompt for, it’s ideal to have both. This is going to be a world where, as I said earlier, the master of many skill sets is going to win.
That's my best guess of where people are going to survive or do well. What I'm going to tell my kids and the kids at Khan Lab School is: Play with these tools. It's helpful to have a major where you’re signaling that you clearly have strong engineering or quantitative skills, but it doesn't necessarily have to be an engineering major. If you can double major, there could be some other practical skills that you could develop. Now, if you have a passion for literature or the arts, that's great, too. But the important thing is to play with these things. The people who will do well are the people who are able to straddle multiple fields.
