Learning in the Age of AI: Hot Takes from Stanford's Dean of Education
How AI will change learning forever.
Growing up, I was a good student — straight A’s — but I hated school. I was always the kid raising her hand and asking, “Why do I even need to know this? What’s the point?” There was only one thing that motivated me to get good grades: I wanted to go to Stanford. When I ended up getting accepted, it was the most phenomenal experience.
Recently, I had another phenomenal experience: sitting down with Dan Schwartz, Dean of the Graduate School of Education at Stanford University (the #1 globally ranked education department), to talk about the most pressing questions in education today.
Where is the school system headed? What is the role of AI in the classroom? What does the learning science say? Where do parents fit in?
You can watch the full conversation here — or, you can settle in, pour a coffee, and read through the hottest takes of our conversation.
“I turned out fine” is the enemy of true educational progress.
Listening to a 90-minute lecture is scientifically one of the worst ways to learn new information. But the common response is: “Well, I learned that way and I turned out fine.”
We’ve all heard it. Some of us have said it ourselves. But Dean Schwartz skewers this idea:
“What no one is going to tell you is that you could’ve turned out better.”
Education is filled with material that has been handed down just because. For example, word problems. Who invented them? How effective are they, really? We’re not totally sure. But there they are. The same is true for much of traditional education. Because it’s “traditional,” we assume it’s true. That’s not really the case.
In the last thirty years, learning has become something that we study. (And the discoveries made have built all of the new AI systems, by the way.) One of the most interesting discoveries we’ve made is that learning is not a singular thing. The way you learn your values is different from the way you learn your multiplication tables. They’re entirely different systems. The trick is figuring out 1) the right instruction to fit the appetite of a certain kind of learning, and 2) the right kind of learning for the outcome you want.
Unfortunately, it's a huge challenge to actually implement educational progress into the system. Why? Because everybody always falls back on the age-old excuse: “I learned that way and I turned out fine.” True educational progress is locked in constant combat with complacency, nostalgia, and defensiveness. For Dean Schwartz, this is why AI is so promising.
“Everybody wants to use AI. Everybody is going to adopt it. Which is why it should be the exact place we put the learning science.”
AI should help students make things, not automate bad teaching.
When it comes to using AI in the classroom, there are two paths we can go down.
One path is digitizing the school day. And to be honest, this is worst case scenario. We don’t want to automate outdated, lecture-based instructional methods. And we certainly don’t want to drop iPads or chatbots (otherwise known as “cheatbots”) into every student’s hands and say, “Here you go! Let me know when you finish high school!”
Another path (the fruitful one!) is using AI to create things. When Dean Schwartz toured Alpha School for the first time, he saw a young student working with AI to write songs that helped her understand and memorize U.S. history.
“This is a great example of how students are using AI to make stuff and augment their abilities. Rather than replacing teachers in a system that doesn’t work, this is a chance to do something different. You learn a lot by making stuff.”
The real opportunity of AI lies in helping students create, like writing songs to learn history or building architectural plans they can virtually experience. AI allows us to expand the threshold of learning.
What the microscope did for biology, AI will do for education.
One of the great challenges of education is lack of data. Traditional schools assess students with grades and a standardized test once a year, which doesn’t lead to a ton of information. Outside of an eight-week report card, parents have no idea what, or how well, their kids are actually learning.
But hey — you know what’s exceptionally gifted at collecting, analyzing, and extracting meaningful insights from mass quantities of raw data? You guessed it. Not only can AI tutors accurately assess and hole-fill students’ education, they can share that data with parents in real-time.
Crazy thing is, this is actually old news. About 25 years ago, they made AI tutors for areas like algebra — things that had a simple rule base. If you missed a question, the AI tutor could figure out what went wrong and where, based on how long it took you to answer other questions. Say a student struggled to figure out the area outside of a square. The AI tutor would analyze the data and reveal the blindspot: “Oh, that’s because the student never mastered perimeter.”
AI has been doing this for years. New AI has just made it faster, easier, and more effective. We can now zoom in on every student’s needs in ways we’ve never been able to before. Continuous, fine-grained, real-time insights into what your kid is learning should be the new standard.
AI could have a big-time positive impact on kids with disabilities.
In our conversation, Dean Schwartz tells the story of one of his colleagues, a pediatrician, who trained Google Glass to recognize facial expressions. It can identify an expression as sad, happy, angry, bored. So, how does this help children on the spectrum? When wearing this glass, children can see the emotion that someone else is feeling. This then gives them the confidence to engage in conversation. And once they engage socially, they start to develop neural pathways.
His colleague has since collected huge quantities of data through Google Glass — and he's made a game out of it. It goes like this:
Parents hold a phone up to their forehead, where their three-year-old can see the emotion. Kids then have to demonstrate the emotion like charades. Based on their reaction, the phone can detect if the child has a neural disability — and then it can start changing the content. It will begin to focus on important social behaviors: for instance, making eye contact.
What’s so amazing about this, you ask? The AI identifies behavioral issues. You generally wouldn’t know your kid has autism until they're five or six years old. And by that time, the neural pathways have already settled down. The mission of this “game” is to help kids who would have otherwise not been able to participate in society be able to participate with confidence. (He estimates it to make about a 40% difference.)
And this is just one way we can use AI to radically impact these kinds of situations.
Schools need two things to adopt AI-powered personalized learning plans:
Some technologies work their way into schools fast. Just look at YouTube. If a classroom has a laptop and a projector, then more than likely, that teacher uses YouTube. And why wouldn’t they? You could show students a grainy, one-dimensional picture of space in a textbook; or, you could show them a video of planets orbiting the sun from different angles in motion. There’s no denying that the face value of using YouTube in the classroom is high.
The question is, can schools accommodate AI like they do YouTube?
It depends on two things:
Willingness. If schools view AI like YouTube, where people recognize it as an immediate value-add for education, then most schools will probably get on board.
Structure. The infrastructure of the school day must support the technology, and vice versa. For instance, Alpha School is set up to enable different learning speeds for every student — helping them catch up and get ahead — but not all schools work like that. Large public schools in urban districts have a completely different infrastructure. We can’t just drop iPads and AI into the classroom and expect the same results. The structure must support the technology.
AI cannot appreciate the Mona Lisa for you.
Why do we learn anything to begin with? To appreciate life! It may sound oversimplified, but it’s true. The more you know about the world around you, the more you can appreciate it. Knowledge begets enjoyment. And you cannot outsource appreciation. No matter how advanced technology may become, it will always be important to learn, educate yourself, and fill your head with facts. AI enhances knowledge; it does not replace it.
“Say you go to a museum and there’s this great painting by Van Gogh. It’s not like you want to put the AI in front of it and say, ‘Appreciate it for me.’ You need knowledge to be able to appreciate it.”
Learning is so much more than mere information consumption. It is a cultivation of judgment, taste, style, and depth. You need knowledge to fully understand and appreciate the world around you. That is the heart of education. AI cannot admire the Mona Lisa for you. Only you, equipped with relevant knowledge and facts, can do that.
Ultimately, students still need foundational knowledge to interpret and appreciate AI's outputs, but they can learn more efficiently through AI-powered personalization.
These are just a few highlights of my conversation with Dean Schwartz — if you want to dive into 65 minutes of spiky educational conversation, you can watch the full video here.
There’s much I appreciate about your work, including the argument in this essay that “turning out fine” impedes progress in education. But you also include an example that exemplifies my caution with your embrace of this technology: using AI to interpret emotions is terrifying, and promoting it as a tool for children is even more so.
AI for emotion recognition is one of those ideas that needs a team of people, perhaps the students in your school, to imagine all of the benefits and harms of this technology. Obviously this tech is already here, so we’re not going to stop it, but people need to be aware and think through its use - if it can be used to increase power and money for those who already have power and money, it will be. A good lesson is to let your students use their own thinking (not AI) to consider the range of possibilities.
A great post, Mackenzie.
I'm in complete alignment with your perspective and I'm so happy to see you reference personalized learning plans. I believe that these are our future, starting yesterday.