The CNS sat down for an interview with the Founder and CEO of Khan Academy and Vision Steward of TED, Sal Khan, to explore the intersection of his work in education and AI with neurosurgery. Sal Khan will speak on Monday, November 2, during the General Scientific Session at the CNS Annual Meeting.
In neurosurgery, as in other specialized professions, overreliance on AI carries risk. How should surgeons balance AI guidance with human judgment? What guardrails would ensure that AI enhances rather than erodes, clinical reasoning and expertise?
Well, I think what we're seeing is folks who engage in AI, and frankly, even folks who engaged in pre-AI technologies just doing web searches, but if they didn't have the critical thinking skills or underlying knowledge, they will over-defer to these outside sources and maybe sometimes be led in the wrong directions, or they might not be able to ask the right questions. I think it's super important, that people have the underlying background knowledge, muscle of clinical decision making, and the skill to push the AI further. I think the combination of the human plus the AI will likely get better outcomes, but we don’t want to over-rely. I think now and, in the future, we need to make clinicians aware of how AI can oftentimes reinforce what your inclinations already are. They can hallucinate. As a decision maker, you need to push it. You really need to make it a skeptic. You need it to push your thinking, not just validate it.
As AI becomes more present in the learning environment, what questions or concerns do you have about its impact on the developing brain, particularly attention and social cognition?
A lot of the debates now on screen time focus on whether AI is good or bad, but I think it's actually been the wrong debate. Instead of saying AI: good or bad, what we should be saying is: cognitive offloading versus cognitive onloading. Even before AI came into the picture, if you had a teacher, parent, or tutor who does your homework for you, that's going to be cognitive offloading and that would be really bad for you. But, on the other hand, if your teacher, parent, or tutor is there to support you, they’ll push your thinking. So, let's say there's a multiple-choice question, but they say, “Okay, you got it, you got it right. But can you explain how you got it? Can you apply this generally? Can you say how this connects to something else?” Well, just as you could do that with a human being, you can now do some of those things in a scalable way with artificial intelligence and an AI that pushes you further than a multiple-choice question does. That makes you really question your assumption and makes sure that you can understand the concept. That's going to be cognitive onloading, not cognitive offloading.
So, everything that we're doing at Khan Academy is to try to use AI to broaden the aperture of one: how you could support someone that makes them think harder, not think less. We also need to put guardrails in place to avoid the cheating use cases, which have become easier with AI. We don’t see this on Khan Academy, but we know there's a whole series of AI tools out there that make it very easy. We're also getting into the act of policing those other things a lot more. But we’re thinking about how to use AI to drive higher cognitive engagement, not lower. So, it's not AI: good or bad. AI can take us in either direction. It's what we as educators and clinicians decide to do with it.
You've used the TED stage to introduce big, forward-looking ideas about AI and education to a global audience. How has that platform changed the way you think about communicating complex technological shifts, not just to educators, but to the broader public?
Well, I think TED talks and the various tools that it has are valuable. I think it is a very high-level idea platform. TED Ed does a good job of breaking down everyday topics more deeply. Khan Academy focuses on the more academic or going one layer deeper than you would get from an “edutainment” type of video. We're also big believers that you need to practice that concept. So, my hope is that TED and Khan Academy can continue to play a major role here. What's interesting about AI is that used well, it can also help introduce people. I think the main barrier for a lot of folks is they aren't asking the right questions, or they're not realizing that they can get AI to actually teach them the things that they need to learn, or they don't know whether those are things that they need to learn. I think there's going to be value in articles, exercises and videos, but also in packaging some of the things that we know are going to be valuable, and it makes it much easier for folks to learn how to use AI.
Neurosurgery demands deep mastery over long training arcs. How does the concept of mastery learning translate to procedural skill acquisition in medical education?
Well, I've got to believe that neurosurgeons have been practicing mastery-based learning from the get-go, because if there's a neurosurgeon who's doing a surgery eighty percent well, I don't think that that's going to be acceptable. You need to keep practicing that, keep getting feedback on that until you get that to 99.999% well. All we've advocated for in everything from neurosurgery, all the way to learning your algebra, is to have that same mindset. If you haven't practiced, if you haven't mastered the more basic thing, it's even harder to, maybe even impossible to, master the most advanced things. So, I think mastery learning is essential in medical education, especially if we think that these are things that people are going to do day in, day out, and lives are at stake.
What insights from learning science can help optimize decision making in high stakes environments like the operating room?
My understanding is surgeons already are keeping stats about complication rates or when the surgery doesn't go right. So, there is a lot of clear feedback in the profession. I think learning science will tell you that if you're getting a lot of practice, you're getting immediate feedback. I would assume that when you're getting the feedback and it doesn't go as expected, you reflect on that. Well, that's when you're going to grow those skills. You know more about neurosurgery than I do, but I believe and hope that given the stakes in your field, you all have already been practicing this, and it's we who want to take some of that and bring it into the rest of the education system.
How might AI help identify and remediate potential knowledge gaps in expert clinicians, building on how it supports students with foundational gaps?
If you're a physician, it would never hurt to test your ideas with a high-quality AI. We already have AI's that can automatically take notes, etc., etc. So, it's not hard. We have AIs that are observing or can hear the clinical interactions. If HIPAA allows it, maybe even observe the clinical reactions interactions. If they are, then it would be a very powerful sounding board. Now, I don't think I or anyone else would say defer to the AI, but after the patient leaves, the AI could say, “Hey, you know, I saw that; that was pretty good. But have you thought about this?” Or, “You mentioned this one thing. I'm not sure if that's exactly right. Let's do some research on it together.” Or maybe even during the clinical interaction, the AI is whispering into the surgeon or physician's ear and saying, “Hey, have you thought about this? This might be another interesting angle to take it.” So, the important thing is it's not replacing the clinician's judgment. It's not replacing the knowledge, but it's another expert in the room. And the AI will be good at knowing what the latest research is. If there's some study, it might be able to pick up on some really weird disease or edge case that a surgeon either has never seen or thinks is a very low likelihood, but it's still it's going to happen to someone, even if it's a one in a million or one in ten million type of thing. I think the combination is what's very valuable.
You've often referenced the dramatic impact of one-on-one tutoring. To what extent do you think AI driven tutoring could intentionally engage the brain's capacity for neuroplasticity?
I think, and there's research that backs this up, we learn the most when we are mentally engaged and forced to use multiple muscles. Maybe it’s communicate orally, maybe you have to draw something, or explain your reasoning. Then when you get immediate feedback and reflect on it, you’ll use that in order to improve your performance on whatever it is going forward. There's good literature that these kinds of interactive sessions are much more effective than just passive learning or even just a question and a response interaction, but to keep going on and on, deeper. And that is one of the benefits of one-on-one tutoring. I think there's another human benefit of one-on-one tutoring that might take a long time for the AI to get close to, which is the knowledge that there's another human being that's taking time out and believes in you. And that's if you have a good tutor! Hopefully, AI can do some of that, but I think it's just meeting you where you are at your learning edge, making you do things, cognitively engage as opposed to just giving you a lecture.
Khanmigo responds to learners in real time, identifying misconceptions as they arise. Do you envision a future in which such systems also serve as meaningful tools for assessing cognitive function more broadly?
Yeah, I hadn't thought about it until this question, but yes, I think that's a very powerful idea. We have thought about using AI to broaden the aperture of things you can assess. And I'm sure that there are neurocognitive assessments that you all have that go much beyond multiple choice tests or things that you would traditionally be able to do at scale. But now with AI, you can have chat interfaces. You can now have conversational voice interactions. And I'm guessing we're very close. In fact, there are probably already specialized systems that can do real time conversational voice and image based, so you can even observe the person's mannerisms or their eye contact and things like that. So, the simple answer is yes, I think, and that could be very, very valuable.
Looking beyond traditional education, do you see a role for AI tutors in supporting patients recovering from neurological events, such as helping to rebuild language or memory after surgery?
Absolutely. I've gotten interesting letters over the years from folks who've had strokes or other types of issues, and where they had to relearn a lot of material. This is more on the academic side, and they were using Khan Academy to do that well before we had Khanmigo and AI. But I see now that with AI, you can just, as I've been saying, really broaden the types of things that you can develop.
What's a key takeaway you hope the neurosurgical community gains from your session at the CNS annual meeting?
Well, I hope that people learn AI is definitely not a silver bullet for everything, but it's also not something to run away from. Many people are increasingly aware of all the ways that AI can be bad in the broader world, in education, etc., but there's also a lot of power behind it to everything we've just been talking about. So, I hope people appreciate there's power behind it. I hope they appreciate that this is a good time to be leaning into what is possible with these technologies, trying things out in a safe, effective, ethical way, and get the best out of it. A lot of what I say is that technology just amplifies human intent. If you have negative intent, if you want to cheat, if you want to defraud someone, if you want to spread misinformation, yes, AI and other technologies will amplify that. But if you have positive intent, want to educate, rehabilitate, and get better at your craft, technology and specialty, AI could have some superpowers for you.
Want to hear more about AI, education, and neurosurgery? Join your peers at the 2026 CNS Annual Meeting in Washington, D.C., October 31-November 4, 2026. Register by September 30 to save on your registration and housing!