Education In The Age of Amplified Intelligence
What Formula One driver training has to teach us about preparing to handle performance amplifiers
Thank you for being here. As always, these essays are free and publicly available without a paywall. If my writing is valuable to you, please share it with a friend or support me with a paid subscription.
Over the last few months, the Brake household has become enamored of Formula 1. Granted, the storytelling of the Netflix Drive to Survive documentary and the F1 movie has played a large part, but nevertheless, we are hooked. Judging by the number of F1-related tiles in my newsfeed, Facebook is on to it, too. (Brutal weekend for Ferrari and for Lando Norris in the Netherlands, by the way. Heartbreak abounds.)
There are a number of angles that fascinate me about Formula 1. Of course, there is the thrill of the race and the storylines about the talented drivers who get behind the wheel to move around the track at peak speeds north of 200 miles per hour. As an engineer, I also can’t help but love that Formula 1 is a team sport where engineering excellence in the lab is as critical to deliver pace on the track as the driver in the cockpit.
But the lens that’s been most interesting to me lately has not been fan or engineer, but educator.
AI does not democratize expertise. But it does amplify intelligence.
Much of the confusion and uncertainty in education right now comes from a misconception about how AI works. Not exactly about the machinery of AI models, but the way that we interact with them as we use them. The common perception is that AI democratizes expertise, offering pathways for novices to perform in areas where they haven’t had preexisting expertise. The truth is more complicated.
AI indeed democratizes access to something, but it isn’t expertise. Large Language Models (LLMs) give us a new way to interact with the corpus of recorded human information. Much in the same way that PageRank, the technology that powers Google, revolutionized the way we can search through information, LLMs provide a different but similarly transformative way to find our way through data. Instead of finding links to other existing pages on the web like PageRank, LLMs enable us to navigate at the level of the probability distribution, searching for likely connections that may not yet exist, instead of only returning the ones that currently exist.
Balaji made a smart point a while back that AI really stands not for Artificial Intelligence, but Amplified Intelligence. What he means is that AI tools are most effective in the hands of those who already have expertise.
If this framing is right, then the question of how to equip students to use AI is totally different from what most people think it is. Most people think you need to teach students how to use AI by having them use it. But the output of an amplifier is only as good as the input that goes into it.
What students most desperately need in the age of amplified intelligence is not some new courses about how LLMs work or a totally revamped curriculum that embraces LLMs throughout their cognitive processes. What they need is a renewed focus on the patterns of thought and habits that build true human intelligence and the wisdom to understand how to engage well with intelligence amplifiers and the distortions that are part and parcel of how they work. It might seem like a subtle point, but it has a significant influence on when and where we introduce generative AI into our teaching
The Anatomy of an Amplifier
Unsurprisingly, this problem exists in other domains. One of them is Formula One. Perhaps the way F1 drivers prepare to drive the fastest race cars in the world has something to offer us in the other areas where we are preparing students to thrive.
In the same way that an LLM is an amplifier for intelligence, a Formula One car is an amplifier for driving ability. To squeeze out every last drop of performance from the car requires an incredible amount of skill. Any small mistake is magnified and any weaknesses in the driver's racing fundamentals get exposed.
So to train kids to drive in Formula 1, we expose them as early as we can to F1 cars, right? Wrong.
If you look at the stories of almost all of the current crop of F1 drivers, you'll start to see a familiar pattern in their journeys to the F1 paddock. Almost without exception, they started out karting as kids. From Michael Schumacher, Fernando Alonso, Lewis Hamilton, to Max Verstappen, nearly all of the modern F1 drivers started this way. Why?
At first glance, you may think that karts are just a scaled-down version of F1 cars. Much less powerful, but the same basic idea. After all, they have four wheels, accelerator and brake pedals, and a steering wheel. But if you look a little closer, you'll see that the go-karts are missing several important parts of an F1 car.
For instance, when kids first start karting, the beginner karts don't have a gearbox. This is to simplify the kart and to allow the driver to focus on racing fundamentals like managing acceleration and traction, overtaking, defending position, cornering, race awareness, strategy, and mental toughness. The karts are designed to strip away some of the more technical aspects that are a part of more advanced stages of racing to narrow the focus on the fundamental skills that are the cornerstone of a successful driver.
Carrying this analogy over to the classroom and generative AI yields some insights. The question becomes: what are the fundamental skills that enable students to live a flourishing life and to prepare them to engage with whatever tools they might need to use, whether that be generative AI or something else? I'd argue that these fundamentals are exactly skills at all, at least in the traditional way we think about them. The things that we most need to teach our students, especially when they are in the earliest stages of their education, are things that are better described as virtues—character traits like curiosity, reasoning, resourcefulness, determination, integrity, honesty, teamwork, communication, and empathy, among others. The things we have them do matter, but the process by which they do them matters more. I've often found this graphic from The Jubilee Centre Framework for Character Education in Schools to be a helpful guide for thinking about character development.
The question then is not about whether we need to expose them to generative AI to help them develop AI literacy. It's not that AI literacy isn't important. It's just that it's not fundamental for our students to learn until they get close to the end of high school. Even if you think that LLMs will be a significant part of the next generation's toolkit, the foundational skills that they need to use LLMs well are not dependent on generative AI at all. Not only are the fundamental skills not dependent on using LLMs, but they can easily be actively undermined by using LLMs in the wrong way.
Innovation is always a bargain. There is no free lunch.
Whenever you think about bringing any new technology into the classroom, you've got to consider both what it will bring and what it will take away. You must consider the tradeoff and calculate whether the expected value is greater than zero. In other words, accounting for all the potential off-target effects, is this new intervention likely to be net beneficial for my students?
I can find a way for this calculus to pan out for college kids. A big part is that I think they've got the maturity and the agency to take responsibility for their own learning. They are emerging adults, and as such, we can entrust them with more and should help them to build their wisdom and discernment in more challenging circumstances. It's a classic inverted triangle that is part of any young person’s healthy development and progress to independence.
But for kids in high school and before? The risk of unwittingly helping these kids outsource their thinking without even knowing it is just too big. Of course, the normal caveats apply. This is not true for every student in every circumstance. I'm sure there are plenty of young people who can use AI well. For them, it will be an amplifier. But for students on the whole (and this is where the policy suggestions need to be aimed), generative AI will be net negative for their learning. The temptation is just too strong. No amount of string around the cookie box will work.
At the end of the day, this is where educators need to step up. Unlike our students, we do need to develop AI literacy, if for no other reason than to credibly and specifically articulate why these tools will be harmful for our students.
Some of the pushback that I've heard since originally sharing the kernel of this idea a few days ago has suggested that I am operating with a naïve perspective. Don't I know that kids have smartphones and they all have ChatGPT at their fingertips?
Yes, I get it. I know that the pressure on our students will be immense and the temptation will be overwhelming. Even more urgent is the need for us to help them carve out the space to build the skills that generative AI is threatening to erode. In the face of that temptation, our responsibility as the adults in the room is to help guide them into paths of flourishing. We need to lovingly, with care and transparency, help our students to understand that we are here to set high standards and offer high support.
Perhaps we've deluded ourselves into thinking that the battle is already lost and our students are already too far gone. I reject that conclusion. Our students are worth fighting for. It's our responsibility to help them chart a path through the choppy waters ahead.
Got a thought? Leave a comment below.
Reading Recommendations
As I was writing this piece, I was reminded of a similar one I penned a few months ago on the idea of scoot bike pedagogy. Many of the same principles exist in that metaphor, namely: match the tool we use to the core skills we are trying to teach.
David Epstein has a new book coming out soon: Inside the Box: How Constraints Make Us Better. I loved Range, so I can’t wait for this one.
I appreciated this post from Sam Barber in The Important Work.
I think that if our teachers could actually talk to us about AI and explain why it’s valuable to do the assignments on your own, we could avoid some of the risks using AI creates. I also want teachers to know that if we could exert more agency over what assignments we were doing, we might all be more interested in what we’re doing and care more about learning.
The Book Nook
I am finally getting around to Shannon Vallor’s (most recent) excellent book, The AI Mirror. In it, she uses the metaphor of AI as a mirror to help us understand both what AI is and is not, and how this shapes the way we ought to respond. I’ve been a fan of Vallor and her work for quite a while (you’ve seen me previously share her earlier book, Technology and the Virtues, which is quite good as well.
The Professor Is In
Another reminder that next Wednesday, Sept. 10, at lunchtime East Coast time (12 pm EDT), I’ll be giving a talk for the MIT Teaching + Learning Lab on generative AI through the lens of the technological critics of the twentieth century.
I’m really looking forward to putting together this talk, and hope that some (many?) of you can join. It’s open to the public, and you can register at the link below. Please share with others who may be interested!
Leisure Line
Sabbatical Friday Field Trips™ are a new feature in the Brake household this year. Last Friday, we took a trip down to Orange County to visit the Discovery Cube. We had a great time, and stopped for donuts on the way at a cool new shopping center in an old reimagined packing house building nearby.
Still Life
I had a great time speaking at Harvey Mudd’s convocation last week. While I believe the talk was recorded, I haven’t seen the video posted anywhere. There have been a few photos shared, however.
Agree with this and the framing here is considerably better than what we see elsewhere—particularly, as you note, with the intersection between K-12 and AI. (I also agree that college is the "marginal space" where more exploration and application-centered approaches seem more useful.)
However, I quick follow-up to this point: "But the output of an amplifier is only as good as the input that goes into it."
My wondering right now is if extended use of the amplifier (AI) decreases the quality of the input (human skill) itself?
Of course all research is inevitably new and ongoing, but cases like the doctors who became less skilled after relying on AI seems to present a conundrum, particularly when thinking about it with education: https://www.npr.org/sections/shots-health-news/2025/08/19/nx-s1-5506292/doctors-ai-artificial-intelligence-dependent-colonoscopy
Short-term amplification but long-term erosion feels like a bad equation, right?
I think this is a very careful and polite way of saying what I think is (finally) starting to cut through the hype cycle(s)- that the actual, ethical, successful, useful set of use cases for actually-existing-LLMs-as-AI is damn small once everyone has worn out the novelty. It's looming large in education because the work that students produce and that LLMs have proven successful in producing *is not actually a product for discerning consumers.* Student papers are expected to have mistakes! The references attached to them (that LLMs fabricate) are not necessarily there to be checked by an overwhelmed TA looking to follow their noses through the maze of scholarship, but to *practice producing references for a time when that's important.* Meanwhile, it makes coders who feel like they are going faster go slower, and all the copywriters I know have more work than ever (that they make without LLMs after the inevitable flirtations).
So while I think you're spot-on that there's not a huge place for these tools in educational environments where they screw up the traditional assessment tools and supplant skill development, I think the notion that they get to graduate to using them in all their power might be overselling the endgame.