I believe the most important area of applied AI will be in education.
In most domains, AI success looks like a cheaper version of what we already have, and failure looks like continuation of the status quo. Education is different. Change is unavoidable, and the outcomes could hardly be further apart.
This article is my attempt to explain why I hold these convictions, and how I think we should approach the future.
0.1x or 10x
What AI fundamentally does is make it possible to automate the outputs of thinking. Work that once required scarce human intelligence is now available cheaply, on demand, and at scale.
In most domains, that is almost exclusively good. The valuable thing is the output.
Education is different because the valuable thing is not the output. It is the thinking that produced the output. School is closer to a gym than to a job. You lift to get stronger, not to move the weights.
That is why the same technology that is benign in most professional domains is dangerous in education. Used as a substitute for the student’s thinking, it removes the thing education exists to do. Used to help that thinking happen, it could raise the ceiling of what education can do.
This is the 0.1x or 10x fork. The possible outcomes are unusually far apart.
If we get this wrong, students will outsource their thinking, grow dependent on the technology, and never develop the ability to work through difficult problems on their own. The troubling part is that this can look like success. Work gets finished. Only later, when they must use what they learned independently, does it become clear that nothing durable was built.
If we get this right, every student will receive a level of personalized attention education has never been able to offer. AI will understand what they know, adapt its pace and explanations, and guide them through difficulty without taking over their thinking. This could make personalized attention a standard feature of education rather than a scarce resource, helping far more students turn their potential into real capability.
Why educational outcomes matter
The cognitive skills of a nation’s population are strongly tied to its long-term economic growth. Countries scoring one standard deviation higher on cognitive tests experienced roughly two percentage points faster annual growth in GDP per capita.1 A 2025 OECD analysis similarly estimates that differences in adult skills account for around a quarter of cross-country productivity gaps, and that raising the average OECD country to the skill level of the three strongest performers could increase productivity by roughly 17 percent.2
The researchers call the aggregate cognitive skills of a population its “knowledge capital.” Like physical or financial capital, it shapes what a country can produce. But this capital is carried by people. Education builds it one generation at a time. As students enter the workforce, their ability to reason, learn, and solve problems affects how well they create technology, improve organizations, and run institutions.
Those national-level benefits are ultimately built from changes in individual lives. A study of working-age adults found that those scoring one standard deviation higher in numeracy earned 18 percent more on average. The association ranged from 12–15 percent in Nordic countries to 28 percent in the United States.3 Higher skills were also consistently associated with a greater chance of employment. These material gains provide people with greater security, independence, and choice.
But they capture only the most measurable part of education’s value. To become capable is to gain greater agency in your own life. It means being able to understand the world, form your own judgments, solve difficult problems, and pursue goals that would otherwise remain out of reach. Higher earnings are one consequence of that capability. The deeper benefit is the freedom and fulfillment that come from being able to act competently in the world.
This is why the 0.1x or 10x fork matters so much. AI will not only impact how students complete schoolwork. It could shape the capabilities they carry through the rest of their lives and, collectively, what society can achieve.
The current trajectory
The most prevalent attitude I see among educators is palpable skepticism and even doomerism around AI.
This is a very understandable backlash. Many educators have already lived through a wave of digitalization that treated technology as an end in itself. That approach was damaging and largely failed to deliver on its promises. And now they see their own students using AI tools like ChatGPT in ways that harm their learning. Furthermore, many intuit that fundamental skills like reading and writing should be protected from technological interference.
Where they go wrong is in their approach to the future. They misread the cards they have been dealt. Already, two-thirds of Swedish high-school students use ChatGPT.4 Students will use AI whether schools want them to or not. There is no going back to a world in which they don’t.
More importantly, even if we could freeze the pre-AI model, we shouldn’t want to. Education is not all that it could be. It’s important to recognize that there is room for improvement. As an example, roughly one in five students in Sweden fail to graduate from high school.5 Used well, AI could be integrated in ways that radically improve what education can do.
If this attitude prevails, we get the worst possible future. Left to their own devices, students will keep using AI in ways that harm their learning. Schools will have surrendered the opportunity to shape a technology that will define how a generation learns, leaving that power to companies with no incentive to care whether students actually learn.
The path forward
The good news is there is another way. We don't have to resign ourselves to fatalism. We have the agency to build a future that incorporates these tools in a positive way and avoids the negative ones. But it has to be pursued. It will not happen automatically.
I believe the appropriate approach starts from the premise that AI is a paradigm-shifting innovation, yet still just a tool like any other technology, not an end in itself. Our task is to discern where it should be integrated into existing structures, and where it is best left completely out of education. Every decision needs to be judged against the impact on the student's long-term educational success.
In practice, this would look like teachers remaining in charge of each student's educational journey, with AI assisting them and each student: seeing where they are stuck, guiding them through difficulty, and not taking over the thinking. It means students having clear boundaries around usage of AI, firmly imposed by educational institutions. It means the companies building the AI are accountable to the school itself.
Taking this path, we ground everything in the impact on the student. We don't get carried away with the promise of the technology or stick our heads in the sand in fear of all that could go wrong. It takes the concerns of the pessimists seriously, while keeping an optimistic, problem-solving attitude toward the future, believing that radical improvement is possible.
Why education matters most
This is why education matters more than any other area of applied AI. The possible outcomes are unusually far apart, their consequences will compound across generations, and adoption is already unavoidable. Yet the direction is not predetermined. We still have the opportunity and responsibility to shape this technology, so that the tools available to students are ones that help them learn.
Sources
- Education and Economic Growth — Eric A. Hanushek and Ludger Woessmann, Oxford Research Encyclopedia of Economics and Finance. Across 50 countries, one standard deviation higher cognitive test performance was associated with roughly two percentage points faster annual growth in GDP per capita.
- Adult Skills and Productivity: New Evidence from PIAAC 2023 — OECD Economics Department Working Paper No. 1834. Estimates that adult skill differences account for around a quarter of cross-country productivity gaps and that closing the average OECD skills gap with the three strongest performers could raise productivity by roughly 17 percent.
- Returns to Skills Around the World: Evidence from PIAAC — Hanushek et al., European Economic Review. Across 23 countries, one standard deviation higher numeracy among prime-age workers was associated with 18 percent higher wages on average and a greater probability of employment.
- Svenskarna och internet 2025 — Internetstiftelsen. Survey finding: 2 of 3 Swedish gymnasium students have used ChatGPT.
- Betyg och studieresultat i gymnasieskolan år 2025 — Skolverket. In 2025, 82.6 percent of students on university-preparatory programs and 75.9 percent on vocational programs completed a gymnasium diploma within three years.