In April I attended the “AI Revolution” (AIR) Show that happened in conjunction with the ASU+GSV Summit. The event painted a futuristic vision of classrooms and schools supercharged by AI. 

And I was… not impressed by the visions most companies were painting.

I’ve attended conferences like these for over a decade and seen time and again edtech promising transformational impact only to fall far short.

Why is this? Because of  the nature of edtech business models and the gravitational pull of conventional schooling.

AI won’t transform schools

Put yourself in the shoes of a company trying to develop AI-powered educational solutions. Your company starts off aiming to help students and teachers unlock self-paced, mastery-based learning by using AI to quickly generate individualized learning plans aligned with learning goals. 

Or maybe you aim to facilitate more real-world learning by using AI to assess students’ mastery of state standards and core competencies across a wide range of in-school and out-of-school learning experiences. These technologies could revolutionize education.

But then you have a problem: most schools aren’t interested in what you’re selling. They want products that are easy to fold into their current classrooms, not products that require a massive organizational redesign. 

A few interesting schools on the margins might be highly aligned with the tools you’re trying to develop, but those schools are small and hard to find within the vast K–12 landscape. 

Meanwhile, your business needs to grow fast to cover its costs and give your investors their expected returns. So what do you do? You sell your products to the large and well-known market of conventional schools. 

Unfortunately, this means your solutions get redesigned to conform to convention rather than transforming education. 

If your business model requires rapid scale, you can only sell products that enhance conventional schools’ operations, not products that reimagine schooling. 

Ultimately, your AI may only lead to marginal worthwhile enhancements in how teachers do their jobs—helping them brainstorm lesson ideas, write lesson plans, or grade student assignments. 

And your original vision of using AI to change single-paced, seat-time-based, teacher-directed instruction? Not going to happen because that’s not what conventional schools are hiring your product to do.

Nonetheless, AI does enable us to reimagine education

While I’m skeptical about the impact of most of the hot and trending education AI, I’m quite optimistic that AI will play a significant role in a broader education transformation already underway. To see this broader trend, we first need to step back and consider how people learn and how technology can enable a fundamental reimagining of formal education.

Most people today believe that very few students are autodidacts. What’s an autodidact? It’s a fancy label for someone who is “self-taught.” 

The commonly-held view is that there is a small subset of the human population whose combination of personality and intelligence wires them for autodidactic learning—people like Benjamin Franklin, Frederick Douglas, Bill Gates, and Mark Zuckerberg. 

Meanwhile, popular opinion and education experts assert that most people are not autodidacts. Therefore, most students will only reach academic achievement through direct instruction from teachers—the kind that conventional schools have been doing for decades.

At first glance, this notion seems to hold. Regarding academic content societies aim to impart through formal education, most students seem to require teacher-directed instruction. Expecting students to learn the core academic subjects through their own devices seems like a recipe for failure.

But if we look beyond the walls of classrooms, it’s clear that nearly all young people are autodidacts. Kids across the world today are directing their own learning to figure out things like how to do tricks on a skateboard, solve a Rubix cube, draw cartoon characters, perform soccer-ball-handling stunts, apply obscure rules in sporting events, memorize and analyze historical events in fictional worlds, or build contraptions in Minecraft. And a vast amount of that learning happens without any formal direct instruction. 

Young people’s brains are wired to learn. Learning has been essential for human survival for millennia before formal education, teacher-directed classrooms, and academic standards were created and propagated. 

In other words, all young people can be autodidacts. It isn’t a matter of personality or intelligence. It’s a matter of motivation. 

The real problem is that most kids aren’t motivated to learn academic content outside of school. 

What motivates learning? 

So, if all students can be self-directed when sufficiently motivated, what is it that motivates students to learn? According to prominent theories in learning science, motivation has two main components: subjective value and expectancy. 

The subjective value represents a goal’s importance to the student, and it can come in different forms. For example, a student may enjoy getting an “A” on a test because of the satisfaction that comes from successfully completing a goal (attainment value) but enjoy working on an art or engineering project because of the innate satisfaction that comes from being creative and solving problems (intrinsic value). 

Some goals are valuable to students because they serve as a means to an end, such as when a student works to bring home a good report card so that he can get a bump in his allowance (instrumental value, commonly known as extrinsic motivation). 

In reality, many goals derive value from multiple sources. The more a goal has subjective value, the more a student will be motivated to work toward that goal. When students have competing goals, those with the greatest subjective value will usually get the most attention and effort.

Expectancy is the other important component of motivation. It is a student’s belief that he is capable of successfully achieving a goal. In short, if a student is considering investing time and mental effort into a goal, the student needs to feel a reasonable degree of confidence that he can reach the goal. 

Those expectations usually depend on two things: 1) Does the student have confidence in the path that leads to the goal (outcome expectancies)? 2) Does the student have confidence in her ability to pursue that path (efficacy expectancies)?

Motivation comes from the product of both subjective value and expectancy. We can picture this by representing each of these elements of motivation on the perpendicular axes of a two-dimensional graph. 

On the graph, there’s a threshold (represented by the downward-sloping curved region) that separates feeling unmotivated from feeling motivated. At any point on the graph below the curved region, the combination of value and expectancy is insufficient to motivate the student to take action toward the goal. But when increases in value and/or expectancy cross the threshold, the student feels motivated and works to make the goal a reality.

To see the theory in action, consider a hypothetical example. Picture a student who doesn’t see math’s relevance in his life but does care about getting good grades. His subjective value of math is moderate, so he tries to do the assignments he’s given. Unfortunately, for a few weeks in a row, he consistently gets stuck on homework problems. He spends over an hour working on each assignment but feels like he gets nowhere. With each evening of frustration, his expectancy drops. Before long, his motivation wanes, and he stops doing his math homework altogether. 

But then suppose he learns about free tutoring through With the help of his online tutor, dead ends turn into paths forward. With his newfound expectancy of being able to complete his math homework successfully, his motivation goes up, and so does his grade. By the end of the semester, he’s getting one of the best grades in his class. As other classmates start turning to him for help, he finds an additional source of subjective value, increasing his motivation even more. 

In short, a student will pursue a learning goal when he values the outcome he gets from learning AND when he feels confident that her effort to learn will pay off. 

Why are academic autodidacts rare?

So, why are so few students autodidacts with academic content? Because for most students, teacher-directed classrooms are the only places where subjective value and expectancy are high enough to motivate academic learning. 

For most students, academic subjects like math, history, and literature don’t seem to have a lot of value in life outside of school. However, teachers and schools have a few levers that boost the subjective value side of learning. 

In the best versions of motivation, skilled teachers make academic subjects interesting and exciting, and hence, the subjective value of learning those subjects goes up. In other cases, students want to please their parents or other respected adults, and teachers tie learning goals to report card grades that those adults expect kids to earn, so the institution of formal schooling increases subjective value. Additionally, when looking smart in the social context of school helps some students gain identity and status among their peers, subjective value increases. 

Similarly, teachers play a big role in boosting expectancy. Most teachers give their students a clear path for achieving learning goals. They show students the steps for solving problems or finding answers, they provide step-by-step instructions for completing assignments, and provide rubrics that clearly articulate how grades will be assigned. 

Good teachers also make learning feel more doable through clear explanations, thoughtful scaffolding, etc. And for students with learning disabilities, the pedagogical expertise of a teacher can be the make-or-break difference on the expectancy scale. When the path to learning goals is clear and the work of learning is made easier, expectancy goes up.

In other words, the conventional schools most people are familiar with have levers to pull that boost students’ motivation to learn academic content. 

Outside of schools, those motivation boosters for academic learning are more rare. And because teacher-directed instruction is the norm in most schools, direct instruction from teachers seems essential for getting students to learn academic content. But this strong correlation does not mean that teacher-directed instruction is the only effective way to learn academic content. 

Unlocking self-directed learning

For all the reasons described above, teacher-directed learning has for more than a century been the optimal model of schooling. But all that has been gradually changing.

Fast forward to the last half century and we’ve seen an explosion of technologies that make  self-directed learning an increasingly accessible modality for learning academic content. These include innovations like audiobooks—which make learning from books far easier for people with various forms of dyslexia, thereby making learning more inclusive. Or consider video, which can leverage multimodal presentation of academic content—spoken word, sounds, animations, and real-world demonstrations—to make concepts more intuitive and memorable. 

Next consider how the internet has put a huge array of learning materials at our fingertips—be they text, audio, video, or any other digitizable format. Wikipedia provides thorough and digestible summaries on practically any topic imaginable. YouTube creates a market that pushes the most shared and sought-after educational videos to the top of search results. 

Now add to the list technologies that provide dynamic feedback to learners—from the spell checkers in word processors to platforms that instantly grade math problems and multiple choice questions. 

Next, consider adaptive learning software that can assess a student’s current level of understanding and then direct the student to learning activities tailored to their zone of proximal development. 

Now, consider what self-directed learning can look like with AI added. Instead of reading a long text on a particular subject, you can dialogue with the AI and get straight to the new knowledge that is on the frontier of your current understanding. 

Suppose you want to learn a new topic but don’t know where to start. With a simple prompt, AI can create a custom-made syllabus of learning activities to guide you to your goal and point you to other valuable learning resources you didn’t even know to look for. 

AI can also be a good tool for checking your understanding of topics you’re trying to learn. Explain what you think you understand to a chatbot (or just dialogue verbally with the chatbot using speech-to-text and text-to-speech features), and you can quickly build your confidence in what you do know, identify the gaps in your understanding, and get quick tutorials on topics you haven’t mastered yet. 

Do you need help understanding where you went wrong on a math problem? Ask a chatbot. Do you need feedback on an essay you’re writing? Ask a chatbot. AI is an incredibly powerful tool for supporting self-directed learning.

Together, these technologies make self-directed learning a more accessible learning modality for increasing numbers of students. Technology rarely motivates self-directed academic learning on its own. However, good learning technologies make self-directed learning easier and more effective,  thereby making it an increasingly viable modality for an increasing number of students to use to learn academic content.

A new model of schooling

With all the advances in learning technologies over the last few decades—including AI—students no longer need to learn in lockstep, constrained by the limiting factor that a teacher can’t simultaneously teach different lessons to each student. Instead, students can learn using the best resources for them and at different paces based on their individual needs. 

Asynchronous learning technologies also break the constraints of the classroom. Students can learn at home, in a library, in a study hall, or wherever makes the most sense for them—including in settings like in museums, internships, or travel.

In other words, when learning technologies enable new models of schooling oriented toward self-directed learning, they break the bottleneck constraint of teacher-directed batch processing that has always shackled conventional schools. 

When learning no longer hinges on teacher-directed lessons, many new approaches to schooling are unlocked. This vision of learning isn’t just a marginal improvement on conventional classroom learning; it’s a radical redesign of schooling. 

Will AI-powered learning replace teachers?

So, what does this mean for teachers? Will they soon join the ranks of other workers whose jobs will be eliminated by technology? Absolutely not. 

Teachers will still be essential in new models of schooling. But their necessity won’t come from their skills as performers to captivate student attention as they explain content. It won’t come from their expertise in classroom management—maintaining orderly and un-distracting classrooms while keeping whole groups of students gainfully employed. 

Rather, teachers’ enormous value in schools organized around self-directed learning will come in fueling the parts of the motivation equation tied heavily to communities and relationships. 

First, teachers are still crucial for tipping their students’ expectancy toward learning. Teachers learn things about their students that AI is far from understanding—things like personality quirks, family backgrounds, character strengths, and day-to-day life circumstances outside of school. In other words, they understand their students on a human level. 

With this understanding, teachers play a critical role in both troubleshooting the barriers to students’ learning and conveying to their students that they believe in their students’ potential. This relationship-based role of teachers is far from being replaced by AI. 

Additionally, teachers foster the relationships and communities that heavily influence students’ subjective value of learning. Two of the biggest sources of subjective value for students are their desires to feel successful and to have fun with their friends. 

Teachers curate contexts where learning is fun and where achievement has meaning. They show excitement for the topics students are exploring. They facilitate group learning activities that make learning fun. They celebrate students’ successes. They organize opportunities for students to show off their accomplishments to friends and parents. They help students see themselves as members of a community of learners. And they invite students to see themselves with new identities as writers, scientists, or historians. 

In sum, self-directed learning powered by AI and other educational technologies doesn’t replace teachers. But it does prompt a radical rethinking of how teachers go about their jobs. 

In new models of schooling, a teacher’s role no longer centers on managing classrooms and covering content. Rather, teachers will focus on shaping the uniquely human elements of learning communities that motivate students to drive their own learning.


It’s exciting to consider the enormous learning gains that could come from unlocking self-directed models of learning. 

In conventional classrooms, teaching can easily feel like dragging a heavy load down a long road. Teachers often labor to move their students through academic content while most students actively try to avoid as much work as possible while still maintaining acceptable grades. Meanwhile, for many students, learning in school feels like something they have to do rather than something they’re excited about. 

But what if all that changed by using technology to shift teachers’ focus from managing academic production to unlocking motivation? 

In other words, what if teaching was no longer about getting students to move efficiently down a conveyor belt of learning but instead about putting students in charge of the production process and helping them feel inspired to design and build their futures? 

Unlocking student motivation in education could be like pouring jet fuel on a wispy candle. When students feel empowered to own their learning, and teachers can dedicate themselves to fostering student motivation—free from the constraints of classroom management and lesson delivery—the potential for accelerated and expanded learning becomes immense. 

Unfortunately, the current wave of AI products is unlikely to revolutionize education within conventional school models. Nonetheless, AI holds tremendous potential for reshaping schooling. 

This post has focused on the possibilities that AI brings to education, but realizing these possibilities is a complex challenge. 

Models of schooling that use AI to enable more personalized, self-directed learning aren’t going to evolve out of existing conventional classrooms naturally. Instead, education leaders will need to create the conditions where new models of schooling can emerge and expand. Check out my other work for insights into how to bring about the transformation of schooling.

Because AI is a nascent technology with implications we may not yet comprehend, I’m publishing this piece as an introductory reflection aimed at spurring conversation and questions. In this spirit, I’d love to hear from you: what are the critical considerations around AI for education? What challenges may educators, school leaders, and policymakers face? 

Let me know—I’d love to hear from you!


  • Thomas Arnett
    Thomas Arnett

    Thomas Arnett is a senior research fellow for the Clayton Christensen Institute. His work focuses on using the Theory of Disruptive Innovation to study innovative instructional models and their potential to scale student-centered learning in K–12 education. He also studies demand for innovative resources and practices across the K–12 education system using the Jobs to Be Done Theory.