Simple idea: AI that generates videos on command would be incredible. I'm imagining something like this:

Prompt: I'm trying to learn about matrices and linear transformations. Prepare a 10-minute video where you go in-depth into this topic.

AI: Here's a video going over matrices and linear transformations, covering linear maps, matrices, matrix representation of a linear map, ranks, eigenvalues, and solving some examples.

However, video content is still too static. It still seems like an “AI for x” approach to solving the problem of education. Why create videos when AI can provide a direct Natural Language Interface + Generative potential for any material? 

Video is a one-way link, from content to user. It still doesn't replicate the tutor-student direct communication line, so it probably doesn't solve the two-sigma problem the way I thought at first.

So, how can we decrease the static? It seems pretty intuitive, actually. The way to replicate a human tutor is simply by creating an artificial tutor. Some might say, "Oh, well that's already available; you simply need to start chatting with ChatGPT, and it can guide you through probably any topic." Yes, probably, but I still think it's not engaging enough. Maybe for a highly motivated student, that's enough; they'll read through Chat's answers and then ask more specific questions, which will lead to a great educational outcome. But when trying to solve the Two-Sigma problem, we're attempting to address the needs of the average student, not just the highly motivated ones.

At the end of the day, a highly motivated student will most of the time find resources to learn whatever is on their mind. So, we need to target the average student, whom we'll call John. For John, ChatGPT is probably not immersive enough. Raw text simply isn't engaging enough for most students, especially in subjects that the average student struggles with the most, such as math, physics, etc. These are visual fields by nature. How is John supposed to learn about matrix transformations when he can't visualize the coordinate space and how it transforms when multiplied by some constant? Is John supposed to infer this from the text ChatGPT responds with?

So, what could be the solution? Once again, we come back to the question of replicating an outstanding real-world human tutor – at least enough aspects of it to immerse John sufficiently so he can learn the topic at hand.

How am I imagining it? John logs into his learning platform, let's call it School.com, which has a minimalistic interface similar to ChatGPT's. John will find a large input box with the placeholder "What do you want to learn next?" When John inputs "Matrices and Linear Transformations," our chatbot friend inside School.com knows John's math level from previous inputs (because the output of this question will vary greatly from a high school student starting linear algebra to a grad student taking real analysis). So, our chatbot friend will generate a dedicated lecture for John. However, this won't be a 10-minute lecture where the bot tries to explain everything it deems useful and related to matrices and linear transformations. Instead, our bot will present a short introduction about the topic and begin explaining matrices with visualizations, asking John questions, presenting examples, and giving exercises to John, etc., until the chatbot deems it necessary (that is, until John has "mastered" matrices and can now start learning linear maps), which will be achieved through "mastery learning" practices. John probably shouldn't be learning about the rank of matrices if he hasn't completely mastered the concept of matrices, and our chatbot inside School.com will ensure this doesn't happen.

Then, John will start building up his knowledge, beginning with matrices and moving on to linear maps, through videos if needed, exercises, or whatever John prefers.

But perhaps John is not very opinionated about his learning style. That's totally fine because why would John have an opinion on how to learn matrices if this is the first time he's been exposed to them? Here, our chatbot friend will compensate for this by providing a structured learning approach, which is precisely what most students learning on their own are missing: a structure, a specific way of learning certain things that allows them to feel like they're building up to a certain knowledge. As any autodidact out there may know, learning on your own often feels a bit too free. There's an obvious need for metrics and success guidelines in such open environments so students know when and how they're progressing. ChatGPT and other AI platforms are not directly targeted at teaching, so they don't offer a continuous progress bar, badges, or other materials that have been proven to provide students with motivation and self-actualization incentives. Current systems are obviously "enough" for the most qualified, advanced, and motivated students who don't need the structure, but why optimize for the top 10% when AI can provide 100% of people with personalized access to a solution for the two-sigma problem: a personal tutor that can immerse you in any topic in your way, your structure, your pace, and through mastery learning?

Our chatbot in our hypothetical School.com has very few rules. It follows mastery learning, teaches from first principles and intuition, and leads you toward mastery of a topic through step-by-step, bottom-up learning. You choose your learning adventure and all the remaining parameters that create your personal experience. That's the beauty of it. No more 45-minute blocks of learning algebra with 60 students at different math levels than you, with a disinterested teacher who barely knows your name. That's an education scarcity mindset. With AI, we don't have to think about scarcity as a limiting factor. We can now solve the two-sigma problem with personalized tutors that will not only help the most advanced and motivated students but also the Johns and every student out there who wants to understand, on a fundamental level, the world around them.

Progressus Scientia.