The Future of Education

Problems, principles, and 20 year vision

Education is broken. It smothers our inborn natural curiosity and love of learning. It homogenizes the broad diversity of interests, abilities, and types of intelligence into conformity. It optimizes for what the masters of society deem to be important: deference to authority, punctuality, sitting down and shutting up, and a definition of intelligence that prizes analytical skills above all else.

There are two core issues. The first is pragmatic: our current system was conceived of to produce factory workers, but our economy no longer has a need for this skillset. Anything that requires rote rule following is either already outsourced or about to be automated. In tomorrow's economy we need people who are creative, empathetic, collaborative, and able to create new ideas across a variety of disciplines. The second is philosophical: people should be supported to explore and develop their intellectual potential, rather than treating it as a defect. Lifting people up to meet their personal potential is a worthy end in itself.

Our education system has certainly made progress in the last century: it's more accessible to all, instruction is much more humane (no more physical punishment), teaching techniques are more sophisticated, and some technology has been deployed to improve lessons and administration. But the world is changing at an accelerating pace, and our school systems are far too bureaucratic and slow to adapt. The students of tomorrow will suffer as a result.

What we refer to as a "school" is actually a bundle of services. It provides children with much-needed daily structure. It provides parents with childcare. It provides common resources, like musical instruments or a library books. It provides social connection and opportunities for emotional development. It provides a credential that, by convention, society requires to unlock downstream opportunities. And last but not least, it's a platform for delivering learning content and assessments.

Where we could be

If we were to rethink any of these from the ground up, I'm sure we could find a better solution. I'd like to focus specifically on the content delivery piece for now.

From first principles, learning is just the process of altering one's neural state to represent new knowledge and skills for later recall and application. We don't yet have nanobots to directly manipulate our neural states (nor would we know how to direct them if we did), so we use our senses as an interface, absorbing information via images, video, speech, and text.

Examples can help inform how we can effectively convey learning content. Vox Explained or 3Blue1Brown are both informative, compelling, and accessible (depending on your background knowledge). However, schools provide a familiar counter-example, with generic lectures that perplex those who are behind and bore those who are ahead, teaching content that serves standardized tests more than students' actual needs. Also, as much as I love it, Wikipedia is less than optimal. Try comparing the 3Blue1Brown explanation of Fourier Transforms to this Wikipedia article on the same topic.

Principle 1: Personalization

It should be personalized. We've known for decades that tutoring provides superior outcomes to group-based classroom teaching, moving the average student to the top of the class (see here). Personalization breaks down into three components.

  1. You should be able to carve out your own learning path, letting your passion and curiosity be your guide.

  2. The content should be delivered at a pace and difficulty that is appropriate for your current level.

  3. Delivery should take into account your current knowledge, skipping familiar concepts and relating new concepts with things you already know.

Principle 2: Quality Content

Content should be delivered in a way that's as engaging as possible. This likely means learning bite-sized concepts, delivered via a compelling narrative, with visual aids and metaphors to help make connections.

Principle 3: Tight Feedback Loop

Finally, our system should also be able to frequently assess how well you're learning, and adapt as needed. If the lessons are clicking, keep going. If not, it can try explaining it another way, or perhaps revisit a more foundational concept. And the system should ensure you're remembering concepts within a lesson, and longer-term, to make sure you're truly absorbing the concepts.

Principle 4: Free

Good news - this already exists! You can call up Sal Khan or some other world-class teacher and ask them to provide private tutoring. But, if they even have space in their schedules for you, it's going to cost an arm and a leg. Costs can exceed $50/hrfor an average private tutor, and if you're getting tutored for most of your learning, say 20 hrs/week, that's $50k/year. Whether paid for by individuals or governments, it's not feasible for private tutoring to become the primary instruction method for kids today.

I believe this is fundamentally a product problem. Software will be able to outperform even the absolute best tutors, and do so for the marginal cost of a few CPU cycles (pennies or less).

To achieve this, we'll need to find a way to decompose and digitize the act of teaching humans, and structure it in such a way that computers can understand, learn, and optimize. We can start with humans providing the bulk of the content for a few specific verticals, but with more data and better NLP, we can improve performance and increase coverage to all of human knowledge.

Here's what this might look like in 20 years' time.

Vignette: A Possible Future

A high school student, Tina, is at her local high school. She has just finished lunch, concluding the half of her day that's dedicated to physical and social development: sports, art, meditation, and unstructured free time. Now it's time to start the "knowledge acquisition" phase of the school day. 

She loves animals and is currently learning about how energy is extracted from food from across the animal kingdom. She has has been making great progress in biology - according to her age she should be in Grade 9, but this topic is at a Grade 12 level.

She goes to the biology room, and sits near some the other students that are working on similar concepts. Everyone is on a unique learning path, but they can bounce ideas and help each other work through problems as needed. And there are many valid paths to earn course credit, since there are many ways of demonstrating knowledge and skills in a given subject.

She puts on her headphones and AR glasses, and sees the digital avatar of her tutor floating on the desk in front of her. Her virtual tutor's name is Jane, and they've been working together for almost 10 years now. Jane knows what Tina knows, and how she learns best. Jane also knows all human knowledge, because she has read and understood the entire internet. Tina loves asking her to explain any concept that she's curious about - her response is always understandable and entertaining.

They dive into their first lesson of the day, describing which nutrients are absorbed where in the reptilian digestive tract. She connects it to their previous lesson, projects 3D visuals to describe how it works, and explains how it can be used in a project Tina is working on. Tina furrows her brow midway through the lesson. "I forget, what are intestinal crypts?" Jane pulls up a 3D model of a lizard and zooms into the right area, and she can tell by Tina's facial expression that it clicked. After about 7 minutes, their mini-lesson is complete. Jane and asks Tina to answer some questions to confirm and reinforce her understanding.

Meanwhile, some students are collaborating to solve problems, and a coach (the "teachers" leave the teaching to the software, after all) is available to help motivate and support the students if they need.

Tina’s family doesn't have much. But she has access to the best education in the world, for free, and she has a very bright future because of it.

Do you agree? Think I got something wrong? I’d love to hear from you, let me know in the comments.