How AI Will Change Education
Education does lots of things for society. AI will change all of them.
Education in the US is a big big deal. It takes up 18-30 years of our lives, employs over 10% of our workforce, and is responsible for 60% of non-mortgage/non-car debt. Even a minor improvement to education could be a big deal.
Education is also something that has changed massively in recent decades. In 1930, only 19% of people graduated high school and only 4% went to college1. If something has changed a lot in the past, it is reasonable to expect that it will change a lot in the future.
And I expect AI to change education a lot.
One-on-one tutoring is known to be far more effective than whole-class teaching2. If someone is listening to a group lecture, half the time they are bored because they are being told stuff they already understand, and half the time they are lost because they missed something important. By contrast a tutor can pace themselves exactly with the student, and focus on exactly the areas where the student is stuck.
The reason why tutoring is not widespread is because it is impractically expensive - or at least it is if the tutor is a human.
As an experiment, I created a GPT that acts like a personal tutor for a subject. You tell in what subject you want to learn, it asks you a set of questions to determine your current level of knowledge, and then it walks you through a personalized curriculum that fills in your gaps, asking questions along the way to track your learning. It’s far from perfect, but it works well enough that it’s become my preferred way of learning about a new topic, and techniques like this will only get better as AI improves.
Students often find education pretty boring. Part of the reason is that they are being taught how to solve problems that they haven’t encountered yet, so it’s not obvious to them why what they are learning is useful.
So why do we have this strange setup where we give people 18-30 years of education without yet knowing what skills they are actually going to need, and then send them into the workforce in the hope that they will still remember whatever skills turn out to be useful. Wouldn’t it be better to send them into the workforce earlier, and then teach them skills on demand when they encounter situations where those skills will be useful?
Part of the reason for this strange setup is that our education system is based on the principles of mass production. We teach lots of students the same things in bulk, and that’s much easier if you can teach everyone the same things at the same time, rather than teaching everyone different things at different times. This is much less of a problem if people are learning from personal AI tutors.
Another part of the reason is that each skill you learn builds on top of another skill you need to learn before. If you learn something in school or college, you’ll be taught things in a good sequence that makes it easy to gradually build up your skills. But if you suddenly want to learn string theory or international diplomacy from scratch then it can be hard to know where to start because everything you might read on the subject assumes you already know something you don’t yet know.
AI changes this too. If you want to read a cutting edge research paper on a subject you know nothing about then you can just paste any sentence you don’t understand into GPT and ask it to explain it. Often this will lead to a back and forth conversation where you and GPT gradually probe the gaps in your knowledge and efficiently fill in all the gaps between what you already know and what you would need to know in order to understand the paper.
In the future it will likely be possible for people to enter high skill professions with vastly less pre-existing education, safe in the knowledge that AI can teach them the skills they need on demand.
The reason why people care so much about getting into top colleges isn’t because they think those colleges will be better at teaching. Indeed often the teaching at a top college is worse than lower ranked colleges, because they select their faculty based on research prestige rather than teaching ability. The reason people want to go to top-tier colleges is because it increases your chance of getting into a top-tier job.
But why is this the case? In a perfect world, employers would intensively interview everyone who applied and determine who was actually the most qualified. But interviewing is ridiculously expensive (many hours of time spent by high paid employees) so you need an easy way of filtering down to people who are likely to be good, before they get to the interview stage, and college has become the de facto filter.
To some extent one can use multiple-choice automatically-graded tests as a filter, but these do a poor job of measuring the skills you care about. They can’t test your ability to create new things, and seeing a list of options is likely to job the test-takers memory of what the right answer is.
As an experiment, I created a simple GPT interviewer that can interview you for any job. Unlike most real interviewers, this one is even polite enough to tell you how you could improve. It’s far from perfect, but it works well enough that I can imagine something like this being an effective first-pass interview filter for many companies, potentially removing the need to use college degrees as a filter.
A big part of the role of earlier years of education is baby-sitting. Schools keep kids safe and occupied, allowing both parents to work out of the house full time.
Can technology help with this too? To some extent it already has. Everyone knows the power of “screen time” in keeping kids occupied. While current “screen time” is often rightfully seen as a poor use of kids' time, it’s unclear if that will always be the case. Future screen time might be “AR headset time”, taking place in the real physical world, while adding enough guidance, teaching, and stimulation to remove the need for constant adult supervision.
Similarly, parents of multiple kids will likely be aware that older kids are often pretty good at entertaining younger kids until things suddenly go sideways. An AI mentor might be able to step in to help prevent things going sideways, or alert an adult at exactly the right moment when it looks like things are about to go wrong.
Or, as a less dystopian-sounding idea, AI might make it more practical for kids to exist in a multi-generational society, actually performing useful tasks around older family members and community members, teaching them the skills they need as they go, freed from the need to be part of a same-age cohort for the convenience of teaching. Kind of like how life used to be in the days when nobody went to school and we all worked together on our farms.
As an experiment, I created a simple GPT that entertains kids with a simple D&D style adventure, while also teaching them a subject of your choice. It’s very simple, but good enough that my older son asks to play it every evening.
That said, I expect that young kids are always going to require fairly close support from adults, even if AI is helping improve the teaching. Direct human contact from adults is just really important.
Education also does a ton of other things.
Some highly paid professions use arguably-excessive education requirements as a way of limiting competition and keeping their own wages high. But it’s not clear this is something we actually want.
College can act as a source of social connection and a way to make life-long friends. But people reported having many more close friends in the decades before the rise of college, so college clearly isn’t necessary for this, and indeed other institutions like churches and local groups may be more effective.
Elite Colleges curate the ruling class, bringing together the wealthy, the well connected, and the highly skilled into a community of people who know each other and can do each other favors. But again it’s not clear that this is actually desirable, or that it couldn’t be done in a better way through earning the right to rise in bottom-up institutions.
In 1930, John Maynard Keynes predicted that by 2030 people would only need to work 15 hour weeks, due to the ability of labor saving devices to substitute for human labor. Technology was indeed able to substitute for a lot of human labor (farming is now only 1.2% of employment), so where did all that spare labor go? Some of it went on providing luxuries we didn’t have in 1930 (e.g. more than one set of clothes), but a lot has gone on educating our workforce to the level needed for us to create all those labor saving devices. If AI can make education more efficient then maybe we’ll get that 15 work week after all.
Or maybe we’ll get an AI dystopia and all get turned into paperclips. I have no idea what’s going to happen, except that I’m pretty sure the future isn’t going to be the same as the present.
https://www.infoplease.com/us/education/educational-attainment-sex-1910-2020 Other sources have slightly different numbers, but in the same ball-park.
Won't that interview filter GPT stop working as soon as someone else makes another GPT which generates answers good enough to pass the filter? :)