The Diamond Age is one of my all-time favourite books. It’s steampunk set in a Neo-Victorian nanotech world. The story begins when a kleptocrat worries that generational wealth will make his daughter weak and lazy. It was the first time I’d heard Andrew Carnegie’s saying that people take three generations to go from shirtsleeves to shirtsleeves: Money earned by the first, hardscrabble generation is squandered on its children, and lost by its grandchildren.
To try and stop this cycle of wealth and laziness, the kleptocrat hires a coder to develop a a small device (called the Primer.) He hopes the Primer will help his daughter grow up bold and clever despite being born into privilege.
The Primer is an interactive, illustrated book. It’s part advisor, part life coach, part co-conspirator, and part therapist. The plot thickens when coder steals a copy for his own daughter, who is subsequently pursued. Adventures ensue, with the girl on the run aided by her copy of the Primer.
In the book, the Primer is voiced by an interactive actor (“‘ractor”), a live human performer who improvises in real time over VR based on a computer-generated script. Basically a paid, personal, motion capture actor working live for an audience of one. What the story misses is that we don’t need a human as an interactive front-end.*
Machines got better at humans
Computers are far easier to use than they once were. In the advent of the computer age, we programmed hardware, and removed literal bugs from the inner workings of our machines. Then we programmed in machine language, columns of glowing green hexadecimal. Human-readable languages like BASIC made programming widely accessible; today, development environments such as Unity feel more like video games than code.
Our interfaces got easier, too. Where once we interpreted stacks of cards, we moved on to paper printouts, then computer screens, and now, we scroll through endless feeds at night in our beds.
Cheaper computing gave us better interfaces. When you use the calculator on your phone, the vast majority of its computing power is devoted to making pixels look like physical buttons. Very little of it is spent doing your math problem.
Humans got better at machines
Computers got better at humans, but we also got better at computers. Our brains are plastic, and children born today will spend more than a quarter of their lives online. Today’s student is weaned on Discord and Roblox and tablets and DMs and memoji. Search is part of how we think.
This generation is perfectly comfortable—perhaps happier—talking directly to a machine. Let’s face it: When you’re accessing information, people just get in the way. The Primer’s ‘ractor is an outdated UX device we don’t need any more. Human agents are superfluous skeuomorphs
A School of One
In 2009, New York launched a math program called the School Of One (now “Teach to One”) where “a computer-based Machine Learning algorithm collects data to generate a daily lesson plan or "playlist" for each student based on what is determined to best meet their learning needs.”
In one year, Teach to One students gained almost 1.2 years of growth in math or nearly 20% more than the national average. A 2013 study showed participants had 1.5x times the national average in math gains.
Teach to One isn’t a computer-taught class. But it is learning tailored around the student, at the student’s pace. It adapts the types of communication, learning environment, and schedule to whatever works best, and adapts with feedback loops.
Stephenson’s Primer is also an adaptive teacher, offering the right lessons when they’re most needed, synthesizing answers, tailoring them to its class of one. In the book, this tech wasn’t available to everyone, because the Primer itself and the interactive human UI were both expensive. But behind it all, the ‘ractor was reading a computer-generated script.
Today, chatbots respond, Lyrebird speaks, Dall-e illustrates, Synthesia personalizes. We have amazing video content, the ability to visit or simulate anything, at any scale, in VR and AR. And far more smartphones than chalkboards. The Primer stack is now basically free to much of the world. It just snuck up on us.
Already thousands of students credit online content from Kahn Academy and Crash Course with helping them finish high school in spite of their teachers. Who needs a real-time human to learn?
Forget the scarce, precious artifact of Stephenson’s book. Illustrated Primers are for the many, not the few. What happens when everyone has a personal teacher that can out-teach all but the best teachers in an ailing educational system, for less money?
As usual, we aren’t thinking weird enough—both about what’s possible, and about what might go wrong.
We worry about someone using AI to write a book, rather than whether we’ll still read books.
We worry about AI writing term papers, rather than whether schools will become obsolete.
Some consequences
Obviously, replacing teachers with Primers has plenty of consequences, unintended and otherwise. Most of the reactions I’ve had to this are polarized, with people doubling down on their existing beliefs about technology or the educational system. The reality is much more nuanced, with plenty of pros and cons.
I’ve added just a few issues that came to mind, but would love to hear any other ideas or concerns you might have.
What if it’s better and cheaper?
Automated teaching may produce good results for less money, which is hard to debate. Even if it doesn’t work as well, it may catch on. Teaching in many countries is woefully underfunded; my parents were both teachers, and I’ve seen firsthand just how little money schools have. Education around the world is in decline at all levels.
The cost savings, and lifestyle benefits, may also be unavoidable. Guided study tailored to each students might mean parents don’t have to wake up to get their kids to class (there’s even good evidence that letting teens sleep in improves their grades.)
And thanks to testing, we can measure whether kids get better results, objectively.
Can it save us from anti-science fundamentalists?
Primer-based teaching might be better than the alternative. What could a personalized learning bot do to educate women prohibited from attending high school in their country; or to teach reproduction or climate change or evolution in an anti-science state; or to help working parents finish college.
Can it brainwash us?
On a darker note, programmatic education could be a tool for mollifying and misinforming an electorate. It could be used to surgically remove awkward histories or inconvenient science.
No teachers also means no righteous objectors inspiring students to question the status quo, something any government needs in order to remain just and tackle corruption.
What if the AI is wrong?
You might also think ChatGPT isn’t accurate—and you’re right. Today’s AI chatbots are probabilistic language models. They’re still really impressive. Maybe a teacher doesn’t need to understand the student in order to set curricula, respond to questions, and analyze feedback.
Most of our discussions today about AI chat are based on experience with OpenAI’s public demo of a chatbot. Google has a bigger chatbot, trained on nearly six times as many parameters. It’s so good an employee tried to have it granted rights in July, and was fired.
ChatGPT is a beta. By using it you’re building it.
Reinventing learning
Our public Primer may start out as a for-profit tutor chatbot. What parent wouldn’t invest a bit of money each month to help their kid study? Most great technological breakthroughs start out as toys, then become essential.
I’d much prefer that educational AI be a moonshot that puts a basic high school curriculum in the public domain, because leaving it to the private sector might further divide our society. If I were a government, my department of education would be making this a top priority immediately.
Whatever your thoughts on the matter, we are only years—maybe months—from a chatbot that’s better at giving students accurate, individual attention which surpasses what a teacher can do. This has wide-ranging consequences for society we need to get ahead of, because worrying about how AI helps students cheat misses the real news: We’re reinventing learning.
* To give Stephenson his due (huge, huge fan) the coder makes copies of the Primer, and the book explores the consequences of giving the tech to the rich, the homeless, and the masses; and the human ‘ractor is rather necessary to move the plot along.
This will save me from having to write a blog post about the exact same thing!