Why EdTech Fell Short and How AI like ChatGPT Could Soon Deliver the Revolution We Were Promised
There was an exciting time in the 2010s when we thought that giving students computers would save US public education. The idea was that some combination of public charter schools, iPads, and Khan Academy could help close the expanding achievement gap in the United States.
Perhaps this was best typified by the $100M tech-infused effort from Mark Zuckerberg and Cory Booker to “save” Newark schools. Unfortunately the effort, to put it kindly, did not work…at all…and even despite the billions of dollars flooding into Venture Capital for EdTech startups, technology has, in my opinion, failed to truly transform education.
I got to experience this first-hand as a teacher during COVID, doing whatever I could to make use of technology to enable virtual education. Now, to be fair, there are some very useful tools out there. Google classroom is a godsend for grading, assignments, and communication. Desmos genuinely makes every aspect of math teaching better. And Nearpod made virtual teaching almost bearable.
The primary benefit of these tools has been in marginally improving the efficiency of all the little tasks that take up teacher time that most people wouldn’t even think of — the hour a day spent managing students who didn’t bring their binder or notes, the hour a day spent organizing papers and printing, the hour a day spent grading physical papers, etc. These tools provided amazing ways to automate all of that and I can confidently say that they allowed me to focus more on the important things in my classroom like actually teaching. But they were far from revolutionary, and I think that if you were to really study the impact, you’d find that the inclusion of tech in the classroom has had marginal-to-immeasurable impacts on student outcomes.
The issue is that education is hard. Really hard. It’s almost impossible to measure teacher quality effectively, some of the best evidence-based interventions often fail, and when interventions do work they have very small effect sizes. It doesn’t help that it is an incredibly political and emotional topic that is full of interest groups that frankly aren’t prioritizing student outcomes.
It’s also just really expensive. Part of this is due to mismanagement and waste, but part of it is due to the sheer fact that teacher productivity can’t meaningfully scale. As in, the max number of unruly 9th grades I can teach now is about 30 at-a-time. That’s the same number it was 100 years ago and 100 years before that. So, while machinery like textile looms improved the productivity of 1 weaver by 10000% thus giving us more textile for way less input, teachers are at roughly the same teaching productivity that they were 100 years ago.
Part of that is baked in and probably essential. In many situations, developing kids can require just as much socioemotional support from teachers as academic support. You can never get rid of that fully. But this general productivity problem points to the sort of holy grail of EdTech and why EdTech has fallen short of being revolutionary → in order for technology to be revolutionary in the classroom it needs to increase teacher productivity by either allowing students to learn things in less time or to allow teachers to effectively teach more students at once. To make this more tangible. I either need to be able to teach kids the full curriculum to mastery in, say, 10 years rather than 13, or I need to be able to effectively teach more than 30 kids at once.
So how do we do that?
The best and most inspiring examples of EdTech come from Science Fiction. Particularly Star Trek and the Neal Stephenson novel The Diamond Age.
There is a brief scene from the 2009 Star Trek remake that shows these immersive “skill pods” that Vulcan kids use to learn. Though this was added in as a sort of afterthought, the scene really stuck with me because it gives what I take to be a compelling vision for how technology could be used to create highly immersive and individualized learning.
Then there’s Diamond Age. [Spoiler alert] The novel centers around the use of an interactive and intelligent book called “A Young Lady’s Illustrated Primer” that adapts to students to provide a perfect and personalized series of lessons that guides the reader to an intellectually more interesting life. It is so good at doing this that its access is tightly guarded, only available to the wealthy.
Take the Vulcan learning pods and the Illustrated Primer. We aren’t there yet but we have the goal (boost teacher productivity in driving student learning), we have the vision (AI for individualized teaching), and I think we are starting to have the tech.
Before two weeks ago (early November 2022), I was extremely pessimistic about the possibility of this tech coming anytime soon. After all, I’d seen the failure of virtual education during COVID, and I’d seen first-hand how hard it is to make computer-based learning work. But then, like many of you, I got to try out ChatGPT and then witness the flood of other incredible AI innovations that I was neglecting out of my own ignorance and (frankly) cynicism.
Now, you can use ChatGPT to write your papers, Lensa to generate portraits, CoPilot to write code, Soundraw to create music, Perplexity to summarize research, and lots of other scary and vaguely dystopian things.
Some are good. Some are meh. Some people think these are genuinely intelligent and some think the tech is just a stupid “if-then engine”. That’s all okay.
Regardless of what the Twitter haters say, the sober truth is that AI has come along much further than most of us thought, and I strongly believe it will soon begin to reshape every single industry…especially education.
So how will that look? [If you haven’t yet played with ChatGPT or any of the other tools, please do so or at least read about it. Not so you can just be in awe and slight terror, but also because this next section will make more sense.]
In the next year ChatGPT and its soon-to-be-released-equivalents will make lots of schoolwork redundant. The primary way that teachers will view these Chatbots will be from the lens of seeing how students cheat with them.
However, some industrious students and teachers will start to try to incorporate these Chatbots into lessons. They will ask the Chatbot to explain things they don’t understand, and the bots will do a fairly good job of identifying their errors and responding within context. It won’t be perfect, and it won’t be able to take much of the inputs that students need (like I scribble on this iPad my answer to a math question, and it tells me what I did wrong).
But then the flood will start once these somewhat commodified models start to cheapen, start to become open source, and start to become better tuned for handling particular contexts. I hope for this in early 2024.
This is when EdTech startups will be able to better use these models for more focused educational contexts. With some tweaking, I suspect that the first groundbreaking tools will be able to take written student work in Math, and then provide individualized feedback about what the student misunderstood. It will allow for an iterative process that allows students to go through multiple rounds of practice followed by feedback. These systems will automatically track student progress and provide metrics to teachers about what students need extra individualized help. In subjects like Math and some Sciences this will be truly transformative. It will decrease time to comprehension and could actually become a tool for allowing teachers to effectively teach and engage more students at once.
The issue, much like the naive optimism about Khan Academy in the past, is that students won’t just learn from a computer on their own. Or at least it is very hard. When I was teaching, the majority of students frankly could not be trusted to do laptop-based assignments by themselves or with groups. So many students needed direct instruction, constant monitoring, and a human touch. So, one of the big keys to unlocking this tech will be in building the skills of teachers to get student buy-in while also tuning these AIs to respond to students in ways that seem genuinely personalized and human.
That brings in what I hope to be the next wave which is likely 5+ years out — Metaverse and VR.
Much like the Vulcan Skill pods, I strongly believe that there is a lot of value in personalized and immersive individual learning. Not only has this been borne out in the research on one-on-one tutoring, but a one-one-one classroom is also the easiest classroom to manage. I say this, partly because classroom management is by far the most important and underrated thing in Education. Quite literally it’s the deciding factor for determining whether a kid actually learns or not, and without good classroom management none of the tech even matters.
The unique thing about the possibility for immersive VR-based tutoring is that it can create safe and effective educational immersion that is easy to monitor. There are now tons of startups diving into the space. I am worried that they are too early. It will just frankly take a long time to get the teachers, schools, students, and tech to the point where it can be easily integrated into the average classroom.
But the promise is that the combination of ever-improving AI-enabled individualized tutoring alongside the immersive and adaptive capabilities of VR could genuinely make something like The Diamond Age within striking distance in the next 20 years.
