AI in education: recent developments
Artificial intelligence is rapidly improving, which is creating a growing disconnect between the trajectory of AI advancement and the response from most (not all) school districts, state agencies, and associated organizations. The discourse around AI taking place in education publications, professional development, and school board meetings is mostly focused on near-term issues such as student academic integrity, and teachers’ use of AI for lesson planning. But the significant advancements in AI capabilities are outpacing these discussions, and this divergence is only going to get bigger over time.
The GPT-4o release is a major step forward
If you’ve not already dug into the education-related details of the latest release from OpenAI, you are out of date. (That’s how easy it is to fall behind on understanding the state of AI. On May 1 you were caught up, on June 1 you fell behind.)
When you have a few minutes, this is the best source to begin with that I’ve found. The two items that most caught my attention are:
Instant tutoring and real-time assistance: this isn’t a new concept in AI, but GPT-4o’s audio capabilities and ability to “see” a math problem are a step forward. How big a step? I don’t know because I’ve not delved into what Khanmigo, for example, was doing previously and can do now. But at the post-secondary level, the stock price of ed tech company Chegg dropped by 50% due to developments related to AI, which is a valuable clue to the thinking of the people who are betting on the future of companies in the education space. AI-driven tutoring is coming and will likely create new products while decimating the value of tutoring companies that rely entirely on humans.
Real-time translation: in the same link above showing the instant tutoring, scroll down to the video of AI translating a conversation between Italian and English. Then, if you or someone you know speaks a non-English language, try it! I had results almost as good as the video, with just a bit more latency which may have been due to our home Wi-Fi. Your mileage may vary, but I was blown away.
Whether I was impressed or not isn’t really the point, though. Instead, consider this article, which discussed challenges of working with English language learners in rural areas of the US, where bilingual teachers are rare. The article never touches on AI, and it also doesn’t propose any easy-to-implement solutions. One would think that in the absence of other potential solutions, AI would be a topic worth exploring. (Although to be fair it was probably written before the GPT-4o release.)
That’s not to say that GPT-4o provides a clear and easy solution—yet. But I suspect that as real-time translation gets better, it will become a key component of the solution. For example, a key challenge for teaching a new language is the value of talking with another person—having a conversation in a new language—as opposed to just learning vocabulary and rules of grammar. This is so hard because having one-on-one or even small group conversations takes so much time; it’s simply not possible to have extensive conversations in most public education settings. But the current version of AI allows a student to have a conversation in many languages, and either now or soon AI will be able to help the student with pronunciation and idioms like “once in a blue moon.”
Some people will respond that it’s not as good as a bilingual teacher, to which my counter is that’s irrelevant because too many of these students don’t have bilingual teachers. Comparing a potential solution to an ideal that doesn’t exist, and isn’t likely to exist, isn’t a useful exercise. But it’s easy to imagine an ELL program that significantly augments the work of human EL teachers.
Meanwhile, a lot of discussion misses the mark
There are certainly plenty of educators and observers who are looking closely at how AI will impact K-12 education and providing valuable analysis. These include people like Jon Fila, who is applying his research and understanding of AI into his current teaching while also supporting other teachers and district leaders.
But it’s also the case that much of the discussion feels akin to horse and buggy drivers wondering how they were going to adapt their carriages to the automobile age. It’s clear that schools, districts, and states have to think about student academic integrity, teacher training, acceptable use policies, and so forth. But in traditional education spaces, too much of the AI-related conversation stops there. Those conversations also tend to over-focus on issues like hallucinations, which are concerning in the short term and need to be addressed, but are not a critical long-term consideration.
Given the trajectory of AI development, it seems to me that the discussions should be a lot deeper than student cheating and teacher lesson planning.
But there are real concerns
That’s not to say that real issues and concerns don't exist or should be ignored. The list is long; here are a few:
Dan Meyer does a great job of exploring the impact of AI (and other technologies) on teaching and learning generally, mostly from a skeptical vantage point. A recent post discusses how AI enthusiasts love to talk about how technology can personalize word problems for each student, with the idea being that such personalization will increase student engagement—but Meyer believes (and I agree) that they are wrong. That post is well worth ten minutes if you think that such personalization is valuable. His narrow point is right on, as is his broader point that technologists often don’t understand the value of the technology in a school setting.
Another Dan Meyer post talks specifically about GPT-4o, demonstrating Meyer’s skill in grappling with what the technology is currently capable of—and its limits. Key quote: “Tech and business leaders are deeply unserious right now. Rather they are deeply serious about new technology and shareholder returns but they are deeply unserious about the needs of learners.”
Wandering around the ASU+GSV air show (basically an exhibit hall for AI-related vendors) in April, I was reminded that many of the technologists and companies creating AI-related education products lack two elements that some people might consider useful: 1) empathy for how schools really work, and 2) a basic understanding of the history of educational technology. Item 1 was demonstrated by conversations I had with people who don’t understand how tech tools are adopted in traditional schools. Item 2 was demonstrated by the use of the exact same terms and phrases that have been in vogue since at least the late 1990s—with no apparent consideration for the irony of using old descriptors for this thing (AI) that is so new.
Teaching content vs teaching students
It’s difficult to make predictions, especially about the future. Predictions are even harder regarding a technology that’s moving as quickly, and is as potentially transformative, as AI.
But I have a strong sense that the traditional education system is not meeting this moment, and that this failure will become clearer within roughly three years. Key players are focused mostly on how AI can be folded into a future education system that looks very much like the current system, instead of truly exploring how AI-driven changes may lay the foundation for some fundamentally different approaches. These different approaches would consider, for example, how much human time (teachers as well as other professionals) should be spent on teaching content, especially relatively simple concepts, versus working with students in a deeper counseling/mentoring/champion role.
Many of the schools and educational organizations that are actually investigating how to use AI in deeply impactful ways are part of the digital learning community. To that end, we are launching a new affinity group within our Digital Learning Collaborative, called AI for Innovative Schools. The group will be made up of schools and organizations that have the flexibility to respond to AI opportunities and challenges, test new ideas, and apply them at scale. At first it will be primarily a discussion and sharing group that we will bring together once a month remotely, and onsite together at DLAC. Eventually we intend to publish findings and help this group lead the adoption of AI in education.
If you’re interested in taking part, email me (john@evergreenedgroup.com) and I’ll get you some details.