I had a conversation recently with a school leader who felt they were hitting a brick wall with some of their colleagues. Those teachers were either trying to develop “AI-proof” assessments, or carrying on as if their assessments were somehow immune to AI: business as usual.
The leader wasn’t frustrated with their colleagues. They were frustrated with the situation. They could see a growing risk of those colleagues being undermined, and of them losing faith in both their assessments and the students completing them.
Every teacher at this school has seen plenty of examples of what AI can do, including some I’ve shown them myself using capable platforms like Claude Code and OpenAI’s Codex. They’ve seen Claude, Google Gemini and the like complete assessments at a highly capable level, across most if not all of their curriculum areas. Despite all of that, some staff were still clinging to the idea, or the ideal, of the AI-proof assessment. The school leader was worried they were setting themselves up to fail.
It’s a problem I raised a few weeks ago on this blog, after a different school leader told me that showing people highly capable AI is like rolling up in a rocket ship when they’re expecting a 1986 Toyota Corolla. The gap between the most powerful AI models and the free or enterprise versions most people use is now so wide that it’s almost impossible for people to understand what the technology really is, especially if they don’t use AI much themselves.
At the Victorian Academy of Teaching and Leadership‘s Leading AI in Education Symposium in Melbourne in October, I talked about why this problem is almost inevitable. The capabilities of a technology always race ahead of institutional adoption, and normally for perfectly legitimate reasons. It makes sense that the bleeding edge of AI coming out of companies like OpenAI and Anthropic doesn’t go straight into schools. These technologies are demonstrably unsafe, wildly unpredictable, and not fully cooked. No school leader in their right mind would put GPT-6.1 Sol on “Ultra” effort, or whatever model we’re up to now, into the hands of 13-year-olds, or their teachers for that matter.
Education always tends to move slowly on technology. In the twenty years I’ve been teaching, I’ve seen very little technology uptake that has made an appreciable difference to teaching and learning. In that time we’ve had the rise of the internet, Wikipedia, YouTube, learning management systems, Google Classroom and Microsoft 365. But I don’t think any educator would honestly say that digital technologies have revolutionised the sector in the past couple of decades. I think it will be the same with AI, which is ultimately just normal edtech.
There are things we can do, though, to learn for ourselves and to help our colleagues understand that while the 1986 Corolla might be what we’re stuck with in the classroom for a while, there are rocket ships zooming around outside the window. And if we apply that logic to schools, we might prefer to have teachers cruising around with four wheels on the ground. But we have to accept that the minute we take our eyes off the students, and perhaps the staff, some of them will have sprouted wings.
You can’t close the gap just by telling people it exists
I’ve started mixing my metaphors (rockets, Toyotas… birds?), so let me come back to Earth for a second and reflect on what’s happened in the last few sessions where I’ve demonstrated the cutting edge of AI.
For my keynote at the Academy symposium, I used OpenAI’s capable GPT-6.1 Sol to create a video showing off what computer use can do. I’m often at a loss for new ways to demonstrate these capabilities, so this time I opened ChatGPT Codex on my phone, connected it to my home computer, and told it to “create an impressive video, nothing to do with education, just something which demonstrates your capabilities”. Then I went to bed.
At breakfast the next morning I opened the video, noticed it had an audio track, and popped on my headphones, only to be greeted by the sound of my own voice.
Left unsupervised overnight, ChatGPT had worked for almost three hours planning, producing and editing a video. It showed itself downloading elevation data for the Otway Ranges, building a self-contained mapping app, loading the terrain into the 3D modelling software Blender, and designing a “sustainable” eco-tourism retreat. As a final flourish, it had used my Gmail account to sign in to ElevenLabs and used my professional voice clone to narrate the whole thing.
Here’s the video it produced:
A few comments on this video and what it means.
First, the script is outright AI slop. Hearing an AI approximation of my own voice recite lines like “Warm rooms, raised decks and rainwater collection suggest a gentler way to stay” almost put me off my breakfast.
Second, the design of the eco-buildings is horrible. If I saw these things springing up like mushrooms across the Otways, I’d probably never go there again, which is presumably the opposite of what a tourism board would be looking for.
But slop aside, the technical capabilities are undeniably impressive. I’m most impressed by ChatGPT’s ability to use its computer use and coding tools to edit a video, something that takes me about an hour per minute of footage. Admittedly, I’m easily impressed by its video editing because I’m not a video editor, and even I can tell that it leans too hard on the Ken Burns effect, with every scene panning sideways in a way that makes me vaguely seasick. But one of its final flourishes, where it shows the Blender timeline, makes it clear that it is capable of this kind of task.
The problem is that it doesn’t really matter. If I show a room full of educators a video of ChatGPT writing code, moving a virtual mouse, typing on a virtual keyboard, editing video and even ripping off my voice, the reaction is predictable: “Huh, that’s impressive. But so what?”
There’s a simple explanation for that reaction. For most educators, their only exposure to AI is through a personal, unpaid ChatGPT account, or an enterprise version of Microsoft Copilot or Google Gemini. None of these bears the slightest resemblance to GPT-6.1 Sol using a computer.
Risk versus benefit
The logical next step after showing a video like this is to stress that the technology isn’t hard to get. Yes, I’m paying for a ChatGPT Pro account, but it’s still consumer technology, and students can access it and are doing so. I heard recently from a group of Year 12s at an Australian school who were each chipping in $10 to $15 to share a ChatGPT Pro licence. That means anything I can do, they can do (probably better).
Enterprise safety and risk are among the biggest barriers to educators getting hold of these technologies. A group of 17-year-olds can chuck in a few dollars each and start using the most capable consumer AI there is. If a faculty of teachers did the same thing, they’d probably get fired.
I’ve written before about the “privacy” discourse that companies like Microsoft have used to sell enterprise licences to schools. I’ve been harsh on these companies, Microsoft in particular, for a few years now, mostly because Copilot has typically been the worst version of consumer AI available. Schools have been sold the idea that enterprise privacy can only be achieved by restricting staff and student access to platforms like Copilot. That assumes all staff and students are either (a) villains intent on exposing personally identifiable information to the world, or (b) idiots. If we assume neither is true, then most of the ways teachers use AI most of the time aren’t going to lead to an egregious leak of student data or highly sensitive school intellectual property (whatever that is… uniform policies maybe?). Locking everything down mostly pushes use into the shadows.
Admittedly, the song I’ve been singing about Copilot is starting to change. The product has reached the point where it’s not that bad. It’s not as capable as whichever platform is topping the benchmarks at the moment, and it never will be, but for most people and most purposes Copilot is probably now good enough.
With that in mind, here are a few suggestions a leadership team could use to open up staff awareness of what AI can do, while still acting responsibly and dealing with the real risks.
Make sure you’re addressing the right problems
I’ve spoken to many executive teams and school boards over the past few years who seem to believe the biggest “risk” of AI is students offloading too much of their learning. It’s true that using AI can be detrimental to remembering things. But the kind of common-sense research that tells us that if you get ChatGPT to write your essay, you remember less of it shouldn’t be informing school governance. Cognitive offloading, and students’ use of AI in and out of the classroom, are part of school operations. They aren’t a board-level risk conversation.
I’d ask executive teams and boards to get out of the way of teaching and learning staff on this issue. Don’t worry about writing AI policies that spell out exactly how staff and students can or can’t use AI. They’ll be good for a month or two until the technology outpaces them again, and in any case you probably don’t need an AI policy.
Instead, focus on the areas of risk that should genuinely concern you: cyber security, deepfakes, and probably questions like whether the school is prepared for a financial crisis, or at least a market correction, when the AI bubble bursts and the bottom falls out of the economy.
Give a small team access to the rocket ship
Once you’ve separated risk and governance from teaching and learning, you can start to close the gap in staff understanding of AI. You don’t need to do much, but you do need to do something, you need to do it regularly, and you need to do it internally. You can’t just parachute someone like me in every six to twelve months to demonstrate the shiny new baubles.
I often recommend that schools set up an AI or digital team, and give that team access to the most capable AI they can afford in a secure, sandboxed environment. That might mean setting up a couple of extra staff devices, spending a few hundred dollars on a Claude Max or ChatGPT Pro account, and letting the team loose. Or, if you’re feeling more confident and you’ve got your security and permissions ducks in a row, you might be ready to enable Claude Enterprise and connect it to your existing Microsoft 365 or Google Workspace environment. You might buy a few Copilot Premium licences, or start experimenting with your own set-up in Azure. I don’t really care which way you do it, but find a way to give staff access to the most capable multimodal, computer-using, coding-proficient AI out there.
Support the team, then let them share
Support this group with additional professional learning. Next year I’m running a cohort for AI leads in schools and universities, whether they’ve been given the role formally or informally, and there are plenty of other opportunities out there for people who want to expand their understanding of what AI can do.
These people then need as many opportunities as possible to share what they’ve learned with their colleagues. An external consultant like me coming in to demonstrate that ChatGPT can make its own videos and steal my voice to narrate them is fine, but it doesn’t get much traction with the teacher who is adamant that their assessment is still AI-proof, for whatever reason. Convincing colleagues that the most capable models on the market can now comfortably handle the knowledge work of the K–12 curriculum won’t be easy. But it’s a necessary part of coming to terms with this technology.
The brutal realities of the capability gap
To come back to where I started: that school leader wasn’t frustrated by their colleagues’ lack of understanding. They were genuinely concerned that if the school didn’t do something to support those staff, they’d be setting themselves up to fail.
There would be an inevitable, growing gulf between those staff and the ones who had accepted what AI means for their disciplines, and a growing gulf between how those staff think their students are using AI and how their students are actually using it. The gap between what people think AI can do and what it actually does isn’t the only thing that matters. These attitudes also tend to produce reactive approaches to AI, such as detection tools or running every assessment under exam conditions. Neither is a suitable or appropriate response.
So closing the gap in understanding what AI can do should be a priority for any education provider. Whatever your discipline, whatever your personal views on the technology, and whatever your approach to teaching and learning, AI will affect the academic fields and industries connected to your subject.
If you’re an English teacher, we’re already seeing how AI is affecting digital media, communications and even traditional publishing. If you teach humanities, you need to be ready to talk about how AI is both helping researchers and implicated in ongoing attempts to rewrite history. In a subject like geography, where academics and industry have used digital technologies and data collection for years, many of those methods can now be augmented or entirely automated by AI.
AI isn’t a chatbot. It’s an interface, an infrastructure, a universal translator between humans and computers. It’s probably the next layer that will sit on top of the internet and all the digital systems that education has only really begun to come to terms with over the last twenty years.
If you want someone like me to come in and demonstrate the shiniest new features of ChatGPT, that’s fine (it’s also my full-time job now, and how I pay for my groceries, so please do keep asking). But don’t expect it to help the staff who still see AI as a free chatbot that’s only good for cheating and writing predictable, easy-to-catch essays.
The years from 2023 to 2026 have been turbulent, and most schools and systems have started to lock in policies and processes. Some of them are flawed, but they’ve laid enough groundwork that we can now get back to what we’re good at: teaching and learning. Executive teams can get out of the way, confident that their colleagues in teaching and learning can handle this, as long as everybody goes in with their eyes wide open about what this technology really is.
Header image: Jamillah Knowles & We and AI / Better Images of AI / People and Ivory Tower AI 2 / CC-BY 4.0
Want to learn more about GenAI professional development and advisory services, or just have questions or comments? Get in touch:
Leave a Reply