What Comes After AI Literacy?

The full narrative

Tim Gander doesn't open with a slide deck. He opens with a question about parking.

How many of you found your way here this morning without using a map? A few hands go up. How many used GPS? More hands. And of those, how many think they could have made it without it?

It's a small, slightly playful moment, but it does real work. Because the difference between someone who could navigate without the map and someone who couldn't isn't just about confidence. It's about whether the knowledge is actually there, built up over time, or whether it was always being carried by the tool.

That distinction runs through everything that follows.

Tim has been working in this space since before most people knew how to pronounce ChatGPT. He traces the arc clearly: 2023 was panic, bans, plagiarism detectors, urgent conversations about academic integrity. By 2024, the consensus shifted to policy and frameworks. Now, in 2026, AI is embedded in teachers' daily workflows, used by government agencies to draft documents, treated as a given. But Tim's concern is that through all of that movement, one question has stayed largely unasked: is any of this actually benefiting learners?

That question is what led him and a colleague to start a community of practice. It began with two people meeting fortnightly. It now has over 800 members. And the questions they set out to explore weren't "how do we make teachers more productive?" They were: how might AI enable effective pedagogy? How might it support inclusive learning? How might it enable equitable outcomes? The word "might" was deliberate. No assumptions. Just genuine inquiry.

What they found wasn't always comfortable.

One teacher's voice from the research stays with you: "Our school is poorly resourced in all areas. Our students are in survival mode. AI is a wealthy school problem." Tim doesn't dismiss this. He holds it up as a necessary corrective. The equity gap in AI isn't just about access to a device or a subscription. It's about the knowledge and skills to use these tools in ways that don't quietly reduce or marginalise learning. And 61% of teachers in the research were unsure whether their students were even using AI at home. Not a confident no. Just: I don't know.

The research Tim shares is pointed. A study by Bastani and colleagues across 1,000 high school students found that learners using AI scored 48% better before an exam. Then the tool was removed. In the exam itself, they scored 17% worse than students who hadn't used AI at all. The AI hadn't taught them the maths. It had done the maths for them.

A second study, from Fan and colleagues, gave 117 university students a tool and asked them not to rely on it too heavily. They all did. The work looked excellent. The learning wasn't there. The researchers called it "metacognitive laziness." Tim also uses the terms cognitive offload and cognitive surrender. The idea is the same: you skip the hard part, and the hard part is where the learning actually happens.

But then there's a third study, only published a couple of weeks before this talk. A large randomised controlled trial run by Google in Sierra Leone, over eight weeks, found learning gains of between 1.2 and 1.8 years. The difference, Tim argues, is in how the AI was used. The tool asked students questions. Students had to respond and think. The AI wasn't doing the work. It was prompting the student to do it themselves.

The implication is significant. It's not that AI helps or harms learning. It's about the design of how it's used.

Tim then turns to his own research, conducted with colleagues at Massey University, looking at 277 students through the lens of the Structured AI Literacy Framework developed by Kathryn McCallum at the University of Canterbury. Of those students, 152 said they hadn't used AI in their assessments. The instinct might be to assume they lacked access or understanding. But that wasn't it. These students had frameworks, guidelines, and tools available to them. They understood AI. They chose not to use it.

One student's response captures it: "I choose not to use AI on my assessments as I want the work to be my own."

This student scored highly on the AI literacy scale. They weren't confused. They were making a considered decision about what they would lose if they handed that work over. Tim describes the moment of breaking through a hard problem — the struggle and the eventual clarity — and says that isn't a side effect of learning. That is the learning. And when students say they want to do it themselves, they're not being precious. They're telling us something important about assessment design. They're telling us that we haven't yet built assessment in a way that preserves the thing that makes it worth doing.

Tim draws on a clip from 1984, a documentary in which a technologist argues that good design is about eliminating uncertainty, and that this impulse extends to eliminating human judgment and human intuition. Tim connects this to Paulo Freire's banking model of education, to bell hooks on education as a practice of freedom, to Loris Malaguzzi's observation that a child has a hundred languages and schools pick one. And to Gert Biesta, the Dutch philosopher, who writes that teaching is always a wager. You cannot control what a student becomes. The risk that they outgrow your expectations, question your certainties, find their own path — that's not a design flaw. That risk is what education is for.

The systems built around education, Tim says, weren't built in bad faith. Standardisation was the only way to scale. You couldn't reach thousands of learners without fitting them into the same-sized hole. But now, for the first time, there's a technology capable of responding to complexity rather than flattening it. The question is whether we use it to run the existing system faster, or to build the system we always wanted.

He uses the image of standing on a rock on the East Coast, fishing. From the shore, someone watching through binoculars might wonder what on earth he's doing out there in the rain. But he knows what's beneath the surface. And that's the point. Assessment, as it currently works, is very good at seeing the surface. It struggles with the thinking that doesn't fit the rubric, the understanding that can't be captured in a mark scheme, the knowledge that is real and present and invisible to the tools we've built to measure it.

Some learners can't bring their thinking to the surface — not because it isn't there, but because the pathway doesn't exist. Different neurologies, different languages, different ways of knowing. Tim describes recent funded research with Ako Aotearoa that built a tool to help schools view assessments through the lens of Universal Design for Learning, opening those pathways. But he's clear that fixing access is only part of the problem. The deeper question is what we're actually trying to see beneath the surface. If the answer is only what the curriculum predicted, then we've made an old system more visible, not better.

The talk closes with three questions for educators to take back to their schools.

The first is personal. Think of a specific student — not a type, a face, a name. What does the current system ask them to compromise in order to be measured? And what would you learn about them if it didn't?

The second is practical. Every school has predictable tasks that AI could handle: report comments, rubrics, administrative work. If AI took those on, what would you do with the time? Not to fill it with more tasks, but to use it for that first student.

The third returns to the map. When the signal drops and AI isn't available, can that student find their own way? What would help them get there?

Tim brings it back to where he started. The pair activity at the beginning of the session asked people to complete two sentences. Person A: "We've been told AI will save teachers time by..." Person B: "But the thing AI can't replace is..."

The first half is real. It's happening. But the second half — whatever people said in that room — that's what Biesta was pointing at. That's what the students in the research were protecting when they chose not to use the tool. That's the beautiful risk of education.

The job, Tim says, isn't to choose between efficiency and that second half. It's to make sure the first serves the second.

And the question isn't whether students use GPS. It's whether, when they do, you're still teaching them how to read the map. Because when the signal drops, they need to know where they are.