What is the most effective starting point for meaningful teacher Professional Development on AI?

It has been said, "you can't outrun AI". There is a truth here, yet I find this mental image exhausting and demoralising. When something is changing as rapidly as generative AI, how can we grab hold of this slippery thing enough to be the guides our students need us to be? Teachers can feel simultaneously exhausted at the thought of staying ahead and guilty because they recognise critical thinking about AI's threats and opportunities is a skill our students need. What is the best way to prepare our teachers to be the informed mentors our students need? How do we equip our learning community for a near-future which could go anywhere–and yet no one actually knows what is ahead?

I am going to be arguing for a pathway to AI professional learning that will be as effective as it will be familiar to educators. Turns out, we've been doing this for a long time.

Spoiler: It is to develop core understanding first. To generate interconnected schemas, iteratively. To begin with nurturing understanding before hammering application. What supports this argument? The biggest themes in learning research.

Having an accurate schema (understanding about generative AI matters... a lot. How we conceive of AI informs how we talk about it. How we talk about it influences the schemas of others and can do unintentional harm. As Sanguinetti (2026) put it, "If we represent AI in misleading ways, we will struggle to truly understand it. And if we don’t understand it, how can we ever hope to use it, regulate it, and make it work in ways that serve our shared interests". I would... add–be helpful guides for our students. So, we need an faithful mental representation of AI (a faithful schema), how do we best go about developing that?

Let's start with cognitive load theory. I can remember John Sweller recalling to a group of us at UNSW how shocked he was as his findings - at time counterintuitive findings - were increasingly validated by the evidence. In a nutshell, human memory is divided into two main parts. Long term memory (which seems almost boundless) and working memory which is limited to a few items (7 ±2) at a time. But, and here is the genius of our brain, a developed schema retrieved from long-term memory takes up only one item or slot.

So a student can retrieve an existing schema about photosynthesis while still allowing six free "slots" for new information about the Calvin cycle. The existing schema both makes sense of the new information and is also updated by it. Iteratively revisiting and developing schemas can make something as fantastic as photosynthesis, possible to grasp. It's only with a solid schema that far transfer is even possible (albeit rare). I love this example of far transfer though, where understanding plant physiology was the inspiration to re-think computer memory systems! The best way to refine and strengthen schemas?... apply them and refine them with new information or experiences.

A schema is a mental framework—a structured cluster of knowledge in long‑term memory that represents a concept, event type, or system. It provides a ready‑made pattern that we draw into working memory to interpret new information efficiently

Takeaway: Schemas are the architecture of understanding which, when iteratively revisited, can build a resilient and agile mastery of the most complex phenomenon. Schemas are the substrate of fluent application.

Schemas are the substrate of fluent application.

Cognitive scientist Michelene Chi's (2006) extensive work on expert-novice differences reveals that what separates experts from novices isn't primarily the quantity of information they hold, but the quality of their schemas. Their knowledge is structured around deep principles rather than surface features, which is exactly what allows flexible application in unfamiliar territory. This ties into John Anderson's (2007) ACT-R framework that procedural knowledge — knowing how to do specific things — is brittle when conditions change. Principled, declarative understanding is what gets recruited when the situation is genuinely novel. A schema which is faithful to the phenomenon it represents is the key to expertise.

Perhaps most relevant, as we consider the dragon of AI pounding down the road behind us, is Schwartz and Bransford's (2005) research on "preparation for future learning". This suggests that even when we can't directly transfer what we know to a new situation, deep understanding dramatically accelerates the learning of genuinely new things. It provides the patterns we can draw upon to interpret new information efficiently. No one can predict which AI tools will matter (or even exist) in three years but teachers with rich, principled schemas won't be starting from zero when the ground shifts–again. They'll be retrieving, connecting, refining, and adapting.

Takeaway: The quality of the schemas is what distinguishes expertise. Schemas built on deep principles (rather than misconceptions) are the best preparation for an unknowable tomorrow.

Finally, Hattie's (2009) synthesis of surface learning → deep learning → transfer is a helpful through-line to bring all this together.

John Hattie's synthesis of decades of educational research describes a learning arc that moves from surface learning — core terms, foundational mechanisms, the basic "what" — through deep learning, where cause-and-effect reasoning and comparison build genuine understanding, and finally toward transfer, where that understanding gets recruited in genuinely new contexts.

Takeaways: Principles before procedures. Understanding before application.

Back to the future

Let's return to our initial question, "What is the most effective starting point for meaningful teacher Professional Development on AI?". We are now ready to propose some guiding principles.

Guiding principles

• Create and maintain a calm culture where trying and making mistakes is encouraged.

  • There are many layers of emotion around AI. Most people are reluctant to admit using to using AI as there is a stigma. Unclear, unformed or prohibitive school policies about AI use can drive it underground. What is your school's policy framework?

  • There are loud evangelists and screaming doomsayers. Avoid them. We are here for the students, to understand this beast so that we can support them. We are here to replace fear with understanding, calmly. I personally hope that "seatbelts" for AI are enacted far more quickly than it has taken (is taking) for social media. Remember that for social media, and literal seatbelts in cars, the path to victory was by developing understanding and expertise.

• Start by providing a schema for an understanding of Generative AI. This forms the "seed crystal" that subsequent experiences and learning form around.

  • Look for expertise in AI's fundamentals that is paired with the skill to make complex science, simple. This is not the place for novice leading novice.

    • Grant Sanderson is a mathematician specialising in visualisations with some very helpful ones of generative AI- his target audience here are software engineers, but this one is more accessible.

  • Go slowly at first and try and remove the anxiety around needing to apply this too early.

• Design (or seek) training that is not only true to the deep principles of the phenomenon, but also to pedagogical principles which are evidence-based (see above).

• Provide longitudinal opportunities for applying and sharing the application of generative AI.

  • Perhaps the best of these will be informal staffroom-based (impromptu) exchanges - how can you increase the likelihood of these arising?

  • This is the place for novice sharing with novice!

  • Remember John Dewey, "We do not learn from experience... we learn from reflecting on experience.”

  • Schemas have to regularly "bump against" the reality they are seeking to capture. This not only strengthens the schema as it is interprets the new experience, but the schema itself gets refined by this process.

• Minimise shallow and brittle procedural learning, "try this tool.. click here, here ,...", and especially, "try our new AI product [billing structure revealed later]". It might feel good to say, "On Monday all teachers used AI to do x" - but it is unlikely to be effective in achieving any meaningful professional learning goals.

If you'd like to see these principles applied, you might enjoy Taming the AI Dragon. In Module One, we explore the science of generative AI to start building that schema, and in Module Two we apply it to the most urgent question for teachers, "How do we meaningfully assess learning in the age of AI?". These professional development courses could start with a live (in-person) session at your school, so reach out to start a conversation.

Bibliography

Anderson, J. R. (2007). How can the human mind occur in the physical universe? Oxford University Press.(Contains ACT‑R overview; canonical source for the procedural vs declarative distinction.)

Barnett, S. M., & Ceci, S. J. (2002). When and where do we apply what we learn? A taxonomy for far transfer. Psychological Bulletin, 128(4), 612–637. https://doi.org/10.1037/0033-2909.128.4.612

Banerjee, R., & Chakrabarti, B. K. (2009). Learning & intelligence of plants: Developments following Jagadish Chandra Bose. Saha Institute of Nuclear Physics. https://www.saha.ac.in/cmp/camcs/banerjee_chakrabarti_PhysicsNews09.pdf

Chi, M. T. H. (2006). Two approaches to the study of experts’ characteristics. In K. A. Ericsson et al. (Eds.), The Cambridge handbook of expertise and expert performance (pp. 21–30). Cambridge University Press.

Hattie, J. (2009). Visible learning: A synthesis of over 800 meta-analyses relating to achievement. Routledge.

Hattie, J., & Donoghue, G. (2016). Learning strategies: A synthesis and conceptual model. npj Science of Learning, 1, Article 16013. https://doi.org/10.1038/npjscilearn.2016.13

Sanguinetti, P. (2026, January 7). We’re talking about AI all wrong. Here’s how we can fix the narrative. The Conversation. https://theconversation.com/were-talking-about-ai-all-wrong-heres-how-we-can-fix-the-narrative-272752 [digitalinf...nworld.com]

Schwartz, D. L., Bransford, J. D., & Sears, D. (2005). Efficiency and innovation in transfer. In J. Mestre (Ed.), Transfer of learning from a modern multidisciplinary perspective (pp. 1–51). Information Age Publishing.(Major source for “preparation for future learning”.)

Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257–285. https://doi.org/10.1207/s15516709cog1202_4

Sweller, J., van Merriënboer, J. J. G., & Paas, F. (1998). Cognitive architecture and instructional design. Educational Psychology Review, 10(3), 251–296. https://doi.org/10.1023/A:1022193728205

Sweller, J., Ayres, P., & Kalyuga, S. (2011). Cognitive load theory. Springer. https://doi.org/10.1007/978-1-4419-8126-4