AI in teacher professional development: The unnecessary and absolutely necessary cognitive load of developing as a teacher

ROBERT CAUDWELL AND PAUL MALLABAND, CO-FOUNDERS, PENROSE EDUCATION, UK

Introduction

As is true in most professions, teachers are increasingly outsourcing elements of their work to AI (artificial intelligence) tooling. As AI capabilities continue to improve, it will become possible to increasingly rely on machines to do parts of our jobs for us. This has the potential for significant benefits – from reduced workload for teachers to increased personalisation of learning experiences for pupils. However, this article will highlight a fundamental question about the use of AI that demands careful consideration. In a world where teachers and school leaders could outsource more and more thought to machines, what do we want to ensure that we are still thinking about? And, crucially, what implications does how we answer this first question have for how we train, educate and develop teachers?

Unnecessary vs necessary cognitive load

A foundational principle of cognitive scienceThe study of the human mind, such as the processes of thought, memory, attention and perception is that (with some exceptions) to learn a new thing, we must think about it first. To successfully store something new in our long-term memory, we must first pay attention to new information from our surroundings and then process it using our working memory (EEF, 2021a). This mental effort to grapple with new information is often called ‘cognitive load’ (Neelen and Kirschner, 2020, p. 216). Thinking about new things is hard work, and a person’s working memory can only handle so much new information at once before it becomes overloaded (Weston and Clay, 2018).

In education, one common application of cognitive load theoryAbbreviated to CLT, the idea that working memory is limited and that overloading it can have a negative impact on learning, and that instruction should be designed to take this into account is to encourage the reduction of unnecessary load on the working memory. For example, the Education Endowment Foundation’s ‘Guidance on effective professional development’ states that ‘strategies that reduce the strain on [working memory], which reduce the burden on teachers’ thinking, can promote better learning’ (EEF, 2021b, p. 15). However, we must not forget that the intention here is not to eliminate the need for thought, because learning often only happens through thinking hard about something new (Kirschner and Hendrick, 2020). In other words, the goal is to reduce unnecessary cognitive load so that the learner can focus their mental resources on the absolutely necessary cognitive load required to learn something new.

A framework for using AI to support professional training and development

When considering how artificial intelligence might support professional learning in teachers, we must pay close attention to both halves of this cognitive load puzzle. What is the unnecessary load, for which an AI tool might be able to offer support? And what is the necessary load that we need to protect? 

There are some emerging ways in which AI might helpfully ease unnecessary cognitive load during professional development. This will be an ever-growing/changing list(!), but some promising examples that we have already seen are:

1. Rehearsal: Using AI-powered simulations with AI-powered feedback could help teachers to reduce the considerable cognitive load associated with facing a scenario for the first time. Imagine being able to teach the exact same ‘class’ multiple times and gradually learn from the mistakes that you made the previous time. The potential benefits here are obviously dependent on the quality of the simulations and the feedback! But there are some promising studies suggesting that this use of AI can have discernible benefits (e.g. Sailer et al., 2023).
2. Reflective practice: Using AI-powered journalling tools to help to guide your reflections on your own practice can reduce the cognitive load associated with metacognition. Thinking about your own development and learning is not always easy. AI tooling that could guide and prompt reflective practice could free up cognitive resources to focus thinking on the reflective focus. A recent study showed the positive impact of using an AI ‘coach’ during school placements (L’Enfant, 2024). (See the work of the Noticing Network for some really interesting and thoughtful examples of how this could work.)
3. Feedback: AI can be used to give instant feedback on teaching practice. AI tools that will help teachers to analyse and reflect on video recordings of their own practice, highlighting certain key moments and drawing in relevant theories, already exist. Again, the impact of this will depend on how accurate and helpful the feedback proves to be – we should know more soon. (See GoReact’s new AI Assistant as an example of this type of functionality.)
4. Sequencing: AI outsourcing can be used to support the sequencing of learning. For example, while we may want trainee teachers to eventually learn how to plan great sequences of lessons (more on this later!), we might want them to first focus their attention on practising a narrower skill (e.g. tailoring content to pupils with special educational needs and disabilities). AI tooling might offer ‘good enough’ alternatives so that cognitive capacity can be focused on the specific thing on which we want to focus thought at a particular stage within a sequenced curriculum.

However, alongside these potential benefits, there are potential dangers. In a world where more and more thinking could be outsourced to AI-powered tools, we must also think very carefully about what we choose not to outsource. This is especially true if the claims of cognitive science are correct: that we only learn what we have thought deeply about. When we outsource thinking, we also outsource the potential for learning associated with that thinking. We have condensed these issues into a framework of three core categories of question:

• What thinking is unnecessary and safe to outsource? What are we happy for teachers to rarely or never think about (and therefore potentially not ever learn for themselves)? 
• What thinking is necessary and should never be outsourced? What do we want every teacher to be thinking deeply about? (Crucially, this may have a very different answer to ‘what is it not possible to outsource?’.)
• If unnecessary cognitive load can be reduced, what do we hope to do with our freed-up cognitive capacity? What do we wish that teachers had more time to think deeply about?

How we answer these questions has significant implications for how we approach teacher education, training and development. We should be designing teacher development programmes and qualifications that make the most of the help that AI tooling can offer. But we must also ground all teacher development on the foundational principle that people cannot expect to learn things that they are not thinking about.

An example: Unnecessary vs necessary thinking in lesson planning

To help to explore some of the implications of the framework questions above, we will use the example of how AI is or could be used to support lesson planning. There are probably some in the sector who would gladly (and do!) outsource the entire lesson planning process to AI. There are also probably others who believe that lesson planning is so intrinsically connected to ‘what it means to be a teacher’ that they would refuse to outsource any part of their planning process. However, there is evidence that many teachers will be somewhere between these two ends of the spectrum.

A recent trial commissioned by the EEF and evaluated by the NFER found that AI tools could reduce lesson and resource planning time by 31 per cent, and found ‘no reduction in quality’ of the lesson resources (Roy et al., 2024). The study found that this time saving was being made by outsourcing work to ChatGPT (the model used in the study) on very specific tasks. Teachers did not, in general, use the AI to plan the entirety of a lesson but focused on specific aspects, such as creating extra questions for a target group in the class or generating ideas for activities for a specific purpose. It’s impossible to say whether these narrow time-savers are the only aspects of lesson planning that ever should be outsourced, but it does seem that experienced teachers tended to not outsource thinking when it came to overall lesson design, sequencing or ensuring that the specific needs of the class in front of them were met (e.g. the teacher identified the need to have an adapted set of resources for certain groups, but outsourced the actual creation of these resources). Establishing what thinking should and should not be outsourced when planning lessons is not straightforward.

This is further complicated when considering that the unnecessary–necessary distinction will not be the same for all teachers in all contexts. For example, the same EEF study referenced here also considered whether teachers with different levels of experience used ChatGPT in different ways (Roy et al., 2024). Fourteen per cent of the participants were ECTs (early career teachers) or trainees, and they were more likely to ask ChatGPT for ‘starting’ ideas for how to teach a particular lesson, where more experienced teachers sought to use the AI tooling to expand and vary their teaching repertoire. While this is unsurprising, it is entirely possible that because the more experienced teachers have previously spent considerably more time thinking about lesson planning, they have a much more sophisticated ‘schema’ of the learning process and effective lesson design upon which to draw (Krepf and König, 2022). What if – because of the ongoing use of AI – the ECTs and trainees in this trial never have the opportunity to develop their thinking about lesson design in the same way? What may be unnecessary thinking to an experienced teacher may well be absolutely necessary thinking to someone at the beginning of their career. When considering what thinking we are happy to outsource and what we need to protect, there will be local, contextual and even personal considerations to take into account.

Finally, assuming that we are able to identify ways in which to carefully and thoughtfully reduce the cognitive load associated with lesson planning, how could any regained cognitive capacity be used? Again, the EEF/NFER study gives plenty of food for thought. One example worth highlighting is the finding that some teachers used ChatGPT to rewrite whole texts to be more suitable to a lower reading age – something that could take a huge amount of time for a human to do (Roy et al., 2024). By doing this, the teacher can draw on a wider range of resources and easily adapt them, thus being able to spend much more time focusing on lesson design rather than construction. Perhaps by reducing cognitive load elsewhere, teachers might be able to spend more time thinking about the individual needs within their class and less time trying to adapt resources in just the right way. As we consider questions about what is and is not necessary for teachers to think about, we should also consider what might be possible to do with any freed-up cognitive capacity.

Conclusion

[I]t is important to recognise that the digital automation of education is not simply a technical matter of how to most effectively design, program and implement systems. Instead, we need to get to grips with questions relating to the nature of education as a profoundly social (and therefore human) process.

Selwyn, 2021, p. 165

Cognitive load theory can be the basis for a helpful framework for considering the use of AI to support teacher training, education and development. If overloaded cognitive capacities reduce teachers’ capacities to learn, then we absolutely should be open to how AI might support the reduction of unnecessary cognitive load. However, if thinking leads to learning, then an obvious consequence of teachers no longer thinking about something is that associated learning may also be lost. Therefore, we must also be careful to identify what is the absolutely essential cognitive load of teaching, and ensuring that teachers and school leaders continue to think deeply and carefully about these fundamental issues. This is what underpins the framework that we have offered in this article: a balance between reducing unnecessary cognitive load and identifying the necessary load of what it means to be a teacher. This can be summarised in three broad categories of questions: 1) What are we happy for teachers to rarely or never think about? 2) What do we want every teacher to be thinking deeply about? 3) What do we wish that teachers had more time to think deeply about? By first wrestling with these issues, we will be much better positioned to use AI to support the authentic development of teachers and school leaders.

The examples of AI use and specific tools in this article are for context only. They do not imply endorsement or recommendation of any particular tool or approach by the Department for EducationThe ministerial department responsible for children’s services and education in England or the Chartered College of Teaching and any views stated are those of the individual. Any use of AI also needs to be carefully planned, and what is appropriate in one setting may not be elsewhere. You should always follow the DfE’s Generative AI In Education policy position and product safety expectations in addition to aligning any AI use with the DfE’s latest Keeping Children Safe in Education guidance. You can also find teacher and leader toolkits on gov.uk .