Student engagement in effective AI use: Prompt engineering and individualised learning

MICHAEL SPIERS, SENIOR HOUSEMASTER, RIPON GRAMMAR SCHOOL, UK

Rationale and preliminary research

I first became aware of the potential of artificial intelligence (AI) in aiding the incorporation of creativity in teaching in late 2022, after the release of the large language model (LLM) ChatGPT. I was familiar with generative AI from conversations with colleagues involved in coding and web development, and had marvelled at the primitive yet remarkable images created using the Dall-E platform one year earlier. Having read Chris Dede’s seminal article ‘The promise and perils of artificial intelligence in education’ (2023), I was aware of the profound impact that AI could have on facilitating personalised learning and enhancing student engagement. If AI could provide students with an individualised learning platform and a dynamic akin to one-to-one tuition, it could, as Sal Khan believes, go some way towards solving Bloom’s ‘2 sigma’ problem (Bloom,1984).

With AI platforms, including the aforementioned LLMs, permeating into popular discourse away from the parameters of a once very obscure and even derided field of academic research, I decided to have recorded conversations with an exceptionally technologically adept cohort of A-level historians about their own interactions with AI. It became evident through these interviews that students were reporting mixed successes in engineering accurate or useful responses to their prompts. The greater the clarity and specificity of the prompt provided, the greater the accuracy and relevance of the response from the LLM. To understand the dynamics of prompt engineering, I looked outside the field of educational research and found the article ‘Artificial intelligence prompt engineering as a new digital competence’ (Korzyński et al., 2023) very useful in establishing the case for educating people in the mechanics of effective prompt engineering. The research of Shehri et al. (2023) in exploring the mediating role of prompt engineering in education provided a more acute focus on the relationship between learners and LLMs. This preliminary reading led me to consider whether research could be undertaken relating to student engagement with AI and the use of prompts to aid individual learning.

The initial plan for the research was devised with the LLMs ChatGPT and Google Gemini, made available to a group of boarding students aged 13 to 15. Out of the 10 students, four had experiences in using generative AI and the rest had not used any AI platforms.

The small-scale experiment

The idea behind the small-scale experiment was to test whether students could learn about unfamiliar concepts more effectively through engaging with AI platforms than through the more traditional method of studying a chapter in a textbook. This approach tallied with my own preference for the Trivium framework of classical education, advocating grammar, logic and rhetoric stages, more recently relabelled as ‘flipped’ or ‘inverted learning’ by Bergmann and Sams (2012) in their advocacy of students learning about coming topics via videoed lectures, freeing up lesson time for hands-on chemistry. AI has the potential to contribute to all three stages of the Trivium: in acquainting students with a foundation knowledge of an unfamiliar topic; in allowing students to critically assess the construction of arguments and defend their own viewpoints; and, finally, in expressing their understanding through longer written exercises.

Methodology

Students were given a sheet of 12 prompts to use when interacting with the LLM, with X used to denote any topic that they would study. These prompts ranged from the simple ‘tell me more about X’ or ‘give me five interesting aspects of X’ to the more complex ‘explain X through an analogy linked to my favourite sport/activity Y’ (with Y denoting their own choice of leisure activity) or ‘explain the concept of X in the style of JRR Tolkien’. These prompts, which could be individualised, were engineered to scaffold the learning process and provide a greater level of engagement. Thus, the students could then choose the prompts that aligned with their preferred working style. Two unfamiliar topics were chosen for this research: the reasons for the fall of the Roman Empire and the formation of volcanoes on certain plate boundaries. Roughly two-thirds of the students opted to learn about the fall of Rome, with the remaining students opting to engage with vulcanology. An important extra prompt was given to the students: ‘Do not make up any of the responses.’ This was a simple measure to avoid any AI ‘hallucinations’, which occur when the LLM encounters an area with which it is not familiar and seeks to embellish the narrative, as opposed to explicitly referencing its limitations. Sebastian Farquhar is one of a growing number of academics who have investigated this quirk of AI (Farquar et al., 2024), and his and his colleagues’ findings and suggestions will no doubt serve to guide educators as AI becomes ever more present in our schools.

The interaction

Throughout the session, lasting approximately one hour, the students were heavily engaged and had great enthusiasm for choosing the prompts that they considered to be most interesting and most conducive for learning. The fall of Rome was made instantly accessible by students opting to learn about the Vandals and the Suebi in the style of a JK Rowling Harry Potter book. Students were asked to try a number of prompts, and in the event of not understanding an aspect, were instructed to ask for a simplification or the use of a metaphor or analogy to help them to understand. One student, who would describe himself as a committed mathematician, asked for the story of Rome’s fall to be presented as algebra. An example follows: D = (Decline in Military Strength) would include the loss of discipline, training and the increasing reliance on mercenaries. The LLM then presented other factors leading to Rome’s fall, with the eventual outcome summarised as F = (D . E) + (G . I) + C. For the purpose of balance, the same student also decided to turn the process of volcano formation at a constructive boundary into algebra, with the equation V = (P . M) + (R . M). It became clear that the process of communicating with the LLM was enabling the students to utilise their areas of interest and learning preferences to access these unfamiliar topics.

Post-interaction evaluation

After the session, students were asked to complete a 15-mark multiple-choice test of GCSE-level questions based on the topic that they had studied. The results were very encouraging, with a mean percentage score of over 70. With such a small cohort, and no comparative study of textbook learning running in parallel, I deemed it necessary to concentrate on qualitative data and student interviews. Students were interviewed in the days after the experiment and all students spoke positively of the experience, with the most common theme being AI facilitating a kind of personalised learning that traditional textbooks cannot afford the students. When questioned on the effectiveness of prompts, the students enjoyed using the outline prompts to develop their own learning paths. As they refined the prompts, their confidence grew and they started to provide more definite parameters to the LLM, which, in turn, led to more effective responses.

Reflection and advocacy 

This small-scale experiment, undertaken towards the end of the academic year, has had a great impact on the learning of the participants. Their familiarity with prompts and their understanding of what effective prompt engineering entails stand them in excellent stead when asked to undertake a research task or when their natural curiosity leads them to explore curriculum topics in greater detail.

Since the study was undertaken in July, advances in AI have resulted in students now being able to use advanced voice functions on the LLMs to interact in an even more sophisticated manner. The latest video mode on ChatGPT-4o can be trained, via prompts, to provide a commentary on insects living on a houseplant in the style of David Attenborough.

The outcome of this study reinforces the suggestion that learning about AI should become part of the school curriculum. With some rather conservative estimates claiming that AI will affect 40 per cent of the job market (Georgieva, 2024), it is vital for students to be taught prompt engineering via the IT curriculum. As a sector, compared with the sciences, finance and healthcare, education finds itself behind the curve on the potential of AI. Unless students are educated in effective AI use and we engage in metacognitive conversations regarding AI as a learning – as opposed to a cheating – tool, then it will remain as a taboo subject for many educators and serve to act as an inhibitor for learning, when in reality it could, when paired with the classical method of Trivium, revolutionise secondary education.

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 .