Categories
AI in education
5 September 2024The pedagogical thinking behind Aila, our AI lesson planning tool
John Roberts
Product and Engineering Director
Ensuring pedagogical integrity with Aila
Why have we built Aila?
Teachers are busy! Lesson planning for our classes is essential but requires a great deal of thought and time to do well. Plenty of AI tools are available to make tasks quicker and easier, but we know that planning is nuanced and requires your expertise.
We built Aila, our AI-powered lesson assistant, to support teachers with some of the more time-consuming aspects of lesson planning, adaptation and resource creation. We wanted to make this process more efficient, giving you the time and headspace to think more deeply about other things that require your attention, such as refining your lesson delivery.
When developing Aila, we knew that for it to be impactful for teachers and pupils, the lesson needed to be well-designed, the outputs had to be accurate and safe, and the content had to be appropriate for UK schools. To achieve this, we have put pedagogical rigour at the heart of its design and ensured that the teacher - as the expert - is always kept firmly in the driving seat.
Foundations in research
A wealth of research has informed Oak's six curriculum principles, which have been fundamental to all our lesson and resource design.
The lesson planning process that Aila will take you through is designed to emulate the thought process of an experienced teacher as they plan a lesson. Aila starts by working with you to define a learning outcome, then chunking learning into manageable chunks - ‘learning cycles’.
Aila will support you in identifying the prior knowledge that pupils will need to access the lesson, building this into a starter quiz before suggesting the keywords and key learning points.
Aila will also highlight common misconceptions that pupils may hold about a topic so that these can be addressed during the lesson. You will then create each ‘learning cycle’ - a simple explanation, some questions to check for understanding and a chance to practice. Aila can help you create model answers or success criteria to share with pupils as part of their feedback.
Sweller’s Cognitive Load Theory, Mayer’s Theory of Multimedia Learning and Rosenshine’s Principles of Instruction have also been integral to our lesson structure and the design of our resources. Research shows that busy resources are harder for pupils and teachers to use because of the cognitive load required to process the information. Our slides, worksheets and quizzes are simple, consistent and easy to adapt, making them quick to produce and easy to use in your classroom.
How have we improved the quality of Aila’s content?
We have used our research base to codify each lesson component within our AI prompt to align with our curriculum principles. This includes clear guidelines on what to include and exclude in different lesson sections. For example, key learning points should be succinct and knowledge-rich, or teacher-talk such as ‘Now we are going to look at the periodic table…’ should not appear on slides to reduce the cognitive load for pupils.
We currently use OpenAI’s GPT4o as the LLM (Large Language Model) behind Aila. This is the same LLM that powers the latest version of Chat GPT. An LLM is a type of AI model that can interpret and respond to human-like text based on a large database of content.
We’ve tailored GPT4o to meet the needs of teachers by using a pedagogically rigorous prompt. The prompt is the instructions we give to the LLM to ensure it returns appropriate, high-quality content.
LLMs work much better when examples and non-examples accompany the instructions within the prompt, so we have included these in the prompt too.
For instance, when creating multiple-choice questions, Aila is instructed to create distractors that are plausible and are of a similar length to the correct answer and highlight common misconceptions. This instruction is followed with a good example;
“What is the periodic table? (A) a table that lists all chemical compounds, (B) a table that shows all chemical reactions or (C) a table that shows all known elements” and a non-example; “What is the periodic table? (A) a table that lists periods, (B) an old wooden table, (C) a table that shows every known element that has been discovered”.
Our code is open source, so if you’re technically minded, you can see examples of this in the prompts we use within our code. There are also some more technical aspects to improving the quality of Aila’s content, including Retrieval Augmented Generation (RAG) and content anchoring.
So, will Aila do all of the hard thinking for you?
We noticed that many lesson planning tools ask you to input a small amount of information at the start, and then they plan your entire lesson for you. You know your class the best. You know what they will find challenging, where they will thrive, and what they will enjoy. You have probably taught similar lessons previously, and this subject-specific expert knowledge is essential.
Aila is powered by AI, but LLMs use pattern recognition to decide on their output - they are not intelligent humans, and certainly not expert teachers. We’ve designed Aila to pause intentionally during the lesson planning process, working with you to tweak and edit content to make sure it is suited to your pupils and your context.
You can ask Aila to reduce the literacy level of your keywords, add additional support such as sentence starters and success criteria to tasks where your pupils might need it or swap in tasks you have previously used instead of those suggested by Aila. It is your lesson; use Aila to help craft something you love rather than something generic and off the shelf.
So what next?
This is the most commonly asked question: what will we do next to improve Aila?
Our next step is to elevate the quality of content produced for specific subjects and key stages, incorporating the nuance of each. We know that a high-quality practice task in year 1 science will look very different to one in a year 11 citizenship lesson and that a PE lesson should include a warm-up and cool-down.
Music lessons should have a focus on listening, composition and performance and so we want to build these subject-specific refinements into Aila going forward.
Our initial focus is on maths. LLMs notoriously struggle with producing high-quality maths output because, unlike subjects such as English or geography, where the individual words chosen for an answer can vary, maths requires the answers to be 100% correct.
We are working to incorporate some additional steps into Aila to improve the quality of maths content produced, specifically focusing on quiz questions as a starting point.
We are also working on translating the richness of an explanation into our slides. We know that teachers will use a range of images and text to explain a concept to their pupils and that this is an area we need to develop further within Aila over the coming months.
Teachers across the country have tested Aila during the closed beta phase of development, which has been invaluable in ensuring output quality. If you would like to be involved in improving Aila in the future, you can register your interest in joining a new early feature tester group!
Plan a lesson with Aila
Explore Aila, your AI lesson assistant, today or watch the demo video in this blog to see it in action. It is, and always will be, free - so why not take a look around to see what Aila can help you with?
Get updates to your inbox
Sign up below to be kept updated on our latest AI and curriculum developments and get other helpful content by email. Unsubscribe at any time. Read our privacy policy.