Perspective by Richard Alberg, CEO, Aptem

What are your thoughts on the use of AI for apprenticeship and training support? At the start of 2023, you might have struggled to find someone that had heard of ChatGPT. You would have had a harder job finding someone who what it could do. Throughout the first months of the year, word spread at office coffee machines and dinner tables about this new tool tool. It supposedly could generate believable written content, images, and technical code, from just a few words. It was one of those things that had to be seen to be fully appreciated. As the intrigue grew, conversations became dominated by speculation about the future. What would this technological leap forward mean for businesses and individuals all over the world?  

ChatGPT on its own was mind-blowing. But for me the lightbulb ‘this changes everything’ moment was when Microsoft demonstrated GPT functioning within Office 365 Apps and using OneDrive data. This opened the possibility of an app, like Aptem, using the power of a Large Language Model (LLM), integrated with Aptem functionality and the structured data in the Aptem database. Whatever could be done standalone using GPT would be dwarfed by the app, the LLM and the database all working together. Fast forward to 2024, and for many, using AI for training is invaluable. The tricky part is now in identifying how to leverage these tools. We need to take them from a novelty or ad hoc productivity tool. We can use it to create a well-conceived, well-embedded strategy that delivers true value.

Supporting a struggling sector

It’s no secret that the apprenticeship sector is struggling. Being asked to do more with less is nothing new for providers. But, with stagnant funding and ever-increasing costs, it has become harder and harder for providers to maintain their quality of learning delivery. Organisations offering apprenticeships are doing an excellent job at adapting given the circumstances. However, many are reaching their limits in terms of capacity. There’s only so far conventional strategies such as increasing caseloads can be stretched to close the growing gap between the resources available to providers, and the high-quality learning opportunities that apprentices require and deserve. 

LLMs and AI hold the key to bridging this gap and supporting future training. This new technology creates a step-change for organisations, only seen once in a generation. By complementing internal expertise with AI-powered tools, providers will be able to unlock new levels of efficiency. They’ll be able to make time for more meaningful relationships with learners, and increase the quality of teaching for all.

The opportunities and applications for LLMs within Apprenticeships 

We have all been amazed by the way that GPT can eloquently, appropriately and accurately answer a broad range of questions. Many of us at Aptem experimented with questions related to vocational training. It was hard not to be impressed by the quality of output. It quickly became clear to us that there are some obvious applications of LLMs within the sector. Using AI for training could help tackle admin-heavy, repetitive tasks, including, for example:  

• Assessment of learner progress 
• Assistance with marking submitted work 
• Summarisation of topics, feedback and reviews 
• Providing guidance for tutors and learners 
• Adapting content according to personalised learning need 
• Validation for compliance 
• Natural language reporting 
• Identification of examples and data for Ofsted 

However, as many of us have found in our early experimentation with LLMs, the output is only as good as the information available to the model. Oftentimes, the most impacting factor was the quality of the prompt or question posed. The detail and precision of language used makes a significant difference. It would be almost impossible to achieve the above list of tasks in a consistent quality way simply using standalone LLMs or products outside of the context Aptem has available. 

The importance of context 

With LLMs, context is essential. We’ve all heard how certain versions of ChatGPT struggle to retrieve simple answers such as who is the current Prime Minister. It can only pull from the information fed into it.

In order to create content that providers can trust, there’s a wide range of contextual considerations. Take a learner currently enrolled on a particular apprenticeship. To provide any meaningful guidance, as a minimum the LLM would need to have:

• Detailed contextual understanding of the apprenticeship sector and the role of each stakeholder in the learning journey. 
• Awareness of every part of the curriculum for the relevant Standard, including every KSB. 
• Information about the learner themselves. Where are they in their apprenticeship? What learning have they already undertaken? Which KSB are they currently working towards? 
• A well curated prompt. What is it we’re asking the model to output? 

This would be very difficult to realise in a standalone LLM.  

Luckily, as a platform, Aptem contains all of this data and more. This rich, secure source of data provides the perfect backdrop for tackling these tasks in a way that is insightful, accurate and personalised.  

The benefits of an in-platform approach to AI 

We’ve already seen that context is key when generating meaningful responses. The Aptem platform is an ideal setting within which to embed language technology. It has a breadth of apprenticeship knowledge and detailed information about each learner’s personal learning journey.  

Language technology

Incorporating language technology into Aptem also gives us the opportunity to tackle some of the roadblocks that have caused some providers to pause in their exploration of AI technology. As generative language tools, platforms like GPT create their responses, based on an extensive set of data. This gives them the potential to misinterpret information. In some cases, they make claims that aren’t true. This phenomenon has become known as hallucination.

LLMs within Aptem are controlled by Aptem. This means that we can avoid usage prone to hallucinations and we can incorporate the techniques that identify or minimise hallucinations. We have access controls that can evolve as researchers develop tools to manage the challenge. This is a far more secure approach than learners and tutors doing their best with a standalone LLM.

It is essential that providers feel they can trust the output. So, we are putting mechanisms in place to flag responses that don’t quite hit the mark. This is helping us to introduce standardised and proven solutions that improve the quality of responses for all Aptem users.  

Security

Another potential concern for providers is security. OpenAI, with GPT, is just one of many organisations offering LLM technology. Each technology provider has its own policies as to what data it uses to train its model and what data is saved and where. Aptem will only have a contract for service with reputable providers with appropriate data guarantees. And we also control what data is sent to our LLM provider. This contrasts with tutors and learners who, outside of Aptem, can choose to access systems that take their fancy, including ones without protections or who breach GDPR. By using language technology embedded within the Aptem platform, data remains secure. Providers can be confident that they remain in control of their data and its use.

Our approach to incorporating AI into the Aptem platform 

Working with language technology is relatively new, and inherently challenging. We have trained and hired staff to create a team of specialists in the field of machine learning, prompt engineering, user experience and data science. We are supplementing our existing strengths in software engineering and colleagues with extensive industry experience. Our team that is more than capable of delivering this much needed lifeline for the sector.  We are driving the industry by adapting AI for training.

Our approach is to create a series of value multiplying features in Aptem Enhance®. These will stack to create greater and greater impact. The first of these solutions, Aptem Checkpoint®, was launched in late 2023. It provides an objective measure of learner progress. This learning tool leverages the extensive knowledge bank within Aptem to generate knowledge-check questions for apprentices throughout their programme. As a result of learners completing Checkpoints, learners, tutors and providers as a whole gain a much clearer and more objective understanding of progress against the requirements of the Standard. The use of generative AI makes this possible without requiring any time for tutors. They won’t need to create materials or quiz their learners in a 1:1 session.

We’re not interested in adopting AI for the sake of it – it has to have a genuine impact for our customers. Early on, we enlisted the support of some of our Aptem customers to join us as early adopters of our first AI-powered enhancements. Their feedback has been invaluable in providing real-life use cases, questions and challenges. This allows us to refine Aptem Checkpoint further. Together, we’re creating a tool we are confident will greatly impact learner achievement.

In addition, we are currently running a series of detailed consultation sessions. These sessions include internal stakeholders, industry experts and a range of Aptem customers. We want to gauge the sector’s most pressing challenges. This ensures our roadmap delivers maximum positive impact as quickly as possible.

What we’re not trying to do with AI 

As we explore what AI can achieve, we’re also mindful of the role people play in the apprenticeship journey. Apprenticeships are about sharing knowledge, skills and behaviours between experienced professionals and learners. Our goal with these solutions is not to replace roles such as tutors and skills coaches. Instead, we allow them more space to do what they got into teaching to do – teach. 

For example, we can reduce the time taken to gather data in preparation for a review. We can give tutors more time to pull together meaningful discussion points and thoughtful suggestions for stretch and challenge. With Aptem, tutors aren’t so focused on summarising notes and inputting information. So, they will have the space for more 1-2-1 time with each learner. If AI can notify tutors when it looks like a learner is beginning to struggle with their training, they have the best possible chance of offering support. This intervention can also happen when it can have the greatest impact.

Amongst all of this, it’s imperative that the learner’s needs remain at the heart of what we deliver. We all share the desire to help providers become more resilient and profitable. However, we agree with the sector that this cannot come at the expense of the learner experience. The use of AI for training support cannot ignore the humans at every stage of the process.  

Summary 

I am more hopeful than ever that 2024 will be an exciting and transformative year for apprenticeships. The sector is realising the myriad of benefits that language technology can offer. This technology unlocks a level of productivity and efficiency never seen before. We’re delighted to be at the forefront of bringing this innovation to the sector. 

For more information about Aptem, or to speak to us about how we’re harnessing the power of AI within our platform, get in touch.