AI: an invisible revolution in workplace learning

Developments in technology are making a huge impact on our daily lives. Yet the use of technology in workplace learning is still – relatively-speaking – in its infancy. That’s about to change, in a technology revolution that will transform the way we plan learning and manage knowledge at work. That revolution will be driven by artificial intelligence – AI – which will power solutions that boost the competency of individuals and the performance of organisations. The only thing holding it back will be the humans who work alongside it, argues David Sharp, CEO of International Workplace.

Learning and development has made slow progress in adopting new technologies. Instructor-led (aka face-to-face, or classroom) training has long been considered the best and most effective method of delivery. We’re used to it. It’s flexible and responsive. It’s controllable. And we understand it.

The development and use of eLearning courses has grown rapidly (my company has been delivering IOSH-certified courses by eLearning since 2008). Technology has embedded itself into the fabric of most organisations in the form of the ubiquitous LMS (learning management system). And yes, the pandemic has accelerated the use of virtual classrooms viaTeams, Zoom and the like to allow training to continue when physical training is not an option.

But early eLearning content was a primitive copy of the classroom course, littered with tick boxes, multiple choice questions and quizzes to replace trainer interaction. Too many of these glorified Powerpoint presentations still exist to this day, masquerading as eLearning libraries. While the LMS has been a go-to learning management tool for L&D professionals, many such systems still rely on the 27 year-old SCORM framework to capture learner data.

What all of the above have in common is that technology has been employed to assist learning for the benefit of the organisation, not its learners. It is time for learning and development to fully embrace technology, rather than merely tolerate it.

The facilities management sector sits at the centre of the debate about employee health and wellbeing, productivity and the impact of the workplace on individual and organisational performance. Learning and development activity is often still managed with administrative inputs and outputs in mind however, rather than performance outcomes.  

Course content is housed in an LMS. L&D will decide which employees should be given access to which courses. They will then enrol them, check that courses are completed, and record performance. Effectiveness is measured in terms of enrolments, completions and scores.

This approach is all about organisational compliance, not learner or organisational performance. The result is what EdTech and AI evangelist Donald Clark describes as a ‘Zombie LMS’ .  

The LMS for most organisations is like a dungeon. Things go into it and nobody likes going there. Why? the problem is that most of the LMSs are just lists of courses. That’s all that’s in them. And that’s a real shame as it cripples the use of more innovative approaches, like use of AI or VR …The LMS is the enemy of innovation, they’re just big course-based cul de sacs.

The primary purpose of health and safety regulation is to set a minimum standard that employers must meet in order to ensure the health, safety and welfare of their workers. To manage the risk of working with computers for example, the Health and Safety (Display Screen Equipment) Regulations require employers to conduct a DSE risk assessment for users of VDUs (visual display units) at regular intervals. How is this largely still done today? Through a survey that every worker is forced to complete once a year, with the results recorded in a risk assessment. While the process is notionally about the worker’s health, its primary purpose is to gather data for organisational compliance. The output is for the manager, not the worker.

Wouldn’t a more useful approach start with the user, encouraging them to review their DSE use whenever it was needed, such as when they change from one setting (maybe using two screens in an office) to another (working on a tablet or mobile phone at home? Or when they start to use a new piece of kit? Then maybe giving tailored preventative advice to the user that might mitigate risks before they become serious, rather than report on them after they’ve happened?

This requires a change in mindset to enable technology to adopt the sort of user-centric approach familiar to us from other apps we interact with every day. The trick is how to generate a legally required DSE risk assessment for organisational compliance purposes (an output) from a set of loosely structured interactions (inputs) entered by a user over time, to create a genuinely useful tool to deliver beneficial outcomes (improved wellbeing and better performance). This is exactly the trick that AI can help pull off.

You might notice I haven’t distinguished between a) knowledge required to develop competence, and b) knowledge required to perform tasks competently.

Workplace learning has largely focused on the former: the development of skills and knowledge, through a number of methods including training, coaching, mentoring, and on the job supervision.

However, the ability to perform tasks competently may not necessarily require any learning at all. In his book The Checklist Manifesto , author Atul Gawande presents some insightful examples of how the humble checklist can be used to great effect: surgeons performing complex operations, pilots carrying out pre-flight checks. Scheduling another training session on the principles of risk management won’t necessarily help them manage the risk; giving them a checklist at the appropriate time very well may.

The same is true of the oft-quoted statistic that 70% of the knowledge we need is now gleaned from a Google search. We don’t necessarily have to retain that knowledge, if we can look it up again next time we need it. Solutions like this – which can aid competence at the point of need – are known as performance support tools.

Knowledge management should be about much more than learning alone (even the Serious eLearning Manifesto  guards against this in the first of its 22 principles) – so why is it still the starting point (and often endpoint) for L&D in so many organisations?

Should L&D not broaden its focus to include all the activities required to develop, maintain and support competence, in order to achieve the performance that comes from that competence? Those activities include performance support, as well as structured learning.

The zombie LMS that Donald Clark refers to is oblivious to performance support. It only yields up content that users have been enrolled on in advance – and that content is likely to be generic rather than task specific. While it might have been written by a subject matter expert (SME) it’s unlikely to have been produced by someone familiar with the risks that relate to a specific task. The result is a knowledge management system that is neither user-centric nor performance-driven. It doesn’t readily allow people to share knowledge they’ve picked up from the work they do; and it risks creating a siloed approach to performance where the content in the LMS (the preserve of the L&D manager) is separated from the guidance found in checklists, method statements and risk assessments (the preserve of the Facilities Manager).

AI will slay the zombie LMS. It has the capability to transform individual and organisational performance by delivering up the most appropriate content, in the most appropriate form, at the most appropriate time.

This is not yet mainstream . But it is happening, and the learning landscape is changing rapidly. New applications such as Workplace DNA® from my own company and the recently announced Viva employee experience and engagement platform from Microsoft  will surely see an end to the LMS as we know it.

A focus on improving competence and task performance will have an impact on the way learning and development is managed at an organisational level. The use of AI applications will drive this change.

Instead of spending time developing and updating content, enrolling learners and monitoring progress, AI tools will empower L&D managers by easing the administrative burden, freeing them up to act on the insights gained from knowledge management activities to further improve performance. In the workplace management sector, it’s likely they’ll need to get closer to operational managers in FM, health and safety and sustainability, to better understand how the knowledge required to build competence relates to tools needed to manage task performance.

AI is likely to see workplace learning become a more strategic function, measuring the impact of the outcomes of knowledge management programmes on organisational performance, using real time data to make continuous adjustments over time.

What does this look like? AI is already being used to create learning resources that would historically have taken months to produce in a matter of days. Curation and updating of content can be readily automated, with online news articles already being written by AI instead of humans . Imagine the law on DSE risk assessment changes next month, and your course content updates itself automatically. Then it automatically sets a deadline for your team to refresh their knowledge, alerting you to the 7% of workers who’ve failed to do so by the target date. Your time can be spent training the AI, approving its recommendations (if you want to), and focusing on the exceptions that need your attention. You don’t need to spend your time performing many of the day-to-day tasks currently associated with the L&D function.

Linking learner data to operational data will afford further insight for organisations. For example, AI could be used to predict the likelihood of an accident at a specific site, based on analysis of historical accident and incident data correlated with learner engagement data. Insight such as this could help a multi-site retailer or logistics provider to focus training resources on the specific locations, tasks and risks identified, which could deliver significant health benefits and performance improvements. Why pay to train the whole workforce in everything, when you can just target the people or places that need it?

And when it comes to performance support tools, the possibilities of AI will be almost endless as they start to transform the simple checklist into augmented reality (AR) tools, for example by overlaying content and data onto a live video feed to help workers complete tasks competently and learn while they’re doing it. 

One of the technological changes that has brought these possibilities to workplace learning is the development of a framework known as xAPI. As a modern, data-driven framework it provides the structure and granularity to displace the outdated SCORM model that has been prevalent for so long. The ‘x’ in xAPI stands for ‘experience’: it means it can be used to track every aspect of an individual’s experience, online or offline, recording every interaction with time-stamped statements to provide a comprehensive history of engagement. Exactly the type and quantity of structured digital data required to train an AI application – data that the SCORM model could never provide.

With detailed learning engagement data, it becomes possible to apply machine learning applications to look for patterns and recommend content to learners based on factors such as their role, location, what others in their team are learning and so on. Combined with location tracking, an AI-driven app could deliver a site briefing to the smartphone of a maintenance operative to inform them of the location of asbestos in a building they’re about to work in. And then store a time-stamped entry to record that the operative had read the briefing (potentially with some checklist questions to confirm their understanding if required).

Or if you’re studying a course like IOSH Managing Safely, an API powered platform will have a good idea from your first interactions how likely you are to complete the course successfully and can tailor any additional content or support you might need to help you pass first time. What that means in effect is that no two people will necessarily ever take the same course, even if it has the same course title (but they may never know it).

The benefits for the learner are not just about saving time and working more efficiently. Serving up the right content at the right time takes away much of the cognitive load that would otherwise divert attention from acquiring the knowledge you need – wasting valuable ‘brain calories’ as they are referred to by Dr Itiel Dror, a cognitive neuroscientist specialising in the application of learning technologies.

Those brain calories are needed not just for the learner to acquire knowledge, but also to retain it . AI applications are very well suited to the use of spaced practice techniques, where knowledge can be broken into microlearning resources with the learner guided through a pathway or campaign, acquiring knowledge in bite-sized chunks that can be tested and reinforced through subtle system intervention. Just as it does with Netflix or Spotify, the AI is working in the background to steer, nurture, check, correct … all of it invisible to the learner.  

While the patterns behind those interactions may be invisible to the learner, the learner engagement data itself is highly valuable to the organisation. Can learning activities be improved? Is learning effective? What is the point of someone retaining knowledge if it has little or no impact on the organisation?   By linking user engagement data with knowledge management activities, individual and organisational performance, AI applications can overcome these limitations to slay the zombie LMS and truly transform learning.

So, what’s stopping us?

Beyond more general concerns about accountability and governance (which are being addressed), there are some barriers that apply specifically to the use of AI in workplace learning.

Data collection

AI requires a large volume of data to be effective: the amount of data will depend on the nature of the application. (How long is a piece of string made of AI? ) But my colleagues at International Workplace quote a million datapoints as a benchmark to aim for so that a system can make meaningful recommendations. Some sectors or tasks generate ‘big data’ and really lend themselves to machine learning. In the built environment, sensors are used to provide real-time data on everything from desk or room occupancy through to the performance building management systems. Combined with external data such as weather information, a huge set of structured data can be crunched and converted into small data for the AI to provide actionable insights at the human level.

Learner records are usually at the opposite end of this spectrum – the very fact that AI is such an opportunity for workplace learning is because the data isn’t usually available in any organised digital form. Historical learner records might be stored in a number of places: personnel files, training files, in one or more LMSs; and in various formats: Excel spreadsheets, scanned pdfs, sheets of paper in filing cabinets; or they might have been lost altogether! Our company specialises in learner data and our mantra is ‘one learner, one record’; but even for us this goal is indescribably difficult to achieve (a subject for another blog!). This will change as xAPI is more widely adopted, but as a framework it is still relatively new, and existing SCORM-based LMSs contain precious little in the way of granular engagement data. 

Privacy and information security

Learning records are about people, and that means personal data. Rightly so, it’s a highly regulated area, and one that still lacks rigour as applied to workplace learning. Adopting an algorithmic approach to the processing of personal data is captured by GDPR and is very much on the radar of the EU, whose High-Level Expert Group on AI produced a set of Ethics Guidelines for Trustworthy Artificial Intelligence in 2019. At an operational level, there are risks around permissions that make it difficult for employers to share data from one system to another. Special category or ‘sensitive’ data (such as information about an employee’s health) should not find its way into a machine learning algorithm, yet all too often that is exactly the sort of thing that can happen. There are practical risks. And there’s a growing amount of regulation. All good reasons for employers to be cautious about AI in learning.    

Fear

There is a natural reticence around embracing AI for learning. People don’t fully understand its application, which makes it difficult to trust. In every sector where AI solutions are engaged it is frequently viewed as a threat to job security. Who needs a driver for a driverless car? It can be disquieting. Andy Lancaster describes one of the benefits of AI in his book Driving Performance Through Learning as a benevolent but invisible force: “As is now the case with online shopping and film selection, learners will hardly be aware of the intelligent systems providing a responsive and relevant learning solution.”  But you don’t have to think too hard about the state of the high street or the fate of the Blockbuster video store to see that these advances in technology have had a huge impact on jobs. And the fear with AI is that surely L&D could be next?

Legacy inertia

Larger organisations are likely to have made a considerable investment in enterprise resource planning (ERP) software such as SAP, Oracle, Microsoft or – for learner records – a corporate LMS. This represents a considerable sunk cost in a legacy system. Workplace learning professionals may be aligned with the chosen solution – they may have procured it, or their role might rely on using it. Intentionally or unintentionally, these are good reasons to delay additional investment in solutions that might disrupt the status quo.

Legacy inertia could also be a factor in the supply chain, where the use of AI might be seen as a threat to established business models. Many awarding bodies - who certify knowledge attainment - continue to measure the effectiveness of learning in inputs which then require training providers to measure and report on compliance. There is an important debate here. If a learner is competent enough to successfully complete an accredited course, and the AI can report in detail on learner engagement and outcomes, does it matter how many hours it took them?

Currently, too few organisations are linking learning to performance. Nearly a quarter of L&D professionals fail to measure learner engagement according to LinkedIn Learning , and less than half measure anything other than learner satisfaction ratings according to CIPD . I hope I’ve shown why, as Donald Clark says, zombie LMSs are the enemy of innovation. They’re compliance systems run to measure administrative inputs, not to support the development of teams and individuals or to drive organisational performance.

AI will transform workplace learning as the value of learner engagement data becomes fully appreciated. It will foster the democratisation of learning in organisations, supporting each unique worker individually, and to recognise knowledge gained over their lifetime and not just from course certificates or training undertaken with their current employer. It will give individuals access to their own learning DNA journey, a history and a record they will take with them wherever they go.

For employers, it promises increased efficiency, targeting workplace learning programmes and performance support tools that make a difference to organisational performance.

For those who are ready and willing to embrace these opportunities, it’s an exciting time to be working in learning and development.  

David Sharp FCIM FIWFM is CEO of International Workplace, developer of the Workplace DNA® digital learning application.