How AI Allows for Reskilling To Be Flexible

October 16, 2020

LearnerShape is focused on using artificial intelligence (AI) to improve reskilling for future jobs. In this blog post I discuss why AI matters for reskilling -- it’s all about flexibility -- and give a basic explanation of how we use AI.

Successful learning and development (L&D) requires active discovery of learning needs and opportunities -- not just passive consumption of content. A catalog of content can help, but provides little guidance to the learner. A better solution is a curated platform: one that categorises content in terms of subject matter, skills or other criteria. There are many learning providers offering curation today, and they often provide useful guidance to the learner -- but curation depends heavily on up-front choices, and personalisation to organisations and learners is difficult to achieve.

A more flexible approach is vital to creating a productive, engaging learning experience. Our society needs a revolution in learning content discovery. Consider the transition, on the Web, from curated discovery on Yahoo! 20+ years ago (about where we are now for learning content) to the flexibility of Google search today. We need to see a similar transition take place for online learning. (Of course, Google can search for learning content, but it is not well-tailored to the content- and skill-specific needs of learners).

At LearnerShape, we are driven by the conviction that this challenge can be addressed now, for the first time, by using AI -- especially, recent and ongoing advances in natural language processing (NLP). AI allows us to deliver flexible learning pathways for any content, using any skills framework, on any platform.

Our approach begins with the skills framework, about which we have written before. Well-known skills frameworks like O*NET and ESCO are excellent research tools which, by necessity, attempt to address the entire economy. However, the result is a framework which is unwieldy to use within individual organisations, lacking the flexibility to accommodate organisational needs and approaches. The collected data is likely noisy if individual employees apply the framework, and quickly outdated if handled centrally.

Our alternative is a flexible ‘skills graph’ that adapts to the needs of each organisation. The skills are chosen by the organisation and easily understood by everyone. An organisational skills graph can use the organisation’s own skills framework, frameworks developed by LearnerShape, public frameworks like O*NET and ESCO, or any combination. With a fit-for-purpose skills graph, individual employees can assess their own skills with minimal training and the organisation can easily update the graph as needs change.

Building a flexible skills framework has always been possible, but connecting it to jobs and learning resources was a laborious process. Now, by applying the latest advances in AI and NLP, these connections can be quickly generated.

NLP has a history dating back to the 1950s, but the past few years have seen significant advances with the application of deep learning and the development of new techniques. Models such as GPT-3 have seen widespread press attention as record performances are set on benchmarks and models are applied to new tasks. These models are able to pick up on subtle cues in the text of a course description and produce very human-like suggestions.

LearnerShape currently uses the BERT model developed by Google, along with other machine learning techniques, to place learning resources in the context of the skills graph. The process includes elements of human intervention to optimise results, with an overall data pipeline which is many, many times more efficient than manual curation. It is this partnership between human intelligence and artificial intelligence, applying each where they are strongest, that enables any organisation to quickly create a relevant and productive learning environment.

Once a skills graph is defined, the scope of content available is crucial for both organisations and individual learners. Whether we are dealing with existing internal content, prior relationships with educational providers, or the entire LearnerShape content catalog, connecting courses to a skills graph should be a simple process. With the flexibility enabled by AI advanced models, courses and other educational resources can be intelligently recommended without requiring humans to curate and annotate a course catalog.

For the individual, this means recommendations can be a powerful guide through the vast sea of course options without being limiting and heavily prescriptive. Learning and the discovery of courses, articles and videos becomes an iterative and ongoing process built on the skills graph and tailored to the needs of the individual. As I explained above for AI-driven content recommendations using a flexible skills graph, human and artificial intelligence can work in concert to produce a productive overall learning environment and experience.

If you are interested in this flexible approach to learning, you can follow our progress by signing up for our newsletter or experience our platform by joining our beta test program as an individual or organization.

Jonathan Street
Head of Data Science, LearnerShape