Online learning discovery
Main Project Content
In 2011, ‘discovery’ would have been the right word for MOOCs (Massive Open Online Courses). That year, Sebastian Thrun filmed himself in his living room in what would become the first big hit MOOC – Introduction to Artificial Intelligence.
Strictly speaking, other open online courses had existed e.g. Lynda (2002) or Udemy (2009), but the combination of topic – Artificial Intelligence – with a Stanford lecturer captured the imagination in a way that previous courses hadn’t, and an unheard-of number of people, 160,000, signed up.
Within a few months, Udacity and Coursera had been born out of Stanford, and edX latterly emerged from Harvard and MIT. MOOCs had changed the perception of online education: by bringing the gold standard education from the top universities in the world online they validated the field at large. Since then MOOCs have been used in homes, schools, on campus and in businesses to upskill workers quickly, and from anywhere with a connection to the internet.
At the same time, a parallel but entirely connected trend was emerging, the concept of the need for ‘lifelong learning’. The trend’s logic was simple: rapid technological change would continually create new tools (such as analytics), new business processes (e.g. cloud) and new industries (AirBnB) and people needed to keep learning to stay abreast of these.
The two trends reinforced each other and by 2017 over 58M people have taken over 7,000 MOOCs. The numbers expand vertiginously if one includes the hundreds of platforms emerging to cater to different specific skills areas (coding, creative arts, corporate training etc). At the confluence of these two trends, the question of ‘What can I learn [online]?’ in 2011 became ‘What should I learn?’.
So, how does the learner know what they need to learn and how do they find it? Broadly, online learning platforms have approached this in three ways:
Social users with a general learning aim or curiosity will find their way to their chosen subject easily enough with search but how to choose? Learning platforms and learning aggregators like Class Central use popularity and ratings to sort courses, much as users do for a product on Amazon – ‘Most popular courses’, ‘Most highly rated courses’. For learning platforms, as with Netflix, these recommendations are easily created using basic user activity clustering algorithms. Social methods of discovery are simple and durable, because it’s inherent in human nature to follow what others do and like.
For the learner with more specific, or granular learning aims, things are getting more interesting. Coursera recently started to use Machine Learning to scrape the content in their courses and pick out skills and themes. If the learner wants to learn ‘Markov Models’, they can now search and find the course that’s contained in. Superficially, it’s just a more accurate search, but in reality it’s the start of a deeper level of specificity for learners. Coursera don’t yet (and probably won’t) allow learners to just learn a single concept outside the context of a course (which is the level at which Coursera’s business model applies), and perhaps they needn’t: learners may wanted to be guided to learn what the educator thinks is important, and be asked to go deeper and learn the breadth of the topic.
But what if the learner knows where they want to end up professionally, but not how to get there from the perspective of skills development or increased conceptual understanding? LinkedIn are having a crack at this problem. If you apply for a job but there are gaps between your CV and the job description – perhaps you don’t have experience in financial modelling – then LinkedIn can offer their courses in LinkedIn Learning to upskill you towards that job. Over time LinkedIn will be able to develop skill pathways that show you exactly what it takes – at least in knowledge terms – to be eligible for each job. Because they have access to a huge range of professional data, they can analyse the skillsets of product managers on the platform to find they all have UX experience, or conversely find out which of their own courses were correlated with successful applications and career progression.
What next? More data, more personal, more assertive
The future of learning discovery is more personalisation, larger data sets on learner activity, better data on employability and the correlation between key skills and knowledge and successful learner outcomes. Course subjects will split into many conceptual clusters, all with the aim of increasing the chance that a recommendation will be relevant for the learner. Learning history, cookie data, profile data, job data will all contribute towards more accurate course recommendations. The platforms that own access to the necessary data sets, and can to some extent control the focus of course production, together with data on how learners discover courses, will help build better pathways to discovery for future learners.
And what will more data mean for LinkedIn style efforts? LinkedIn could combine global data on professionals to inform their recommendations for you. Perhaps you’re looking to switch jobs? LinkedIn might know the skill a professional in Digital Marketing needs right now more than the professional themselves, perhaps even more than the company. What if its spots a trend in silicon valley that’s yet to manifest in London?
More data will be more personal – but what would make it more assertive? One area could be the continued development of adaptive learning systems that work with the recommendation algorithm. Adaptive learning systems (ALS) work by adjusting what you learn according to your performance. If I’m taking calculus and struggling with differentiation it can address my weakness by looping me back to refresh my knowledge and understanding of key concepts that it knows I am weak at. This approach is used by Khan Academy and works very well to help a learner move through a structured body of learning content.
As ALS functionality becomes further built into online courses, algorithms will learn where your strengths are and be able to suggest skills and jobs that play to those strengths – think of it like a diagnostic tool. Someone may start out with an ambition to be a data scientist, but their skill with building data warehouses is offset by their struggle with statistical concepts. Consequently they may be sent more and more courses dealing with data warehousing, and encouraged to consider becoming a Data Engineer instead. This is like the careers officer at school, but one that has access to your real learning record and that of millions of other learners, coupled with the latest view on professional roles from all over the world. If LinkedIn took on this approach, they could inform you about the right job for you, but could also inform the job about you based on your strengths and learner records. LinkedIn are already suggesting courses based on profiles. Their algorithms may be crude but they’ll improve with time and data.
In this scenario the path and the destination were both made mutable by ever more informed algorithms. There are profound consequences for outsourcing the choice of what we learn to algorithms, much as there are for outsourcing what knowledge we find on Google. Accountability and transparency will be critical to building trust. Ensuring that learning isn’t a reductive process that only focuses on super specialisation will also be important. How data is created, shared and used will be extremely critical to the improvement of learning discovery, and it will be critical that the learner is aware and informed in order to be empowered during their learning process.
Chris Fellingham is Strategy Analyst at FutureLearn