Education is complex – really complex. We humans struggle with both the scale and the granularity of educational options, which result from huge variation in learning content, learning targets, individual skills and learning styles, as well as other factors in the learning environment.
Experienced human teachers in physical classrooms, or those teaching small groups over the Internet, can address this complexity well in the subject areas where they have experience. But this solution does not scale well to online education, which is a growing imperative. There is an explosion of excellent (and not so good) online education content, but getting the right content to the right people, in the right way and at the right time, is an extremely challenging problem.
We are building LearnerShape based on the idea that AI can help address this problem. But AI is a tool. The solution needs to work for humans, requiring an approach that both:
Our approach to reconciling these two critical requirements is to build our technology around core educational concepts that are both broad and flexible:
All of these are common-sense concepts that are easily understandable by humans and reflect real learning experiences. We explain more what each means below.
Equally important, these concepts are broad and flexible enough to capture the full complexity of online education. Machine learning algorithms work extremely well with broad frameworks that allow embedded complexity, and are able to spot patterns in that complexity in a way that humans cannot easily do.
An excellent illustration of this “superpower” of AI is the tremendous progress of deep learning algorithms in image recognition, catalyzed by the 2012 ImageNet competition and now offering much better than human-level performance on various tasks. The approach of these algorithms is to treat every image as a set of pixels (without any rules or categorization of different types of images), building high-dimensional classification models that learn from human labels or other information.
LearnerShape is taking advantage of a similar revolution that is occurring in AI for natural language processing (NLP) – among various other AI and data science techniques that we use.
This brings us back to the four terms we mentioned above, and how they can capture the complexity of the educational environment in a way that both humans can use and AI can parse effectively.
Skills are at the center of our model. A skill can be any ability that a human may have, such as knowledge (e.g. Mary is an expert on China), competence (e.g. John speaks excellent Chinese), soft skills (e.g. Ellen is a great leader), learning styles (e.g. Peter is a fast visual learner) and others. Many existing skills frameworks (such as O*NET or ESCO) draw granular distinctions among different types of skills. We believe it is much more effective to take a ‘model-free’ approach that avoids arbitrary line drawing and supports any skills framework. We do this by analyzing skills using advanced NLP and other AI techniques.
A learning Resource is any unit of learning content, such as an online or offline course; video or audio content; a paper, article or blog; a practical / on-the-job learning program; virtual or augmented reality content; and others. LearnerShape uses AI and data science to recommend learning resources that teach specific skills, to assess quality and level of those resources, and for other related tasks.
An Objective is any set of skills that form a learning target. This may be a set of requirements for a job, or any other set of skills that an individual is seeking to learn. Objectives are central to our workforce planning offering, which allows individuals to forecast which current employees are best suited to be reskilled for specific expected future job requirements.
A learning Pathway is a sequenced set of resources aimed at achieving an objective. Because there are many paths to any learning goal – inherently involving preferences, judgements, differing learning styles and multiple learning resource options – charting a learning pathway is a very difficult task. LearnerShape is making strong progress towards automating this task.
Broadly speaking, in the LearnerShape universe, our users have a set of starting skills and define the set of skills they wish to acquire (their objective). Our technology then creates highly flexible learning pathways (made up of adjustable resources) to meet those objectives.
Together, these concepts – and the technology LearnerShape uses to apply them – form the backbone of our powerful AI-based reskilling offering. We are on an exciting journey at the leading edge of AI progress in solving online education problems, while already offering solutions that cut through the complexity of online education for businesses and learners now.