Quantified Learning

How can machine learning be useful in improving learning and education outcomes?

We are at an amazing cross-roads of human history.

We have perhaps reached (or passed?) peak consumerism — where, with the simple click of a button, we have access to all the cheap stuff we need — and many things that we don’t need. It will likely be downhill from here where people will increasingly crave less “stuff” and more experiences and richness of life.

With the disruptive rise of automation, the skills and capabilities that we have relied on in the past will no longer be sufficient to make meaningful paid work/living wage contributions and millions will be looking for opportunities and re-training.

And last, with the competitive environment that children are facing in the future they, and parents, will continue to look for ways to grow, learn, and be comfortable, contributing members of society, with some safety, security, and predictability of leading a good life with a good job.

One thing is common in all these trends: the need to continue to feed the human spirit and to encourage individualized, personalized learning and personal development.

I envision a world where we can buy or acquire the opportunities for certain knowledge, skills, and experiences in a more tangible, predictable way. The choices about education will be less about how do you acquire — and more about what you acquire — leveraging your personal strengths and interests in a variety of ways — much more broadly then the limited range of current schooling, courses, and careers. Instead of being defined by the stuff we have, we will increasingly value and define ourselves about where we spend our time, and which knowledge and skills we acquire.

=======================

I’m interested in people working in ed-tech specifically on the problems of quantified learning. My home-grown definition of quantified learning is the use of technology to help provide more granular and predictable education and education outcomes. This is different from the quantified self — but draws from similar inspiration about quantifying aspects of our learning and cognitive development. Couple key areas:

• Breaking down the acquisition of a skill or knowledge area into small pieces
• Identifying, delivering knowledge courses for each of those areas
• Assessing and diagnosing learning and mastery
• Providing appropriate feedback — based on the diagnosis, suggestions for practice, improvement of methodology, velocity, etc.

Khan Academy and Duolingo are examples of companies that are working in this area. Some MOOCs such as Udacity do a good job trying to keep students motivated using game mechanics, participant engagement, and high leveraged, but light weight, personal interactions, and accountability. Some companies in China (Squirrel AI) are doing a great job breaking down and assessing tasks more granularly as well.

A few things are missing though:

• A predictable path to mastery. e.g. how much money and time would it take to achieve real, practical mastery of a language? If you put in 1 extra hour a day — how much faster could you get to the next level? This predictability is critical to the productization of knowledge acquisition. If after an initial diagnosis an agent could tell you that it will take $X and Y hours at your current rate to acquire a skill with 90% confidence level — you are much more likely to invest that time and money.
• Assessment and diagnosis. While some ML inspired education can perhaps recommend courses that are relevant and helpful to you — the mechanisms are quite crude — and they really never get to the fundamental underlying diagnosis of WHY such a course could be important or meaningful for you. In learning to acquire a knowledge or skillset — do we understand why a student is struggling? In which area? Is it a conceptual understanding issue? A missing earlier skill? A self-reflection issue e.g. not understanding that they are doing the wrong thing and not sure how to self-correct?
• And most importantly, and the big benefit we get from 1–1 instruction, is motivation and relevancy — how a tutor or a coach can understand WHY an individual is not progressing, how to tap into their intrinsic motivations and, curating or right sizing the relevance and learning content or intervention or practice for them at that time. Most importantly is inspiration — making the skill seem accessible and fun — and relevant.
• A vision to a wider range of skills — language and math are very important and instruction in those topics need improvement for sure — but also music, art, soft skills, creativity, public speaking, leadership, etc.

I recently learned about some interesting work in Stanford in Computational Education — which appears to be quite related — using machine learning to understand human learning. I think the next frontier is equally, if not more important: the ability to use machine learning to dramatically aid and improve human learning.

While many machine learning activities are focused at automating human tasks and making humans less and less relevant — seems more important to explore how we can use machine learning to enhance, accelerate and bring out those capabilities that are intrinsically and uniquely human.