Sunday, 8 February 2015

Education as customized paths through knowledge graphs

Lately very frequently I've been involved with, and witness to, discussions about the upsides/downsides of online learning in comparison to traditional classroom learning. I'd like to summarize a few of my main views on this point.

The traditional classroom has a 1:N ratio of teachers to students, where N grows large for many basic-level courses. Tutoring can provide a 1:1 ratio, and has been found (by multiple quantitative studies) to be more successful at getting concepts across to students. Why? Tutoring provides customization to the individual, and thus can build off of the knowledge base of that individual. New information can hook onto any existing understanding the individual already has, and this is what can let the concepts stick. New concepts become more tightly intertwined with what the individual already knows (and perhaps cares about), and are thus more relevant than concepts presented in the most general setting, with no customization.

In a recent talk of Peter Norvig's that I went to (Norvig is originator of MOOCs: massive open online courses), he indicated that even artificial tutoring systems can have the same benefits as human tutors, with statistically-significant benefits over the traditional classroom. This is very promising, because artificial tutoring is a potentially infinite resource (unlike the finite number of good-quality human tutors). In the same talk, Norvig put up a slide of a dense knowledge graph of all the information that can be available to a student on a particular topic in a particular course(s). He drew some squiggly lines through this graph, standing in for unique paths that could be taken through that material. This is the same visual representation of customized learning that I envision, and deeply believe in, for the future of education.

There is no reason why different individuals should take the same paths through learning. Different types of information may be relevant to different people, and a different ordering of material may make more sense to some individuals but not others. Naturally it should be possible to constrain which points an individual should definitely pass through for a particular course/subject matter (to cover the fundamentals), but the paths themselves should be less constrained. This is the diversity that I referred to in my previous post, which is why I believe that online education is the way forward.

We already have almost all the tools to make this a reality: (1) sophisticated machine learning algorithms that can pick up on trends in user data, detect clusters of similarly-behaving individuals, and make predictions about user preferences; (2) thorough user data through logging and cloud storage, integration of physical and virtual presence and social networks, integration of all of a user's applications and data (and the future of the "internet of things"), universal login systems, etc.

Thus, the question is only one of time.

No comments:

Post a Comment