How a Math Algorithm Could Educate the Whole World — for Free | Po-Shen Loh | Big Think

About three years ago, I became the national coach of the United States International Math Olympian Team. I was very happy for a day thinking this is very interesting. But the next day I started to think that maybe I should do something with this. And I decided that I wanted to focus not only on training an elite group of students but trying to do as much as I could to boost the baseline mathematics capability in this entire country.

Unfortunately, I had no money, no connections, and only one person. So the only thing I knew was mathematics, algorithms, and this probability and network theory. So after thinking for some time, I actually came to an idea which was based on using these core mathematical areas that I'd been working with to actually build a solution for education that could be delivered for free on every smartphone.

This is actually the project I'm working on right now called Expii. Our principle is that actually you could turn that smartphone into a virtual tutor, which automates what a person would get if they hired a tutor. It wouldn't be as good as a tutor, but it could get very close. And if you could deliver a free almost tutor on every smartphone in the United States, you might solve equity problems. You might be able to allow everyone, even if they live in a different ZIP Code, to be able to access this tutor, which previously had only been accessible to people who are quite wealthy.

Because today the cost of a tutor is in the $30 an hour, $20 an hour, $50 an hour depending on how you look at it. If you can reduce that to zero dollars an hour, you would actually open up this accessibility to everyone. If we realize that what we're trying to build is this virtual tutor, then you actually, again, can start to conceptualize well knowledge happens to be all of these concepts linked together in this network.

Then the problem becomes if you have this network, how do you mathematically analyze where a person should go next? That can be done by using probability and statistics to find new ways to measure how much each person understands about each concept. Statistics, because the way that one would measure this is by asking them questions.

The experience someone has is they indicate what they want to learn, and then the system starts to pitch questions at them—questions that they would need to know how to answer in order to understand what they claim they want to understand. As the questions come, based on people's responses to the questions, the system adjusts the difficulty of the questions and where the next questions come from, in the same way that a human tutor adjusts their line of questioning based on whether a person is successful or not successful at the previous question.

If the student reaches a point where they are hopelessly confused, meaning they don't know how to do this question at all, then the system suggests that maybe they could read some explanations. As you can see, it turns the lesson flow upside down. It's not that the class comes first, and then the homework, and then the exam; the first thing that comes is the exam, essentially followed by these practice problems, which adapt to you, followed by the class for anything that you don't know.

The idea is that this should cure boredom at the high end and also cure confusion at the struggling end. I actually started this with a brilliant Carnegie Mellon undergraduate student, and then the two of us built this system together. But when you start with no resources, you need to think of ways to actually generate all of this content in a way which doesn't cost an enormous amount of resources.

And we took inspiration from Wikipedia. Our system aggregates all of the questions and explanations that anyone in the world might want to contribute, uses voting like a website called Quora in order to find out which content is strong, and uses statistics, the algorithms, to figure out what questions are easy and difficult.

So actually, in the end, it turns out that it sucks in all of this content, it licenses it all with the Creative Commons license like Wikipedia, and the...