The Future of Education

Educators talk a lot about "differentiation" or "differentiated instruction," which means adjusting instruction to match the needs of individual students. Often it means having students work in groups, presenting extra explanation or materials to some students, or using different homework assignments or even tests. (Just as a note, this is very similar to "tracking," which does this for class-size groups of students by putting them in different classes, but lots of people think tracking is evil, while everybody likes differentiation, as far as I can tell.)

The problem with differentiated instruction is that it's hard for teachers. The reason it's hard for teachers is that it isn't possible, even in theory. You can't do two different things with two different people at the same time - let alone a class that might have thirty or more students.

So, like at Spock's school, some places are starting to differentiate using computers. Two examples:

School of One, a New York City Department of Education program, has a big database of math skills and lessons. The skills are keyed to state standards, and the lessons are pulled from a variety of sources - some are delivered by computer, some are delivered by teachers. All the skills have multiple lessons that can be used to teach them, and there are assessments that help determine when students are ready to move on to the next skills. The lessons that each student does each day are determined based on all data available up to that day - and this scheduling process is, from what I hear, one of the most time-consuming parts of running the program.

Wireless Generation is a private company in DUMBO (Brooklyn) that develops technology/curricula for reading, writing, and math. I don't know as much about them, but they seem to have some interesting stuff. One thing seems similar to what School of One does, but for reading. It updates the plan for each student only once every ten days of instruction.

Both rely on a lot of formative assessments (aka tests) delivered via computer. Both involve students spending some time with computers one-on-one, but also have students working in groups and interacting with teachers (Wireless Generation more than School of One, it seems.)

I think there's a lot of power in really differentiating instruction to the student level in this way. I think School of One has had some success, and they continue their work. There are some issues for consideration:

* There is a lot of content available already from places like Khan Academy, both lessons and exercises (which could also be used as assessments). Heck, there's a lot of content all over the place, both publicly available and owned by companies.

* The principal difficulty, it seems to me, is creating the system that orchestrates the presentation and interaction of materials with the student. It is fiendishly difficult to work out the right progression, and then to decide what to do if a student gets something wrong, for example. This is the kind of thing that a tutor would do, one on one with a student, and even a good human tutor is not so easy to find. So far most computer systems have relied on human expertise, drawing out a curriculum map, that sort of thing. I think eventually the best solution will incorporate human expertise, but also utilise machine learning to determine lesson/problem sequences that adapt to students not every day or every ten days, but moment by moment.

* It's a bit of an aside, but I think this technology should probably incorporate what is known about spaced repetition and memory, to help maximize the amount that students retain from what they learn. The current system of learn-test-forget is rather awful, and really kind of seems to suggest that it doesn't really matter if adults know anything.

* One consequence of this kind of differentiated instruction is that it is no longer nearly as normative. Some kids go a lot faster than others. For School of One, I understand that this has meant that some students finish all the material "too early." I think their response so far has been the correct one: introducing more material. There is no reason that a curriculum of this type shouldn't extend all the way to open questions in the field.

I don't pretend that this kind of thing would work terribly well for creative writing or art or human-to-human communication skills, but I do think this kind of computer-differentiated learning will be very powerful for large pieces of what students need to learn. I hope that people will start working together to develop good and open technologies to implement this kind of thing.