Online Computer Science for 100,000 7th Graders?

In the wake of the success of the online Stanford AI course and some great discussion on how to build on it at the Google Faculty Summit, I find myself asking more seriously: what would it take to get 100,000 seventh graders to take an online introduction to computer science? And not just 100,000 seventh graders, but 100,000 diverse seventh graders - kids who represent the cultures, genders, and socio-economic groups in the United States. The Stanford AI course (and several of the courses offered since) was in part motivated by the rising and increasingly prohibitive costs of a college education. But I think there is also a significant opportunity to use well-designed online education to provide opportunities that simply would not exist for many students.

To take one example that's near to my heart, current opportunities for K-12 students to explore computer science are exceedingly rare. And there are some very real obstacles that prevent widespread computer science education through schools: 1)a lack of computer science in existing education standards, 2) a lack of K-12 teachers with a background in computing, and 3) a lack of curricular time. While important efforts are under way to address these issues, the fact remains that today only a small sliver of our kids have the chance to explore computer science at all prior to college. Could we create an online experience that could change this? What would it take?

Supporting Learner Motivation through Open-Ended Assignments

In discussing the Stanford AI course, Peter Norvig emphasized the importance of motivation. For adults, the motivation to take a course could be to earn a new credential, to learn material relevant to an upcoming project, or simply to feel challenged to learn something new. In my work with kids and programming, motivation has also been a theme. But the motivations at play for the current online courses are less likely to be successful for a younger audience. Different programming systems have explored a variety of ways to tap into the motivation of younger students. In my work, I've found storytelling to provide a motivating context for computer programming. The opportunity to explore programming as a means to the end of storytelling enables middle school aged kids to apply the tools of computer science toward their own lives. We've seen kids use stories to cement their relationships with others by creating animated movies around shared jokes and memories. We've also seen kids use stories to think through more serious issues in their own lives - a parent's divorce, what it means to have a romantic relationship, how to handle a friend who is becoming more distant. Storytelling can be a powerfully motivating context. But, one of the keys to the success of storytelling as a context is providing an open ended enough context to enable kids to write personally relevant stories. And I think this suggests an important challenge in bringing a Stanford AI course experience to a K-12 audience: we will likely need to support a much more open ended problem space, one that allows kids to still create story-programs that are personally meaningful. This creates an interesting challenge in determining how to assess learner projects. While in the general case, assessment of open ended projects is hard, in the case of a programming system, static code analysis may go a long way.

Providing Flexible and Personalized Learning Pathways

Beginning learners in any domain often struggle to identify and set appropriate goals. It is only with experience that we can begin to make (semi-accurate) judgments about the difficulty of a particular task in a given domain. One of the (many) important supports that a formal learning environment can provide is an appropriately graduated path through a new set of skills and ideas. In a classroom setting, it is difficult for teachers to provide truly one on one teaching. At the highest level, online learners could potentially follow different paths through the same material. This might enable a student to explore content of particular interest or to revisit background material where a given learner doesn't have a solid enough foundation to build on. But we can potentially go much deeper than that by using that learner's history to personalize how we present the material. Much of our instructional material has been one size fits all. There are of course notable exceptions in intelligent tutoring systems and educational recommender systems, but these aren't in broad use. As collecting and storing finer grained learning data becomes widespread, we may naturally begin to break that. I'm not an AI person, but from my potentially naive perspective, this opportunity seems to parallel classical AI vs. machine learning in some ways. Much of our ability to customize education to date has been model based. The ability to collect and find patterns in large data sets creates an opportunity to begin to change this. As we collect data about the common mistakes learners make in mastering a new topic, we can create better feedback around those mistakes.

Peer Teaching and Learning for All

One of the tricks that many colleges use is to recruit students from a previous iteration of class to help the students taking the next one, often as teaching assistants. This is valuable for both the students in the new class because learners who have mastered the material recently have a stronger memory of what was difficult and how they got past it. Do you remember what specifically was hard about long division? I don't .But kids who have learned it recently do, and that puts them in a position to bring a greater kind of empathy to their answers. As it turns out, it's also a good thing for the students who serve as teaching assistants. The saying that the best way to learn something is to teach it to someone else has some truth. Anticipating questions and trying to answer them encourages students to broaden and deepen their mastery. Right now, we reserve this opportunity for a subset of the students in a class, and generally the ones who are the most gifted. But online courses can potentially provide the opportunity to learn by teaching to everyone. You learn once when you try to complete a task yourself. And you deepen that knowledge when you help learners behind you to master the same idea.