A person who I find quite insightful (and occasionally inciteful ;) is Donald Clark. He built and sold Epic, an elearning company, and now he leads a learning AI company, Wildfire. He’s knowledgeable (for instance, having read up and summarized centuries of learning theorists), willing to call out bad learning, and he’s funny. And so, when he reported on a new study, I of course looked into it. And I find that it points out the plusses and minuses of learning science research.
To be clear, this is about his product, so there’s a vested interest. However, he’s got integrity; he’s not going to sully his reputation with a bad study. And, it’s a good study. It rightly demonstrates an important point. It’s just that it stops short of what we need for full learning.
So, his product does something pretty amazing. You give it content, and it can not only answer questions about the content (as, for instance, some chat tools do), it can turn the tables and ask you questions about the content. That is, it can serve as a sort of tutor. Which is all to the good.
What it can’t do, of course, is design meaningful practice. As Van Merriënboer’s Four Component Instructional Design (4C/ID) points out, you need to know the information, and you also need practice applying it. And I reckon we’re still far from that. So, while this is part of a whole solution (and Donald knows this), it’s not the full solution. He’s subsequently let me know it can do language tasks, which is impressive. I’m thinking more of contextualized scenarios, however.
The study demonstrates, as you might expect, that breaking up a video into reasonable chunks, and having system-generated questions asked in-between, led to 61% better retrieval, going from getting 8 to 14 questions right. That is a big improvement. it’s also impressive, since it’s generating those questions from video! That is, it parses the video, establishes a transcript, and then uses that to generate a knowledge base. Very cool.
And it’s a well-designed study. It’s got a control group, and a reasonable number of subjects. It uses the same test material, for an AB comparison. Presumably, the video chunking was done by hand, into four pieces. The chunking and break might account for the difference, which wasn’t controlled for, but it’s still a big improvement. Granted, we know that watching a video alone isn’t necessarily going to improve retention (except, perhaps, over some other non-interactive way of dumping content). But still, this is good as it’s an improvement and a lot of work was saved.
What I quibble about, however, is the nature of the retrieval. The types of questions liable to be asked (and it’s not indicated), are knowledge questions. As suggested above, knowledge is a necessary component. But using that knowledge to make decisions in context is typically what our goals are. And to achieve such goals, you basically have to practice making decisions in context. (Interestingly, the topic here was equality and diversity, a topic he has complained about!)
Knowledge about a topic isn’t likely to impact your ability to apply it. What will make a difference are actually doing things about it, like calling it out, having consequences, and actively working to remedy imbalances. And that requires separate practice. Which he’s acknowledged in the past, and rightly points out that his solution means you can devote more resources to that end.
Thus, the plusses of learning science research are we nibble away at the questions we need to answer, and find answers about the questions we ask. The minus, of course, is not necessarily asking the most important questions. It’d be easy to see this and say: “we’ve improved retention, and we’re done”. However, it won’t necessarily lead to reducing the behaviors being learned about, or building ability to deal with it. There are plusses and minuses of learning science research, and we need to know the strengths, and limitations, of it when we hear it.