To help shape the field and create a network of learning engineering, we have created Learning Engineering Google Group. Email ulrich@the-learning-agency.com if you want to join the conversation.

Inspired by work in other fields like bioinformatics, academics are leading the push for learning engineering, and many universities like the University of California, Irvine and Carnegie Mellon are either already offering learning engineering programs or working on creating one.

At the same time, a number of foundations like the Chan Zuckerberg Initiative (CZI) and Schmidt Futures are funding new areas of work in learning engineering. Indeed, Bror Saxberg, Vice President of Learning Science at CZI, has long been one of the leading figures in the field of learning engineering.

Even the federal government is getting interested, and the head of the Institute of Education Sciences at the Department of Education, Mark Schneider, posted a blog item recently touching on, among other things, the need for more support for learning engineering.

“As researchers consider applying for the systematic replication grants,” Schneider writes, “we are particularly interested in technologies that can test many more students more quickly and more cheaply. New platforms are emerging that can do this, perhaps leading to changes in our ‘standard’ model of RCTs [randomized controlled trials].”

In a way, learning engineering seeks to solve a very simple problem: We don’t know enough about how people learn  — and technology can help solve this issue by leveraging big data.

So far, ed tech platforms have not invested much in a learning engineering  approach. As a computer scientist Neil Heffernan argues in a recent op-ed, “educational technology platforms don’t share enough of their data.”

“It doesn’t need to be this way,” Heffernan continues. “Educational technology companies and researchers can open up the back end of their platforms so researchers can access data and run experiments.”

This back end is at the heart of learning engineering. A bit of history will help us understand why.

It seems that AI expert Herbert A. Simon coined the term “learning engineering” about 50 years ago.

Simon was one of the first to apply technical competencies to learning, and he also played a central role in the development of computer science more broadly.

What’s new are two things. For one, there are large-scale ed tech platforms that are beginning to generate a huge amount of data.

What’s more, there is the new and powerful tools of computer science — rich computation, nascent AI, natural language processing. These tools are having a big impact on fields like biology, physics, and astronomy, and they promise the same benefits for education research and development.

Take, for example, a new study titled “Language as Thought: Using Natural Language Processing to Model Noncognitive Traits that Predict College Success.” Released this year, the researchers, including grit expert Angela Duckworth, studied how language can help to predict a student’s future by studying noncognitive traits.

The researchers built off of years of studies that suggest that thought and language are inextricably coupled. As part of the work, the team used natural language processing, machine learning, and manual coding to analyze college application essays and infer  writers’ noncognitive traits, such as goal orientation and perseverance.

The study concluded that language can indeed reveal noncognitive traits and help gauge a student’s learning potential. To be more exact, a student’s college essay predicted their success in college.

This information can be helpful in learning how factors other than GPA and academic record can impact a student’s career as well as in counseling students and building careers. 

“Learning engineering” is a relatively new term, to be sure, and there are a number of competing definitions. Indeed, some believe that learning engineering is the same as saying “data-driven instructional design.” Others believe that is deserves particularized attention.

From my perspective, learning engineering takes place at the intersection of computer science and the science of learning. Much as computational biology did not replace biology as a discipline, learning engineering  will not replace the study of the science of learning.

Learning engineers must, then, have a broad knowledge of learning science, data science, and computer science. They must be able to understand different aspects of engineering processes, including everything from pedagogy to artificial intelligence.

The main aim is to use data to improve learning and teaching. Learning engineers use technologies, standards, and science to propose, test, and implement solutions.

One of the most useful things learning engineering can do is to take some of the guesswork out of education. In the past, educators have largely been in the dark about how well different instructional tools and resources are working.

For example, it can be very difficult for teachers to know how much to credit their students’ ratings of different educational resources. Educators may worry that students find a tool useful that is not in fact helping them learn, or, vice versa, they might dislike one that is nonetheless teaching them successfully.

Learning engineering can change this dynamic. For instance, a study from Jacob Whitehill and colleagues suggests that teachers should take students’ responses seriously. Students, it seems, know what’s working for them.

In that study, the researchers compared learner’s subjective ratings for open educational resources with the learning gains students realized from those resources. They found a strong correlation. Subjective ratings were highly predictive of students’ level of improvement on tests taken before and after they used the resources.

Although there has been an explosion in the number of digital learning platforms, not enough attention is paid to learning science in their design and implementation. These platforms should be built so that they can both provide data and experimental environments for learning scientists, and accommodate research-based insights in future updates to the platform.

In other words, too little has been done to take advantage of the data created by learning platforms. By implementing learning engineering principles, these systems can create feedback loops that promote our knowledge of learning while improving the platforms themselves.

Learning engineering can also help fill serious gaps in education research, offering researchers a built-in experimental infrastructure, with large sample sizes, low set-up costs, and customizable conditions. Consider for instance this recent learning engineering  study that looks at the impact of formative feedback and elaborative questioning, with more than 5,000 subjects.

In the last few months alone, there have been a number of other impressive studies showing the power of learning engineering. A reminder system based on information learned via learning analytics boosted student grades by 0.4 standard deviations.

Another research team used learning analytics to provide tailored feedback to students about their progress, significantly increasing student satisfaction with feedback and improving student achievement.

Digital tutoring software has shown a lot of promise as well. Meta-analyses suggest that intelligent tutoring systems can lead to better outcomes than conventional classrooms. Moreover, they can be integrated by teachers into a variety of pedagogical environments and “perform as effective classroom assistants.”

Another study suggested tutoring software might be useful in addressing achievement gaps. The researchers compared math scores from groups of students in two different after-school math programs: one which was teacher-led and the other which used ALEKS, learning software that personalizes instruction through adaptive questioning. They found evidence suggesting the gaps in performance between white and African-American students were lessened in the group using ALEKS.

Although these successes show the promise of learning engineering, there is still a long way to go. A big reason is that what works for one subject area or one student population will likely not be generalizable. Strategies that improve learning in high school algebra class may not work for college-level data science or middle school civics.

To find sustainable solutions to a variety of different educational problems in a variety of different educational contexts, we can’t rely on quick fixes. What’s required is, rather, an approach that acknowledges the need for fine-grained analysis and long-term experimentation. By leveraging data analytics and technology, learning engineering can generate genuine insights.

Although we are not close to the finish line, we are not at square one either. Learning science has already generated a number of instructional design principles that can be applied to different kinds of courses. These principles include, for example, the importance of spacing out learning, integrating different elements in a course, and varying examples.

But it will take time, patience, talent, ingenuity, and, perhaps most importantly, cooperation to implement these and similar insights in new technology. We will need to  find out what works in different contexts, and iterate for continuous piecemeal improvement. It’s time, then, to get learning engineering to work.

Follow Along and Join the Google Group

At the Learning Agency, we have been fostering a Learning Engineering Google Group.

Email ulrich@the-learning-agency.com if you want to join. We’d love to have you!

READ MORE ABOUT
The Science of Learning and Learning Modalities
Deep Processing and Shallow Processing