Going Beyond the Engineering Model of Personalized Learning - Interview with Larry Berger (Amplify)

Nafez Dakkak
Going Beyond the Engineering Model of Personalized Learning

With over 20 years in the field, Larry Berger is among the world’s pioneers of education technology. He co-founded Wireless Generation in 2000, which was one of the first companies to make use of data analytics and assessments to support student learning. In 2010, Wireless Generation was bought by News Corporation, which rebranded the company as Amplify. Five years and 1 billion dollars of investment later, News Corporation sold Amplify to private investors.

Today, according to Berger, Amplify’s work is focused on two pillars. First, Amplify continues to build on its work from 2003 around providing observational assessments that allow teachers -- and the system as a whole -- to develop a view on student progress. Second, Amplify has invested resources in the past decade in becoming a “curriculum company” - going beyond just textbooks to interactive simulations and hands-on material. At present, Amplify works on K-8 English language arts curriculum and K-8 science curriculum for the US market.

Berger sat down with QRF for a very informative discussion about the lessons learned during his journey and where he sees the sector headed. The interview has been edited and condensed for clarity.

QRF: You’ve recently made some headlines about a message that you had delivered to a group of experts in the United States earlier this year that was later published online. In the piece, you interestingly admit that you no longer really ascribe to the “engineering” model of personalized learning because the underlying components for that model are still missing. Can you tell us more about that?

Larry Berger (LB): To start off, it’s important to emphasize that I made a careful exception that there are things in early reading and numeracy where an engineering model of education could actually work. It’s really when you try to broaden it into an all-encompassing knowledge map that is entirely technology driven that the model starts breaking down. Once you go beyond teaching the foundations of literacy and numeracy, other things start to matter more like the coherence of the learning experience, or the social dynamics of the students you are learning with.

For example, when we try to identify what adaptive learning looks like or should look like, we should think of what great teachers do and how we can support that. Really good teachers are effectively coaches. They think about the importance of creating a shared experience in the classroom that is still individualized with the right scaffolding for each student - all while paying attention to the feedback they are getting from different students. It is a shared activity with different modes of nuanced feedback - which of course something machines are good at. As such, we should use technology to make this kind of teaching broadly accessible to teachers which results in a better more engaging classroom for teachers and most importantly students.

QRF: It's certainly in our interest to have high-energy engaging classrooms. How can we make our learning technology and education systems more effective to serve that purpose? What data do we need to capture that we aren’t now?

LB: The set of data that I believe is very powerful, but often underrepresented, is “productivity data”; i.e. how much work are students doing? If I had to choose between productivity data and academic data, I would choose the data on effort as it ends up being quite significant. Additionally, its important to capture data about teacher behavior - as that is what we want to change. For example, can we capture how much feedback the teacher is giving students? I can guarantee you that the classrooms where students are getting more feedback from their teacher will be the ones where the students are performing better.

Finally, I would add that it’s important to have a clear theory of how students learn and progress through a certain concept, or set of concepts, before we start gathering data on the progress of their learning. The rationale behind that is that capturing multiple data points within a validated learning sequence allows us to be more confident in our diagnoses of the gaps in learning. For example, if I’m testing you on addition, I need to make sure that I take a holistic approach of the missteps you make to ensure you actually don’t understand addition - rather than simply being distracted or having problems using the technology.

QRF: So effectively, we are better off using data systems for formative assessment rather than more isolated summative assessment. Keeping within the focus on providing quality learning at scale, how effective do you think the massive open online courses or MOOCs movement has been?

LB: It’s important to say that I’m not an expert on MOOCs but my understanding is that the completion rates are a real problem. People are dropping off at a rate that is questioning the entire model - and that seems to be primarily problematic for learners in new subjects that they haven’t studied before.

I would start designing MOOCs from  the assumption that nobody has showed up before to tell the learner this stuff. If we make that assumption, it’s pretty clear that putting a smart person on your screen to tell you stuff, and giving you  opportunities to practice will not be enough. Of course, there are a lot of other factors for why you might not engage with a subject or understand it - in that sense a more hands-on constructivist approach to learning could make MOOCs more successful.

QRF: That’s a very valuable perspective. Moving on, if we can take a high-level view and look at the ecosystem as a whole: are education startups in your view different?

LB: Education businesses take a lot more time to figure themselves out - and that’s primarily because the problems are complex human-centric problems - think of how long drug research cycles are. There is a patience that’s needed which entrepreneurs struggle with because they are by nature impatient people. Similarly, investors get nervous about education startups and struggle to find the patience for the reward/risk timeline. Additionally, it’s important to realize that education products are also unique in that they have different sets of interconnected customers. Finally, it’s important to realize that education needs to be localized or regionalized for the relevant population - so contextualization is almost uniquely important. We need to listen very carefully to local communities to understand what they want.

QRF: Within that incredibly complicated set of considerations, what advice do you have for entrepreneurs on this space on how they can find product-market fit?

LB: My closing piece of advice is to really listen to teachers. There are lot of other stakeholders of course, but if what you are doing doesn’t matter to teachers then it doesn’t matter to the system. You also need to go beyond just listening to what they say - because sometimes they’ve been taught to say certain things - but actually observing what they do. I’ve never met an education entrepreneur who has spent too much time in a classroom. Student behavior matters too of course, but teachers end up being the linchpin.

QRF: That is a great note to end on. Thank you so much Larry for your time today.