AI platforms will supercharge learning
A conversation with Candace Thille (Mar 8, 2021)
By John Mitchell and Maxwell Bigman (Decade Ahead Project)
“The pandemic happened too fast,” says Candace Thille, Director of Learning Science at Amazon, “since the infrastructure wasn’t there to leverage the technology in the way … it should be leveraged.” Although Candace is concerned that the rapid shift to Zoom and similar platforms might leave people thinking that “online sucks,” she is also contagiously optimistic about the value of learning technologies for colleges, universities and working professionals. “To make a good instructional differentiation decision requires factoring features about the learner, the thing being learned, and features about the context in which the thing is being learned,” Candace explains. “Humans can’t manage all of that but the computer can support us to be able to make those decisions better.” That sounds great. How will Candace Thille’s dream come true?
Candace Thille founded the Open Learning Initiative (OLI) at Carnegie-Mellon University in 2002, creating an online platform and digital resources to support students in a number of college classes. As former Princeton University President Bill Bowen and collaborators famously showed in a six-campus randomized control evaluation, a hybrid approach using these courses produced equivalent learning outcomes in terms of pass rates, final exam scores, and performance on a standardized assessment of statistical literacy, with less instructor time and effort.[1] After a few years on the faculty at the Stanford Graduate School of Education, Candace is now Director of Learning Science at Amazon, building the technology infrastructure and work processes to support Amazon in developing its workforce. “As a learning scientist,” she says, “ it’s an amazing research environment — there’s a great diversity of background knowledge, skills they want to learn, diversity of learning context.”
Online courses are appealing because they offer access and convenience: learners can access learning experiences from any place and any time. “While these two factors are important,” Candace says,” the most important is not access or convenience, but that technology can really accelerate learning.” She explains the importance of making good learning decisions by comparing authentic practice and deliberate practice, using tennis as an example. “If you want to be good at tennis, you can learn by playing but expertise won’t accelerate quickly.” That’s authentic practice. “On the other hand, if you have a tennis pro watch you play, stop you, and give feedback about a specific problem, you can practice to solve that one problem.” That’s deliberate practice. “A good teacher will engage in authentic practice, but pull you out for deliberate practice, focusing on sub-skills and knowledge to make authentic practice better.” This is exactly what Candace’s online systems aim to do: “give you an opportunity for authentic engagement and identity by observing engagement, where you need specific deliberate practice and when you need it.” This process involves collecting data from learner interactions and using powerful methods from AI to tailor learning to the learner.
While this could sound like the machine is telling the human what to do, Candace is quick to emphasize just the opposite. “Learning is essentially an authoritative act,” she says. We “always want the position of the learner to be a position of authority and agency and action.” A successful learning experience “gives the learner agency to make decisions while also letting the computer support them to use their time effectively to meet their learning goals.”
Explaining what happened to learning in the enterprise, Candace tells us that, “pre-pandemic, a lot of learning inside a work environment happened in 3 ways:”
- Apprenticeship model: In a work environment, someone can learn directly from others on their team who have more expertise.
- Instructor-led training: Someone can go into a classroom where a teacher with expertise will help them learn.
- E-learning modules: Learning designers can produce e-learning modules on a specific topic so that someone can look for topics they need to learn and work through the modules on that topic.
During the pandemic, in-person apprenticeship and instructor-led training become impractical, increasing the use of learning design and e-learning modules in Amazon and other companies. Although Candace and her team had built the infrastructure for successful e-learning modules, not all of the learning experiences needed at Amazon were on the platform. One successful example she points to is a course on how to interview people. Interviewing over 350 people a day, Amazon clearly needed a widespread program to develop systematic, effective interview practices. Pre-pandemic this was done using a four-hour in-person course, assembling employees in groups of 25 at a time before they serve as interviewers.
Fortunately, Candace’s team had an online alternative ready to go before the pandemic hit. The online course has now trained over 60,000 people, and Candace expects that, “interview training will never be an instructor-led program again.” The learning designers thought about the knowledge and skills needed for a good interview and designed them into the web-based environment. Learners “need to be able to practice, get feedback, practice. So the new course is different, and people like this experience better. It has access and convenience, but it also supports better learning by building the knowledge and skills they need.” Leaving us with some food for thought, Candace reminds us that “learning is complex” and while humans can’t manage all of the relevant information, the technology can support learners and teachers alike to make better decisions — and ultimately more agency as well.
Footnote
[1] William G Bowen, Matthew M Chingos, Kelly A Lack, and Thomas I Nygren. 2014. Interactive learning online at public universities: Evidence from a six-campus randomized trial. Journal of Policy Analysis and Management, Vol. 33, 1 (2014)
