An international virtual conference with 1500 attendees, EmpoweringLearners.AI, was held on December 5-7, 2022 and featured three keynote talks which elaborated on SEERNet’s work:
Dr. McNamara, an Affiliate Faculty Member; Professor, Learning Sciences Institute and MLF Teachers College at Arizona State University, made the case for large-scale learning sciences and engineering in the conference’s opening keynote. She started by sharing her own work in developing tools such as Writing Pal, a tutoring system that combines game-based strategy instruction with automated writing evaluation. In the course of developing Writing Pal, McNamara became aware of how many variables—both variables relating to individual learners and to their broader learning context—must be addressed when engineering effective learning technologies. McNamara observed that building systems that include multiple variables is very hard for learning scientists and learning engineers. She described the promise she sees for AI to help: “the use of AI in combination with principles of learning engineering offers strong promise to augment our capacity to test multiple theoretical assumptions at scale.” In SEERNet, Dr. McNamara leads ASU’s Learning@Scale team which is opening up a very large and powerful multidimensional dataset at ASU so that third party research can explore the relationships in the data, and possibly test multidimensional models that would help learners to succeed.
One key slide from Dr. McNamara’s talk, reproduced below, illustrated the range of variables that would be important to include.
In his keynote, Dr. Roschelle, Executive Director of Learning Sciences Research and Principal Investigator of SEERNet, amplified and extended these points. Dr. Roschelle also focused on the need for research that includes multiple learner and contextual dimensions, and shared an example from his own Scaling up SimCalc research of how many kinds of teacher variables have an influence on student mathematics outcomes when they use digital visualizations to learn challenging math concepts. He also discussed the implementation and system-level variables that mediate generalization, replication, and sustainability of positive effects across different school system settings.
Dr. Roschelle argued that AI will be especially beneficial when used to develop models that include the multiple kinds of variables that are important. Dr. Roschelle recognized building multidimensional AI models would require richer data sets than are commonly available to educational researchers. He argued that SEERNet, by opening up opportunities to work with data from real large-scale digital platforms, could be part of the solution.
One key slide from the Dr. Roschelle talk illustrated his main idea using the graphed curve typically called “a long tail.” In the slide, shown below, the horizontal dimension considers the kinds of variables in the AI model. At the left, a system may have only one type of variable, such as user log data, and may help a meaningful number of students to experience positive outcomes (shown vertically). Yet many more students may be left out when the only one type of variable is addressed. Moving to the right, AI models would capture many more variables. As variables are added, each new vertical bar shows that more students are now achieving powerful learning outcomes. Adding up bars to the right requires adding more and more variables, but ultimately ends up serving more many more students and better addressing equity, especially compared to systems designed for “the typical learner in a standard classroom setting.”These types of multidimensional models are nearly impossible for humans to build on their own, but would offer tremendous advantages in serving all students equitably. AI will be most valuable, Roschelle argued, when it augments human abilities to adapt teaching and learning to build on students’ strengths and address their individual needs.
Kumar Garg’s keynote at the same conference introduced a policy perspective on the directions suggested by Drs. McNamara and Roschelle. He noted a “scientists’ dilemma:” research is slow and hard to generalize, plus it is difficult to obtain sufficient data and to translate findings into practice. He also noted an “engineer’s dilemma:” that apparatus for hypothesizing and experimenting, measuring, and continuously improving learning systems is often not available to those who are building the systems. He advocated driving improvements in education outcomes by leveraging advances in computer science that accelerate the pace, relevance and use of learning sciences at scale. He spoke of SEERNet as one the key federal efforts to bring the scientists and engineers together around the goal of harnessing digital learning platforms for more rapid and effective advances in digital technologies used at scale in schools.