Reflections on the past two decades of Mind, Brain, and Education

Abstract

In the early 2000s, Kurt Fischer and colleagues founded the Mind, Brain, and Education (MBE) field (Blake & Gardner, 2007; Fischer & Bidell, 2006), including a flagship journal, society (International Mind, Brain, Education Society [IMBES]), and a master's degree program at the Harvard Graduate School of Education (Harvard). The MBE program was the first-of-its-kind, focused on the intersection of neurobiology, psychology, and educational research and practice (Blake & Gardner, 2007; Fischer, 2009). Between its first cohort in 2004 and its final cohort in 2022, the program graduated 668 students from around the world (see Figure 1). Contemporaneously, scholars developed MBE or related Educational Neuroscience initiatives in several US states, Canada, the United Kingdom, Austria, The Netherlands, China, Israel, across Latin America, and other locations around the world. A central question in education revolves around what instructional approaches work for whom and under which conditions. MBE seeks to address these inquiries by leveraging cross-disciplinary methods and frameworks with rigorous scientific precision. Yet, the progression of scientific knowledge is often limited by the tools at hand. Herein lies one of the most significant contributions of MBE: It acts as a catalyst for methodological innovation, highlighting the limitations of current scientific approaches in tackling some of the most pressing educational challenges. The need to model the complex probabilistic relations that underlie educational response propelled innovation in the application of sophisticated computational techniques to highly dimensional student data. To illustrate, in a recent study, we used machine learning (ML) to improve methods of prediction of instructional responses in first-grade students at risk for reading disabilities (Shangguan et al., 2023; Zhongkai et al., Under Review). Due to challenges such as incomplete data from young participants, our initial ML models struggled to forecast student outcomes accurately. It wasn't until we integrated methods adapted from the fields of natural language processing and computer vision that we saw a significant uptick in accuracy. We trained our models to recognize hidden patterns between known and unknown data points, resulting in an enhanced out-of-sample prediction accuracy of 80%. In another example, innovative ML was applied to video recordings of math lessons to identify what teacher discourse features were important for supporting a positive learning mindset in students (Hunkins, Kelly, & D'Mello, 2022). In a different domain, the need for more ecologically valid neuroimaging to measure how cognitive and neural processes that underlie learning unfold in naturalistic settings, has facilitated the optimization of portable neuroimaging methods across a diverse range of contexts. For example, portable electroencephalography has been used to study student engagement during live classroom instruction (Davidesco, Matuk, Bevilacqua, Poeppel, & Dikker, 2021; Landi et al., 2019), and functional near-infrared spectroscopy has been used to document neural signatures of reading development in children growing in environments with a high risk of illiteracy, rural Côte d'Ivoire (Jasińska & Guei, 2018). The implementation of portable neuroimaging in these settings posed methodological challenges, and the response to those challenges fostered innovation. In-school EEG data collection is noisy, so the team has leveraged high-density EEG tools, which capture thousands of data points per second, combined with ML classifiers, to dissociate signal from noise. This allowed for a more precise characterization of individual differences in the neural substrates of reading as they unfold during instruction (Davidesco et al., 2021; Landi et al., 2019). To address the challenges of setting up a portable neuroimaging laboratory in low-resource contexts, researchers developed comprehensive protocols addressing problems su