One of the things that I’ve been thinking about more often is how to better use the way that brains encode memory to improve learning design.
Learning is simply making our memories accessible and applying them to our lives. The way that courses are designed increasingly ossifies over time, unless you make conscious effort to constantly reexamine teaching practices and activities.
A recent article in Quanta Magazine discussed a new paper on the formation of memories. Classical theories of memory consolidation posit that experiences become memories in the hippocampus and are consolidated over time into long term memory in the neocortex. But these theories, while useful in some cases, do not account for the fact that certain memories remain hippocampus-dependent. The study referenced in the article aims to explore why certain types of memories are consolidated in the neocortex, while others aren’t.
In this paper, the researchers conceptualize the process of building neuronal memory in the neocortex as taking inputs from the hippocampus and developing weighted associations on different, specific neuron associations. The word “bird” associates with wings and flight, for example. So Sun, Advani, Spruston, et al. then built an artificial neuron network and developed a mathematical algorithm to replicate this process; the weighting systems are reportedly fairly common in machine learning, so the novelty of the application seems to be in the segregation of the networks (i.e. one for the hippocampus and one for the neocortex), and using the communication between them to mirror the structures of the brain.
…the standard theory of systems consolidation assumes that generalization follows naturally from hippocampal memorization and replay; it does not consider when systems consolidation is detrimental to generalization.
Sun, W., Advani, M., Spruston, N. et al. Organizing memories for generalization in complementary learning systems. Nat Neurosci 26, 1438–1448 (2023). https://doi.org/10.1038/s41593-023-01382-9
The main takeaway I get from this is that as our brain seeks to make generalizable patterns, it is aided by establishing a solid data set of related memories. From there, our brain can make inferences about new inputs based on the totality of the encoded memories.
Importantly, unique, outlier memories remain outside of the consolidation, so as not to contribute noise.
Patterns in Learning
What fascinates me about this paper is how we might apply this to learning experiences. After all, imparting knowledge is all about giving learners tools to recognize patterns and make inferences in novel situations. We teach about a business skill, say SWOT analysis, using a given industry and situation, and expect learners to generalize the concepts to their own contexts.
But to aid them in this, we should seek to understand what aspects might be unique, and therefore resist being consolidated in the same neuronal pathways. How might we enable learners to better apprehend content? Is there a way to improve consolidation for long term efficiency?
