Coherence vs interference problem of teaching
This text is part of: "I would never send my kids to school" by Piotr Wozniak (2017)
Interference and coherence are two opposite aspects of neural network connectivity in learning. These competing processes have a monumental impact on valid education strategies.
Coherence expresses meaningful connectivity. The positive quality of coherence contributes to building meaningful, well-connected, and well-abstracted memories of high stability.
Interference is the process in which memories compete for survival. The welcome aspect of interference is generalization by which a model is perfected by forgetting its sub-components that are inconsistent with the incoming information.
Due to coherence, memories become stronger by improved connectivity. In interference, memories become displaced by new memories or forgotten by chance.
In free learning, the learn drive reward system determines the learning choices that maximize coherence of knowledge. The process of interference is employed in generalization, and in improving consistency. In free learning, coherence is maximized, and interference is a positive force.
In coercive teaching, external forces determine the flow of knowledge. This magnifies the powers of interference. As a result, there is a negligible inflow of stable knowledge, interference becomes destructive, and coherence is undermined.
At the synaptic level, in free learning, coherence favors good connections, while interference eliminates unwelcome connections. In coercive learning, interference is the dominant force. It may destroy welcome connections, and overtake connections that would otherwise belong to welcome memories.
In terms of the jigsaw puzzle metaphor, the coherence-vs-interference interaction may be illustrated as follows:
- in free learning, the student finds the best matching pieces of the jigsaw puzzle and gets rewarded for building a coherent puzzle
- in coercive teaching, the student is forced to assimilate unmatching pieces, which destructively interfere with the arrangement of the puzzle, penalize the student, and make it impossible to build large and coherent knowledge structures
Figure: Memory connections can be enhanced or suppressed in specific contexts. We may wish to generate different associations depending on the situation a select concept lights up. This is part of pattern recognition. The context is part of the pattern. For example, when speaking English, we may wish the concept of the dog bring up the word "dog" in speech, while when speaking Russian, we may want to bring up the word "собака". In the picture, Concept A activates Concepts B or C depending on Contexts 1 or 2. Concepts B and C should mutually suppress their own activations sparked by Concept A (dark blue connection). In the context of coercive learning, unwanted activation of Context C will perpetually suppress the emergence of the association between Concept A and Concept B. For example, if we wrongly learn that Magellan started his trip in Portugal, not in Spain, bringing up the Magellan in the same context, e.g. of a classroom, will perpetuate the wrong connection, even if a teacher provides instant corrective feedback. The remedy comes from new contexts, which provide a chance for establishing new memory routes free of interference
See also:
- Passive schooling
- Why schools fail?
- Fundamental law of learning
- Free learning
- Exponential acceleration in free learning