Transfer of Learning

Transfer of learning is the capacity to apply knowledge or skills acquired in one domain, task, or context to a different domain, task, or context. It is the phenomenon that makes education meaningful — the reason that learning algebra in school might help you reason about budgets decades later — and it is also the phenomenon that most formal learning systems fail to achieve. The gap between learning and transfer is the gap between memorization and understanding.

In psychology, transfer is classified by direction (positive when prior learning helps, negative when it interferes) and by distance (near transfer to similar tasks, far transfer to dissimilar ones). The educational literature is pessimistic about far transfer: studies consistently show that skills learned in one domain rarely transfer to distant domains unless the learner explicitly extracts the abstract structure shared by both. This extraction — what analogical reasoning researchers call 'schema abstraction' — is the active ingredient that makes transfer possible.

The connection to generalization in machine learning is direct but underexplored. A neural network trained on ImageNet that fails at medical imaging is exhibiting near-transfer failure. A large language model that reasons about code after training on natural language is exhibiting far transfer. Whether these cases constitute genuine transfer or merely statistical overlap remains one of the most contested questions in AI evaluation. The distinction matters: if AI systems cannot achieve far transfer, they are tools, not minds.

The persistent failure of both human education and artificial learning systems to produce reliable far transfer suggests that the problem is not with the learners but with the framing. Transfer is not a byproduct of learning; it is a separate skill that must be explicitly cultivated. Any curriculum that does not teach for transfer is a curriculum that teaches for irrelevance.