Near and far transfer

What the distinction means

Transfer is the use of something learned in a new situation. Near transfer means the new situation closely resembles the training situation: same skill, slightly different numbers, a modest change of surface. Far transfer means the gap is large — the underlying principle is the same but everything around it looks different, or the skill is supposed to migrate to a wholly unrelated domain.

The distinction matters because the popular case for many learning interventions rests on far transfer. Cognitive training companies have claimed that practicing on memory tasks in a laboratory will make someone sharper at reasoning in daily life. Formal logic courses have been pitched on the grounds that arguing about syllogisms transfers to arguing about politics. These are far-transfer claims, and the evidence for them is, to put it charitably, thin.

Where transfer reliably shows up — and where it doesn’t

Near transfer replicates well. If an adult practices identifying drug interactions from a specific set of medical records, performance on different but structurally similar records improves. If someone spends time retrieving vocabulary items in one format, recognition in a slightly altered format benefits. The findings from retrieval practice — one of the better-supported results in this domain — are mostly near-transfer findings: the material tested is close kin to the material learned.

Far transfer is where confidence should drop steeply. The n-back training literature is the most studied case. N-back tasks, where a person must track whether a stimulus matches one presented n steps back, seemed in a much-discussed 2008 paper by Susanne Jaeggi and colleagues to improve fluid intelligence — the domain-general reasoning ability measured by IQ-style tests.

That would have been dramatic far transfer. Subsequent replication attempts at independent labs produced much smaller effects or none, and a 2013 meta-analysis by Monica Melby-Lervåg and Charles Hulme found that working-memory training produced reliable near-transfer effects on similar working-memory tasks but no credible transfer to general reasoning or academic performance. The headline claim did not hold.

That insight implies transfer is never automatic; it requires the learner to recognise that the new problem belongs to the same family as the old one.

Near transfer can show effects in a range that most researchers consider educationally meaningful. Far transfer effects, when they appear at all, are typically small enough that the phrase “statistically significant” is doing a lot of work.

Why the gap is so wide

Several accounts compete. One, rooted in cognitive load theory, holds that working memory is a bottleneck: learners build representations tuned to the specific demands they trained under, and re-encoding for a different context costs the kind of cognitive effort that often just doesn’t happen. Another account, developed through the work of Robert Bjork at the University of California, Los Angeles, frames encoding as context-dependent: retrieval cues are tied to the conditions of original learning, so the further you stray from those conditions, the weaker the cue match.

Frederic Bartlett at the University of Cambridge showed, decades before the term “transfer” was standard, that memory is reconstructive rather than reproductive — recall is shaped by schemas built from prior experience. Recognising that a new problem has the same underlying structure as a familiar one is itself a skill, and training studies rarely teach it directly.

None of this means far transfer never occurs. Expertise built over years in a domain does generalize, albeit unevenly, and there are documented cases of structural insight carrying across problems. But those cases are built on deep, well-organised knowledge — not a few weeks of a targeted task.

The honest summary: near transfer is real and teachable; far transfer is occasionally real and barely teachable. The claims that fill most marketing copy sit in the second category.