Retroactive interference grows with the similarity between new and old content, and shrinks as the difference grows

What it is

Retroactive interference (RI) runs in the opposite direction from proactive interference: instead of old interfering with new, newly learned content damages memory for what was learned earlier — learning a new way of doing something makes it harder to recall how the old way worked. The key variable that determines how strong retroactive interference gets is the similarity between new and old content under the same cue or context: the more alike they are, the stronger the interference; the more different, the weaker or even negligible it becomes.

Why it happens

If newly learned content shares the same set of cues and context as old content, encoding the new material directly competes with the old material for control of that cue — the more solidly the new content is learned, the stronger its grip on that cue becomes, and the more competition the old content has to overcome to be reactivated through the same cue. This is the same competitive-retrieval mechanism as proactive interference, just mirrored in time: proactive interference is the old material claiming the spot first and suppressing what comes later; retroactive interference is the new material arriving later and suppressing what was there first. When new and old content share no cue or context, they aren't in the same competition pool, and retroactive interference has nothing to act on.

Studying it

The classic paradigm: participants first learn one set of paired material (list A), then learn an interpolated new set that either reassigns different answers to the same cues (high similarity, high competition) or uses an entirely unrelated set of cue-answer pairs (low similarity, low competition), and recall of list A is then measured and compared against a control group that learned no interpolated material at all. The independent variable is the similarity between the interpolated material and the original material; the dependent variable is the degree of recall loss for the original material. This line of research also finds that the size of retroactive interference depends on whether the original material got independently reviewed or used during the interval between the two learning episodes — the mere passage of time doesn't fully explain interference strength; whether the original memory got protected during that interval matters too.

Where it stops holding

Retroactive interference has its strongest evidence base for declarative material with an explicit cue-answer structure; for skills that have become highly proceduralized and no longer depend much on explicit cued retrieval, the interference pattern doesn't fully match what's described here, and the same similarity-based conclusion shouldn't be applied directly. Most of the evidence also comes from paired-associate list tasks completed within tens of minutes in a lab; transplanting the similarity-to-interference relationship measured at that timescale directly onto real product transitions spanning weeks or months needs discounting.

Applying it

• When rolling out a new workflow that closely resembles an old one (same trigger entry point, only the internal behavior changed), expect users who still occasionally need to fall back to the old workflow (using an old device, following outdated documentation) to have noticeably more trouble recalling how it worked — the more alike new and old are, the more this group's old memory is damaged.
• If business needs require the old and new workflows to coexist for an extended period, don't deliberately design the new one to look like the old one in the name of "lowering the learning curve" — that surface-level friendliness is exactly the variable that worsens retroactive interference and damages memory for the old workflow. Do the opposite: give the two clear, easily distinguishable differences in appearance and interaction.
• Validation: in a scenario requiring old and new workflows to coexist, specifically test error rates on the old workflow among users who have already been using the new one for some time, using similarity as the sole manipulated variable — this directly tests whether, and how much, retroactive interference is at work.

Related

• Same group: A6.16.1 Proactive interference accumulates with the number of previously learned similar items, not a single one-off effect · A6.16.2 Switching to a new category clearly different from the old content can partly release the buildup of proactive interference · A6.16.4 Successive version iterations stack proactive interference, leaving memory for the newest version the most fragile
• Nearby: A7.14 Recognizing and correcting an incorrect mental model · A6.09 Procedural memory and automaticity
• Search terms: retroactive interference · similarity-based interference · unlearning