Interference: Why New Numbers Push Out Old Ones

How big is the effect, honestly

Big enough to notice, small enough that it is not the whole story, and dependent on the paradigm.

Joshua Hartshorne measured it in visual working memory. In one experiment, capacity estimates dropped about 17% from a low-interference to a high-interference condition. In a large web-based sample of more than three thousand participants, the drop was about 12% from the first trial of a block to the tenth. Item-specific interference decayed entirely within four or five trials. His conclusion is the balanced one: interference certainly plays a role in visual working-memory performance, but it does not dramatically change capacity estimates.

The counterweight comes from Po-Han Lin and Steven Luck, who tested the same question in the canonical change detection procedure — a brief sample array, a short retention interval, a same-or-different judgment — and concluded that proactive interference does not meaningfully distort capacity estimates there, with performance depending virtually exclusively on memory for the current trial.

Both results can be true because the paradigms differ in exactly the way Craig and colleagues predict. Where the task must discriminate the current trial from recent ones along a shared dimension, interference bites. Where it does not, it largely does not.

Interference as a game mechanic

IQ Test & Brain Training Games is our own app, built by Reign Creative, and several of its modes create interference deliberately rather than by accident. It is worth naming them because they make the phenomenon easy to feel.

Running Total Memory presents numbers one at a time and asks you to hold the total. Every new number overwrites the last total, and holding both the list and the running sum is precisely the mistake that makes the mode hard. Take the run 7, 4, 9, 6: the totals are 7, then 11, then 20, then 26. Hold only the current total and the task is four easy additions. Hold the digits as well and you have eight competing items in a store that fits three to five.

Endless Memory rotates between task types across up to two hundred rounds with three lives, so the content changes shape repeatedly. Memory Boss Rush runs eight escalating memory tests back to back with no repeats. Both of these arrange for the previous round to be a live competitor with the current one.

Delayed Answer Challenge inserts a blank pause between solving a problem and typing the answer, which is the classic way of showing that an answer you definitely had can still be gone moments later.

These are game modes. Alongside them the app keeps personal bests, accuracy, streaks and response speed, and it also has a memory test, a reaction time test and an IQ-style test whose results its listing describes as estimates based on app activities, not clinical or professional assessments. What the modes offer is a demonstration you can feel in ninety seconds, which is more convincing than a diagram.

What to do about it

Four things follow directly from the research above.

• Switch the content type, not just the topic. If digits have started to blur, move to colours, letters or arithmetic rather than to more digits. You are changing the dimension the task discriminates on, which is what Craig and colleagues' result says actually releases you.
• Expect the third block to be the worst. Interference builds across trials and then plateaus. A dip partway through a session is the normal shape of this, not a sign that you are tired or losing ability.
• Give competing items different cues. If two similar facts keep swapping, attach each to a distinct image, sentence or location. The elaboration methods in memory techniques that actually work are useful here for exactly this reason: they manufacture the distinctiveness that similar material lacks.
• Space the sessions. Distance in time reduces how many close competitors are active at once, which is part of why spacing outperforms cramming on almost every measure.

Dates are the textbook case of all four, because a long list of four-digit numbers attached to similar events is close to a laboratory interference stimulus. Our guide to remembering historical dates without rote drilling works through the alternative.

The reframe worth keeping

If forgetting is competition, then the question changes from "how do I hold more?" to "how do I make this one easier to pick out?". The first question runs into a hard capacity limit described in working memory and training. The second is a design problem, and design problems have solutions.

More on span, encoding and attention is collected in the memory and attention hub, the arithmetic side of the same app is covered across the education and brain training category, and the modes described above are in IQ Test & Brain Training Games.