Source
Risko & Gilbert (2016), LIT-1059 (Trends in Cognitive Sciences 20(9)), read in NOTE-887 from the authors’ manuscript; the typeset version’s Trends and Outstanding Questions boxes were not seen. The studies the review summarises were not read, so every result below is the review’s account of them, with values read off its Figure 2.
What was actually shown
Definition. Cognitive offloading is “the use of physical action to alter the information processing requirements of a task in order to reduce cognitive demand”, split into offloading onto the body (here-and-now: tilting the head to read rotated text, pointing, gesture) and into the world (writing down, setting reminders, saving to a device).
Offloading tracks demand. In three paradigms the share of trials on which people offload rises with the load that impairs unaided performance. Head tilting rises with display rotation and with the number of items (Risko et al. 2014). Reminder setting rises from about 33% to about 73% of intentions as load and interruption push unaided accuracy from about 95% to about 79% (Gilbert 2015). Writing items down rises from about 42% at 2 items to about 87% at 10, while no-choice accuracy falls from near 100% to near 0 (Risko and Dunn 2015).
Offloading departs from demand. Three results could have come out the other way and did not.
- Words rotated in a rotated frame and in an upright frame gave similar rotation costs and a similar physiological measure of demand, “Yet spontaneous rates of external normalization are much higher when both the words and frame are rotated” (Dunn and Risko 2016), and people misjudged the former as harder.
- People wrote items down “about 40% of the time when they had to remember only two items, a memory load at which performance was already at ceiling without offloading”, and “erroneously judged that offloading would improve their performance” there.
- “Individuals with lower confidence in their memory abilities were more likely to spontaneously set reminders, even after controlling for any influence of objective ability”. Lower confidence in a perceptual task, with accuracy held constant, also went with more reminders.
The review offers three explanations for superfluous offloading (an undetected benefit, a bias against effort, a mistaken belief) and favours the third without a test that separates them.
Offloading changes what is remembered. People who typed statements expecting them saved recalled them worse than people expecting them erased (Sparrow et al. 2011), and saving can improve memory for where information is “at the expense of remembering what it actually is”. Photographing museum objects impaired memory for them a day later (Henkel 2014). A turn-by-turn navigation group drove a route better, remembered its scenes worse, and drove it worse a second time without the aid (Fenech et al. 2010).
What this does not say
- That people offload when it pays. The review’s point is the reverse: when people offload and when offloading pays come apart. It gives no account of when it pays, no cost of offloading set against the cost of the internal route, and no effect sizes.
- That the metacognitive model is shown. Figure 3 is a box diagram with three arrows; it is not formalised and makes no quantitative prediction. Belief is never manipulated apart from ability in the studies as reported; the confidence results are correlational after statistical control.
- That reliance drifts. The self-reinforcing drift away from internal capability (“the model predicts a kind of self-reinforcing pattern”) is a prediction of the model, illustrated by the GPS example. No study reviewed tracks it over time.
- That offloading is computation reused over time. Every study reviewed offloads for one use within a task or session. The number of times an offloaded result is used is not a variable anywhere, so nothing here prices expected reuse.
- That the effort being minimised is known. Box 2 leaves open whether the choice minimises time (Soft Constraints) or memory load specifically (Minimal Memory).
- That the memory cost generalises. The saved-or-erased and photo results are single studies as reported; the record holds no replication.
Connections
- QUESTION-095 asks whether one cost model covers leaving a result in the world and amortizing a computation. This account adds two constraints on any answer: a model of when offloading pays is not a model of when people do it, and offloading erodes the internal route it replaces, which a fixed reconstruction cost (CLAIM-303) leaves out. It also names a retrieval cost: “you no longer need to remember the content of the file, but you do need to remember where to find it.”
- THEORY-359 supplies a benefit-minus-cost objective under which the decision to offload could be written; this account says the decision as people make it is driven by belief about the cost.
- Automation bias, in the review’s Box 3, is the nearest relative of the staleness cost QUESTION-095 names: reuse of an external aid’s output without checking it. The review does not frame it as staleness.