Research Article: Rapid instructed task learning (but not automatic effects of instructions) is influenced by working memory load

Date Published: June 6, 2019

Publisher: Public Library of Science

Author(s): Maayan Pereg, Nachshon Meiran, Claude Alain.


The ability to efficiently perform actions immediately following instructions and without prior practice has previously been termed Rapid Instructed Task Learning (RITL). In addition, it was found that instructions are so powerful that they can produce automatic effects, reflected in activation of the instructions in an inappropriate task context. RITL is hypothesized to rely on limited working memory (WM) resources for holding not-yet implemented task rules. Similarly, automatic effects of instructions presumably reflect the operation of task rules kept in WM. Therefore, both were predicted to be influenced by WM load. However, while the involvement of WM in RITL is implicated from prior studies, evidence regarding WM involvement in instructions-based automaticity is mixed. In the current study, we manipulated WM load by increasing the number of novel task rules to be held in WM towards performance in the NEXT paradigm. In this task, participants performed a series of novel tasks presented in mini-blocks, each comprising a) instructions of novel task rules; b) a NEXT phase measuring the automatic activation of these instructed rules, in which participants advance the screen using a key-press; and c) a GO phase in which the new rules are first implemented and RITL is measured. In three experiments, we show a dissociation: While RITL (rule implementation) was impaired by increased WM load, the automatic effects of instructions were not robustly influenced by WM load. Theoretical implications are discussed.

Partial Text

In the past few years, there is growing interest in the human ability to perform actions immediately following instructions and without prior practice [1–3]. This ability is mostly evident in relatively simple tasks that combine a small number of familiar elements in a novel association and is termed Rapid Instructed Task Learning (RITL; [4]). RITL can be exemplified in many real-life situations, such as operating a machine or software for the first time, or assembling a new piece of furniture. Moreover, instructions are sometimes so powerful that they even produce automaticity, reflected in activation of the newly instructed rules in an inappropriate context [2,5], henceforth “automatic effects of instructions”. Since the instructed task-rules have never been executed before, their representations (at least in their first implementation) cannot rely on long-term memory (LTM) traces from past performance [6,7]. Thus, instructions-based performance was hypothesized to rely on active maintenance in WM [1,8,9], a neuro-cognitive system responsible for holding novel representations [10].

Following Shahar et al. [16], we loaded WM by manipulating novel-rules load. To do so, we increased the number of novel arbitrary task-rules from two task-rules (as in the original NEXT paradigm [5]), to four task-rules. We directly compared low and high WM-load conditions (i.e., 2 vs. 4 novel task-rules) in a within-subjects design.

Experiment 2 involved two groups of participants. In each group, we employed a within-subjects manipulation of Number of Task-Rules, and the groups differed in whether an increase in Number of Task-Rules was associated with an increase in WM load. In one group, (novel-rules-load group) we replicated Experiment 1. This group accordingly involved two conditions–low WM-load with 2 novel task-rules (lWM2R), and high WM load with 4 novel task-rules (hWM4R). In the second group of participants (familiar-rules-load group), the 2 novel task-rules condition was the same as in the novel-rules-load group, but the 4-task-rules condition was different. It involved 2 novel task-rules + 2 familiar task-rules that are thus not assumed to rely on WM (lWM4R, meaning low-WM with 4 task-rules). Our prediction was that, if the critical variable is the number of Task-Rules and not WM load, the effects seen in Experiment 1 would replicate in both groups. If, however, the critical variable is WM load, then the effects of Experiment 1 would replicate only in the novel-rules-load group and would be absent or substantially reduced in the familiar-rules-load group.

The 2 novel task-rules conditions in both groups were similar to that used in the previous experiments, except that the left and right response keys were placed closer to each other. The 6 task-rules conditions included 4 additional rules involving responses that had to be simultaneously performed with both hands, such that they could not be represented as right/left (see Method for an elaboration). In the novel-rules-load group, these 4 additional rules were all novel and arbitrary, as in the previous experiments. In the familiar-rules-load group, these additional rules comprised of S-R mappings that were non-arbitrary. Specifically, the stimuli were designed in such a way that they would clearly remind what the response should be.

The results from three experiments show that increasing the number of novel arbitrary rules held in working memory influences instructions-based performance (RITL). Importantly, increasing the number of familiar rules that hypothetically should not (or minimally) rely on WM resources, did not influence RITL thus pointing to the unique contribution of WM. In contrast, increasing WM load did not differentially influence the automaticity of instructions (manifested in the NEXT compatibility effect), suggesting a dissociation between RITL performance and its associated automaticity effect. The results regarding RITL and automaticity are further discussed separately, followed by broader theoretical implications.