The shaping of cognitive control based on the adaptive weighting of expectations and experience.

Abstract

Existing approaches in the literature on cognitive control in conflict tasks almost exclusively target the outcome of control (by comparing mean congruency effects) and not the processes that shape control. These approaches are limited in addressing a current theoretical issue-what contribution does learning make to adjustments in cognitive control? In the present study, we evaluated an alternative approach by reanalyzing existing data sets using generalized linear mixed models that enabled us to examine triallevel changes in control within abbreviated lists that varied in theoretically significant ways (e.g., probability of conflict; presence vs. absence of a precue). For the first time, this allowed us to characterize (a) the trial-by-trial signature of experience-based processes that support control as a list unfolds under various conditions and (b) how explicit precues conveying the expected probability of conflict within a list influence control learning. This approach uncovered novel theoretical insights: First, slopes representing control learning varied depending on whether a cue was available or not suggesting that explicit expectations about conflict affected whether and the rate at which control learning occurred; and second, this pattern was modulated by task demands and incentives. Additionally, analyses revealed a cue-induced heightening of control in high conflict likelihood lists that mean level analyses had failed to capture. The present study showed how control is shaped by the adaptive weighting of experience and expectations on a trial-by-trial basis and demonstrated the utility of a novel method for revealing the contributions of learning to control, and modulation of learning via precues.

The many faces of learning-guided cognitive control.

“…As is evident from these studies, control learning takes various forms, and shapes how people overcome conflict (Bejjani & Egner, 2021; Chen et al, 2021; Spinelli & Lupker, 2021; Suh & Bugg, 2021) and switch flexibly (Bejjani et al, 2021) based on the statistics of the environment. The study by Trach et al (2021) suggests that learning about likely demands can occur at several levels (possibly independently), including one that sits above the level of single task sets, specifically learning about sequences of tasks.…”

“…In the study by Suh and Bugg (2021), the authors probed for relative contributions of experiential control and explicit control (based on cues provided before each abbreviated list) to reduced congruency effects. They revealed dissociable yet interdependent effects: both experiential and (smaller) cue-guided control effects were observed, and experiential control was often relatively disengaged when explicit cuing was provided.…”

“…While reinforcement events like reward (Chen et al, 2021) may contribute to learning about control demand, neither feedback nor reward sensitivity was related to the magnitude of control learning (Bejjani & Egner, 2021). In contrast, the prospect of reward may affect the use of explicit cues (Suh & Bugg, 2021). This suggests that control engagement based on explicit cues may be more dependent on reward-sensitive decision-making processes (i.e., choosing to use or not use the cues [i.e., engaging or avoiding effort; Kool et al, 2010]) than learning-guided control, which could account for why the latter may well represent the most common and powerful mode of control engagement (thus supporting the raison d’être of our special issue!).…”

“…Another open question concerns the time-course and persistence of control learning. The results of Suh and Bugg (2021) suggest that at least some forms of control learning are observable within relatively few (10) trials, but generally it remains unclear both how much experience is needed to produce evidence for control learning and how long control learning persists. The study of Bejjani et al (2021) may inspire additional studies that examine control learning following some delay.…”

“…Spinelli and Lupker (2021) developed a new confound-minimized design by varying the proportion of neutral and incongruent trials in a Stroop task to reaffirm and strengthen the assumptions that individuals are learning about list-level conflict frequency and such learning supports (proactive) control. Suh and Bugg (2021) applied a new analytic approach to demonstrate incremental, trial-by-trial control learning in abbreviated lists of the Stroop task, thereby showing how control is dialed up or down as a function of accumulating low (congruent) or high (incongruent) demand trials.…”