Reactivated past decisions repel early sensory processing and attract late decision-making

Abstract

Automatic shaping of perception by past experiences is common in many cognitive functions, reflecting the exploitation of temporal regularities in environments. A striking example is serial dependence, i.e., current perception is biased by previous trials. However, the neural implementation of its operational circle in human brains remains unclear. In two experiments with Electroencephalography (EEG) / Magnetoencephalography (MEG) recordings and delayed-response tasks, we demonstrate a two-stage ’repulsive-then-attractive’ past-present interaction mechanism underlying serial dependence. First, past-trial reports serve as a prior to be reactivated during both encoding and decision-making. Crucially, past reactivation interacts with current information processing in a two-stage manner: repelling and attracting the present during encoding and decision-making, and arising in the sensory cortex and prefrontal cortex, respectively. Finally, while the early stage occurs automatically, the late stage is modulated by task and predicts bias behavior. Our findings might also illustrate general mechanisms of past-present influences in neural operations.

Serial dependence persists longer for imaginary versus real stimuli

“…This response-driven attraction was observed regardless of motor (Figure 2A-F) or visual (Figure 2D-I) correspondence between the stimulus and the response. Thus, one key finding of our studies is that serial biases emerge from a decisional level representation of task relevant information that are not solely attributable to low-level sensory or motor factors (Braun et al, 2018;Feigin et al, 2021;M. Luo et al, 2024;Moon & Kwon, 2022;Pascucci et al, 2019;Urai & Donner, 2022;Zhang & Alais, 2020;Zhang & Luo, 2023).…”

“…In addition to the influence of sensory uncertainty and memory maintenance, the relative contribution of other factors has also been debated. For instance, some work suggests that serial dependence is largely due to changes in perceptual processing (Collins, 2020;Fornaciai & Park, 2018;Liberman et al, 2014;Murai & Whitney, 2021;Suárez-Pinilla et al, 2018) whereas others suggest a decisional or response-level account (Ceylan et al, 2021;Feigin et al, 2021;M. Luo et al, 2024;Moon & Kwon, 2022;Pascucci et al, 2019;Sheehan & Serences, 2022;Zhang & Alais, 2020;Zhang & Luo, 2023) see also (Cicchini et al, 2021).…”

“…Similarly, although attention has been shown to enhance serial dependance (Bae & Luck, 2020;Fritsche & de Lange, 2019;Makovski & Jiang, 2008;Suárez-Pinilla et al, 2018), attractive biases have also been observed for passively viewed or even task irrelevant features (Fornaciai & Park, 2018;Murai & Whitney, 2021). Lastly, the neural basis of serial dependence is still under debate with different studies pointing to sensory (Ranieri et al, 2022), association (Akrami et al, 2018), working memory (Barbosa et al, 2020), motor (Neto & Bartels, 2021), executive (Schwiedrzik et al, 2014), or more generally post perceptual origins (Hajonides et al, 2023;M. Luo et al, 2024;Papadimitriou et al, 2016;Sheehan & Serences, 2022;Zhang & Luo, 2023).…”