Our experiences from the recent past, even if no longer relevant, can powerfully shape our current choices. Such “serial dependence” has been observed ubiquitously in various cognitive phenomena [1]. Recent work suggests a causal role of the PPC in mediating serial dependence in rodents [2]; similar evidence in humans is correlative at best [3]. Does the PPC causally mediate serial dependence in humans? We analyzed data from participants (n=26) performing a spatially cued change detection task [4]. Participant received either 40-Hz transcranial alternating current stimulation (tACS) over the PPC or “baseline” sham stimulation, in different sessions [4,5] (Fig. 1A-B). We trained a novel recurrent deep-learning model – an embedding-augmented long short-term memory network (LSTM) – to predict trial-wise response times (RT) from the sequential history of task variables (Fig. 1C). Robust serial dependencies in RT emerged at baseline, as quantified by predictive correlations and accuracy (Fig. 2A-B, Sham, p<0.001, signed-rank test). Remarkably, 40-Hz tACS over the left or the right PPC significantly reduced serial dependence effects (Fig. 2A-B, Sham-Stim; p<0.01). Control experiments revealed that serial dependencies: i) returned to baseline in a post-stimulation “washout” session (Fig. 2A-B, Sham-Post; p>0.1), and ii) were not affected by a control frequency (60 Hz) of stimulation over the rPPC (p=0.560, rmANOVA). Explanatory analysis with SHAP (SHapley Additive exPlanations) [6,7] revealed that, although PPC stimulation did not alter current trial variables’ influence on RT (Fig. 2C, Sham-Stim, T-0, p=0.625), it accelerated the decay of past trial variables’ history effects (Fig. 2D, p=0.016). Additionally, PPC stimulation reduced the influence of cue validity history, but not cue side history, on current trial RT (Fig. 2E-F). Our findings suggest a causal role of the human PPC in serial dependence. In particular, the PPC may mediate history-driven selection [8] by which attentional choices from the recent past influence current behavior.