“…For example, if the friend you are searching for wears only warm colored shirts, but pants of any color, prioritizing the low-variability distribution of shirt colors over the high-variability distribution of paint colors will maximize predictive information in the target template (i.e., minimize target uncertainty). Target features encoded into working memory are known to be remarkably sensitive to the task-relevance of information (Becker et al, 2010; Boettcher et al, 2020; Geng & Witkowski, 2019; Nobre & Stokes, 2019; Rajsic et al, 2020), and it has been shown that object features are flexibly removed or compressed so that only the most informative features are used to guide attention (Bravo & Farid, 2012, 2016; Kerzel & Cong, 2021; Reeder et al, 2017; Woodman & Vogel, 2008; Zivony & Lamy, 2016). Previous work has also shown that when sensory information differs in reliability, perception is optimally weighted toward the most reliable cue (Fetsch et al, 2012; Hillis et al, 2004), resulting in a “push-pull” dynamic for attentional priority (Foley et al, 2017; Kozyrev et al, 2019; Mehrpour et al, 2020; Pinsk et al, 2004).…”