Attention and perception

When attention is biased to a particular option during information search preceding preferential choice, this option is often more likely to be chosen—even if its value is objectively lower than that of the alternative. Here I demonstrate that although attentional biases—even to lower-valued options—may reduce accuracy (the tendency to choose the highest-valued option), they can increase reward rate (the amount of reward obtained per unit of time invested in the choice). To achieve a higher reward rate it is often preferable to choose a lower-valued option quickly rather than spend time trying to identify the highest-valued option. Attentional biases are typically associated with faster choices, and in terms of reward rate, this reduction in response time can often compensate for the accompanying decrease in accuracy. This relationship between attention, response time, and reward rate is modulated by features of the choice environment and by individual differences in choice boundaries and in the attentional amplification of evidence accumulation. These patterns are predicted theoretically by the attentional drift diffusion model (aDDM). A reanalysis of empirical data from several eye-tracking studies shows that these predicted patterns also hold empirically across various domains of preferential choice (riskless and risky choice, options of monetary rewards and of food items). It may therefore often be beneficial for decision makers to allocate their attention in a biased manner—that is, to deliberately ignore information on some options—in order to reduce the time cost of choice and thereby achieve a higher reward rate.

Research in categorization, memory, and visual cognition typically employs isolated, static stimuli, whereas most events that are experienced in life are extended in time and potentially overlapping. A challenge in studying these more ecologically valid kinds of events is that it is difficult to relate their physical dimensions with their psychological representations. We begin to address this gap by developing a set of novel auditory stimuli with experimentally controlled physical features that can be related to psychological representations. The auditory modality is not only integral to many real-life events (e.g., speech and music), it is also well-suited for combining stimuli both within and across time. Our stimuli were generated by manipulating the frequency bands above a 200hz fundamental of seven electronically generated sounds such that they exhibited different degrees of spectral overlap. In a scaling study, participants listened to pairs of these sounds and rated their subjective similarity. We used non-metric multidimensional scaling to obtain a three-dimensional psychological representation of these stimuli. One dimension appears to correspond to timbral roughness, while other dimensions do not admit simple verbal labels. There were individual differences in the degree to which participants attended to these dimensions, potentially related to degree of musical expertise. Implications for using these stimuli in memory and categorization paradigms are discussed, particularly in relation to how they may be combined either sequentially or simultaneously to create artificial “events” that mimic the complexity of more naturalistic events.

Personalization algorithms are broadly used on the internet to generate recommendations fine-tuned for individual users. However, these algorithms have also been discussed as the cause of the limited content diversity, which possibly leads people to confirmation bias and polarization (e.g., Pariser, 2011). The mechanism of how such algorithms affect cognitive processes and internal representations has not been understood well. In this study, we investigate how personalization techniques can hinder optimal category learning via an online behavioral experiment and a model-based analysis. In the experiment, participants first studied categories of aliens under different levels of algorithmic personalization, in addition to randomized (i.e., control) and self-directed learning conditions. After the learning phase, participants’ knowledge was tested using an independent stimulus set. The result shows that participants in the algorithmic personalization conditions develop selective sampling profiles and more distorted representations of categories. Also, participants in the personalization conditions tend to show inflated confidence, especially when they make incorrect categorization decisions. In particular, the frequency with which each category is presented during the learning phase is a key variable in explaining overconfidence. To pursue a mechanistic explanation of the personalization effect, we also fit the Adaptive Attention Representation Model (AARM; Galdo, Weichart, Sloutsky, & Turner, 2022) to the collected data. The model-based analysis suggests that it is important to comprehend how human evaluates the similarity between exemplars when stimulus information is only partially encoded. If learners assume that unobserved information would be similar to what they have already encoded, this tendency will likely result in higher confidence.

This study investigates how people processs the value and probability information and to what extent their processing strategies are affected by these factors. We employed Systems Factorial Technology and conducted the redundant-target task to infer the participants’s decision strategy. Participants were required to detect the presence of any target presented at the top or bottom of the screen and feedback (gain/loss point) was provided after their responses. Target location probability and value were simultaneously varied across two conditions while the expected values of the top and bottom locations remained the same. Specifically, in the first condition, participants were instructed that in the single-target trials one location would have lower probability than another with a ratio of 2:3 and the corresponding payoff was ±105 or ±70. In the second condition, we adjusted the relative differences by increasing the ratio (1:4) and its corresponding payoff (±280/ ±70). Our results showed individual differences in strategy adoption. That is, three participants maintained a parallel self-terminating strategy across conditions. The other five participants alternated between the serial and parallel processing across conditions. These individual differences were less likely associated with the relative probability and reward manipulations since most participants were consistently more sensitive to the stimulus saliency at the high probable locations. Although our results are inconclusive to provide a mechanistic explanation about the individual differences, the shift between parallel and serial processing implies the potential interaction between value and probability processing on decision-making strategies.

Previous research reported conflicting evidence regarding whether Chinese characters are holistically processed. In past work, we applied Systems Factorial Technology to examine the processing efficiency for Chinese characters and English words. Our results indicated that native Chinese speakers exhibited limited capacity processing both characters and words. To identify the source of that limitation, our current research further investigated the mental architecture of processing Chinese characters and English words. Specifically, we hypothesized that observers’ performance would be indicative of a coactive processing architecture, where all information is pooled together to reach a single decision process. This architecture is often considered a benchmark of holistic perception. In Experiment 1, participants were asked to make a same/different judgment on the sequentially presented characters/words which either both or neither of the left and right components differed. The results indicated that participants adopted a parallel self-terminating strategy (i.e., same or both-different structure). Experiment 2 complemented the findings of experiment by examining performance with added conditions that either the left or right component could now be different (i.e., same, left-different, right-different, both-different). With the decisional uncertainty, the results indicated that most participants processed the stimuli with a parallel exhaustive architecture and a few participants exhibited the coactive architecture. To conclude, our current work provided evidence for weak holistic processing (parallel processing) for Chinese characters and English words, with the stopping-rule (self-terminating/exhaustive) dependent on the task and presentation context.

Researchers in facial perception foster competition between holistic and analytic encoding. Despite the popular belief that faces are perceived in a holistic fashion, both neural organization of the visual system and the phenomenological experience indicate that faces can also be examined analytically in terms of facial parts. This view is further corroborated by the concept of hierarchical object representation, in which selective neural populations are fine-tuned to detect visual properties ranging from simple features to more complex combinations of features. Thus, theoretical developers face two major challenges. The first one is to determine how to integrate both holistic and analytic encoding within the same framework, relying on the idea of hierarchical facial representations. The second is to further integrate these facial perception stages with the higher-level cognitive processes, such as memory and decisional processes. To answer these challenges, we proposed a computational framework of Modular Serial Parallel Network (MSPN), which is a synthesis of several successful approaches in both perceptual and cognitive domains that includes signal detection theory, rule-based decision making, mental architectures (serial and parallel processing), random walk and process interactivity. MSPN provides a computational modeling account of four stages in face perception: (a) representational (b) decisional, (c) logical-rule implementation, and (d) modular stochastic accrual of information, and can account for both choice probabilities and response time measure predictions. In a facial classification task, the MSPN model showed an impressive ability in fitting choice response time distributions, over other facial perception models. The MSPN can be used as a tool to further the development and refinement of hypotheses in facial perception. The analysis of the model’s parameter values, estimated from data, can be used to explore distinct properties of the perceptual and cognitive processes engaged in both analytic and holistic encoding. The MSPN could be generalized to other domains in both cognition and perception.