Decision Making

Talk Session: Saturday, May 17, 2025, 5:15 – 7:00 pm, Talk Room 1

Schedule of Events | Search Abstracts | Symposia | Talk Sessions | Poster Sessions

Talk 1, 5:15 pm

The theory of subjective inflation in peripheral vision: accept, reject, or revise?

Brian Odegaard¹, Joseph Pruitt¹, Doyeon Lee¹, Angus Macgregor¹, Trevor Caruso¹; ¹University of Florida

Several years ago, the theory of subjective inflation was presented as a possible explanation for why observers may think they perceive the visual periphery better than they actually do (Odegaard, Chang, Lau, & Cheung, 2018). Based on findings that observers have (1) liberal detection biases and (2) overconfidence for decisions in peripheral vision, the theory posited that observers have an inflated sense of perceptual capacities outside the fovea. Here we ask: how well do predictions made by this theory hold up when put to the test in wide-field visual displays? Using a monitor which ranged from -40 to +40 degrees, we presented observers with either Gabor patches or noise and asked them to judge the presence (or absence) of the stimulus, and rate confidence in their responses. Across two experiments, when comparing attended and unattended parts of the visual field, observers exhibited higher numbers of false alarms for unattended (compared to attended) peripheral locations, providing evidence of liberal detection biases in the periphery. Further, two experiments with feedback showed that these detection biases were resistant to different forms of feedback, including not only trial-by-trial feedback, but also explicitly telling observers about the liberal detection bias before they performed the task. However, in a final experiment which matched performance across peripheral locations, confidence was shown to decline with increasing eccentricity, challenging the idea that observers may overestimate their perceptual capacities in the periphery. In light of these new findings, we review the viability of subjective inflation, as some predictions appear to be confirmed (i.e., a liberal detection criterion in the unattended periphery), but others appear to be refuted (e.g., overconfidence in the periphery). We conclude by discussing the importance of using different tasks and subjective measures to further probe perceptual decisions in the visual periphery.

This work was supported by an Office of Naval Research Young Investigator Award (#N00014-22-1-2534)

Talk 2, 5:30 pm

Accumulating evidence over space and time requires primary and secondary visual cortices

Joao Couto¹, Lillian Wilkins¹, Anup Khanal¹, Anne K Churchland¹; ¹University of California Los Angeles, Department of Neurobiology

Accumulation of evidence relies on neural circuits that use time flexibly and strategically. Genetic dissection tools for mice make them an appealing model to understand such circuits, but encouraging mice to accumulate evidence over time and space has proved challenging. Further, not all visual tasks in mice recruit cortical structures (and instead rely on subcortical pathways), making it difficult to draw parallels between mammalian species. Here, we ask if the visual cortex is required for spatio-temporal integration of cues in a novel decision-making task for head-restrained mice. Our task uses localized visual patches (25-30˚), drawn stochastically at different azimuths/elevations, inviting accumulation of evidence from distinct locations in the visual field. After a delay period (0.5-0.7s), a correct choice is reported by licking to the spout corresponding to the side with the highest patch rate. We performed optogenetic suppression of primary visual cortex (V1) or the anteromedial visual area (AM) using an inhibitory opsin (stGtACR2) selectively expressed in excitatory neurons. Unilaterally suppressing V1 during the stimulus period (1 second) strongly biased responses to the ipsilateral side, consistent with a role for V1 in spatio-temporal integration (2 mice/13 sessions). When the manipulation window was restricted to the initial or late stimulus periods, we observed a similar impairment (2 mice/8 sessions), albeit smaller in magnitude. This result, together with psychophysical regression, suggests that mice utilize the majority of the stimulus period. To test whether V1 activity was required after the stimulus period, we suppressed activity in the delay period, and no performance impairment was seen. Similar to V1, suppressing AM during the stimulus period evoked a strong bias towards ipsilateral choices. These results demonstrate that mice can accumulate visual evidence and that this computation relies on visual cortical structures.

This work was supported by NIH (award R01EY022979).

Talk 3, 5:45 pm

Overconfidence for filled-in information in the foveal rod scotoma in mesopic vision

Hui Men¹ (), Alexander C. Schütz¹; ¹University of Marburg

Different types of retinal photoreceptors contribute to vision, depending on the level of illumination: cones function in photopic vision in bright light, while rods function in scotopic vision in dim light. In intermediate mesopic vision, rods and cones are active simultaneously. Due to the lack of rods in the fovea, a foveal scotoma occurs under scotopic vision. It has been shown that this scotoma can be filled in with surrounding information, and that humans tend to trust such inferred central information more than veridical peripheral information. However, it remains unclear whether these effects also occur in mesopic vision where cones are active in the fovea. We investigated filling-in of the rod scotoma and the confidence preference in photopic, scotopic, and mesopic vision. In the mesopic condition, rods and cones were independently stimulated using a tetrachromatic projector. Our stimuli comprised two concentric circles of sine wave gratings, with the center smaller than the size of rod scotoma to allow foveal filling-in by surrounding information. The orientation of the center and the surround was parallel (continuous stimulus) or orthogonal (discontinuous stimulus). Participants had to determine the continuity of two successively presented stimuli and to choose for which stimulus they felt more confident. For rod-isolating stimuli in the fovea, participants were unable to discriminate continuous and discontinuous stimuli and were biased to perceive stimuli as continuous, both under scotopic and mesopic conditions. This suggests that the absence of rod information in the fovea was filled-in by the immediate surround, even when cones could supply information at the fovea in mesopic vision. Moreover, participants tended to trust the centrally inferred information more than veridical information in the periphery, both under scotopic and mesopic conditions. This suggests that the preference for inferred information in the foveal rod scotoma persists even under mesopic conditions.

This work was supported by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No. 101001250).

Talk 4, 6:00 pm

Probing metacognition with the “psychophysicist’s microelectrode”

Tarryn Balsdon¹, Paolo Bartolomeo², Vincent de Gardelle³, Pascal Mamassian¹, Marion Rouault²; ¹Ecole Normale Superieure and CNRS, ²Institut du Cerveau and INSERM, ³Paris School of Economics and CNRS

Self-evaluations of perceptual decision accuracy are thought to rely on a general metacognitive mechanism that computes confidence as a common currency across tasks. As a subjective variable, it is a challenge to directly manipulate confidence. Rather, the study of confidence has relied on correlational approaches, indirectly manipulating confidence by varying the available decision evidence. Here, we tested whether we can directly manipulate confidence by appropriating the classical psychophysical method of sensory adaptation to confidence adaptation. We tested 100 observers’ metacognitive sensitivity in a forced-choice confidence paradigm: Participants were asked to choose which of two orientation discrimination decisions was more likely to be correct. This followed a long sequence of numerosity and colour discrimination decisions, in which the task difficulty was manipulated to either be varied (baseline), relatively difficult (low-confidence adaptation), or relatively easy (high-confidence adaptation). The presented stimuli, orientation discrimination sensitivity, and thus the underlying decision evidence relevant for evaluating confidence, was unchanged following each of these prolonged exposure sequences of numerosity/colour discrimination decisions. Yet, we found a systematic change in metacognitive sensitivity following prolonged exposure to high/low confidence decisions compared to baseline, consistent with cross-task confidence adaptation. The pattern of this change in metacognitive sensitivity disambiguates models of the neural coding scheme of confidence, suggesting a dual-channel coding scheme involving tuning to both increasing confidence and increasing uncertainty. Greater channel responsiveness and more channel overlap predicted better metacognitive sensitivity. This cross-task adaptation provides direct evidence for general metacognitive computations in human observers, and demonstrates adaptive resource allocation for metacognitive processes.

This work was supported by the Fyssen Foundation

Talk 5, 6:15 pm

Confidence accurately tracks increasing internal noise in peripheral vision

Luhe Li¹, Michael S. Landy^1,2; ¹Department of Psychology, New York University, ²Center for Neural Science, New York University

Visual sensitivity declines with increasing eccentricity, yet it remains unclear whether we are aware of this. Previous studies report both overconfidence and underconfidence in peripheral vision, with mixed results complicated by conflicting definitions of confidence. Here, we used the most common definition of confidence, the perceived probability that a decision is accurate. We tested the Bayesian-confidence hypothesis that confidence is derived from the posterior probability distribution of the feature given noisy sensory measurements. The first-order task was visual localization and confidence was measured using post-decision wagering. Participants fixated on a central crosshair monitored by an eye tracker. A 33 ms Gaussian blob appeared at a random horizontal location (uniformly from –31.5 to 31.5°). Participants adjusted a cursor to localize the target and reported confidence by setting a symmetrical range around the localization response. They earned higher points for narrower confidence ranges but received zero points if the range did not encompass the target. Points decreased linearly with range length, which incentivized both accurate localization and confidence reports. Feedback was only given during practice to ensure participants learned the cost function, with no trial-wise feedback in the main experiment. Both localization variability and confidence range increased with eccentricity. We estimated the sensory noise from the localization responses assuming that sensory noise scales with eccentricity. We then compared two confidence models. The ideal observer uses accurate estimates of sensory noise and combines the posterior distributions of location with a subjective cost function to maximize expected gain. The heuristic observer determines the confidence range proportionally to the distance of location estimates from fixation. Model comparison shows that confidence reports were better fit by the ideal-observer model. Humans can accurately monitor the increasing internal noise in peripheral vision and use it to make optimal confidence judgments.

NIH EY08266

Talk 6, 6:30 pm

The influence of response bias on confidence and accuracy in multi-alternative tasks for humans and artificial neural networks

Bogeng Song¹, Dobromir Rahnev¹; ¹Georgia Institute of Technology

How does response bias influence confidence judgments in perceptual decision making? Most previous research has used 2-choice tasks, which makes it difficult to address this question. Indeed, in 2-choice tasks, bias for one category is equivalent to a bias against the other category, meaning that one cannot independently assess the bias for each category. Therefore, to determine how response bias influences confidence judgments, here we used two datasets with multi-alternative tasks. In the first dataset, 60 subjects performed an 8-choice digit categorization task using noisy MNIST images. In the second dataset, 37 subjects performed a 16-choice image classification task using blurred ImageNet images. We quantified the bias for each stimulus category as the rate at which a category was chosen when the stimulus was selected from all other categories. Unlike for 2-choice tasks, this procedure allowed us to estimate the bias for each category independently from the remaining categories. We then used the response bias for each category to predict the confidence and accuracy for that category. We found that a bias towards a given category predicted higher accuracy. Counterintuitively, however, being biased towards a category predicted lower confidence ratings for that category. These results were robust to variations in task difficulty and condition. We further observed that artificial neural networks (ANNs) trained on object recognition replicated the positive effect of response bias on accuracy, but, in contrast to humans, also showed a positive effect of response bias on confidence. These findings suggest that when rating confidence, humans – but not ANNs – actively discount evidence for stimulus categories that they are biased towards. More broadly, our results demonstrate the utility of complex, multi-alternative tasks in discovering the mechanisms driving visual confidence.

This work was supported by the National Institute of Health (award: R01MH119189) and the Office of Naval Research (award: N00014-20-1-2622).

Talk 7, 6:45 pm

Response duration: A ubiquitous implicit measure of confidence

Hanbei Zhou¹ (), Rui Zhe Goh², Ian Phillips³, Chaz Firestone⁴; ¹Johns Hopkins University

Among the most reliable connections between internal mental processing and external behavior is *response time*, with easier, more accurate, and more confident judgments typically made faster. But which aspects of response time are relevant? Whereas psychophysical studies traditionally focus on the time taken to initiate a response, an underexplored measure is the duration of the response itself—not just the amount of time between stimulus onset and keypress (reaction time), but also how long one holds down the key before releasing it (response duration). Response duration is a ubiquitous and freely available data source, yet almost no studies report or analyze it (Pfister et al., 2023). Here, 3 varied experiments demonstrate that response duration reliably predicts subjective confidence, independent of reaction time. In Experiment 1, subjects detected faces within white noise, with difficulty manipulated by varying face opacity. Subjects responded with a keypress (with both keyUp and keyDown events recorded separately), followed by a confidence judgment. Remarkably, subjects held down the response key longer during trials in which they subsequently reported lower confidence, as if making these face-detection judgments in a tentative fashion. The same pattern held in another visual task (judging the coherence of random-dot motion; Experiment 2), and a cognitive task (classifying American cities as geographically Eastern or Western; Experiment 3). In all cases, response duration accounted for variance in confidence that was not predicted by reaction time. Response duration has distinct advantages as a measure of confidence: It taps confidence at the time of judgment (rather than retrospectively), it can be used when traditional confidence judgments are difficult to elicit (e.g., in animals or infants), and it may be less affected by biases associated with explicit reports. Our results suggest that response duration is a valuable and untapped source of information, raising many avenues for future investigation.