Detection & Recognition

Recent studies on human-AI interactions have strived to acquire data about how domain specialists in a task would follow an AI advice. They use the judge-advisor system (JAS), an experimental paradigm formerly interested in regressions, and analyze reliance behaviors on classification tasks. In this stream of research, few studies focused on developing models of influence, as effects found are often mixed, and difficult to generalize. Tejeda et al. (2022) proposed a consistent probabilistic model of this JAS applied on a lay image classification task ; however the reliance matrix they introduce still needs refinement in order to make sense of unexpected results. Iterating on this work, we propose a probabilistic modeling approach to quantify how advice influence decisions. We test on both Tejeda et al.’s dataset (open acces), and on an application of JAS to mine countermeasures. These data comes from an experimental study (soon to be published) performed with a sample of sonar analysts with varied degrees of experience. In this setting, three advisors (2 humans, 1 AI) help the operator classifying sonar images as mines or non-mines. Results show the feasibility of parsimonious models to predict human behavior, aimed at estimating impacts of organizational changes. Our models also allow us to assess whether individuals or data items are the predominant factor in this kind of decisional setting.

This is an in-person presentation on July 20, 2026 (15:00 ~ 15:20 EDT).

Sternberg’s classic short-term memory scanning experiment has been frequently used to evaluate the mental processes underlying human memory. Generally, participants are presented with a list of stimuli who indicate whether a probe stimulus was part of that list. Here, we adopt Sternberg’s approach for memory scanning of musical tones that differ only in their fundamental frequency, and therefore, their pitch, in musicians and non-musicians. In this experiment, the stimuli used are perceived on a continuum, rather than the more commonly used categorical stimuli. We apply the tools of Systems Factorial Technology (Townsend & Nozawa, 1995) to evaluate whether musicians have different mental architectures or workload capacities for musical tones from non-musicians, given their extensive auditory experiences. In this experiment, participants were given a potential set of 12 notes selected based on the C-major scale. Sequences were 1, 2, or 3 notes. After presentation of the memory list, a probe note was presented. Participants indicated with a button press whether the probe was in the sequence or not. In this talk, we present the workload capacity measures of musician and nonmusician participants’ data on the YES and NO responses by comparing their performance in the 2- or 3- note conditions to performance in the 1- note condition utilizing the Townsend & colleagues (& Nozawa, 1995; & Wenger, 2004) capacity functions. Preliminary results indicate the musicians have superior workload capacity compared to nonmusicians, suggesting that even for stimuli perceived continuously and presented sequentially, musicians perceive musical notes as true gestalts.

This is an in-person presentation on July 20, 2026 (15:20 ~ 15:40 EDT).

General Recognition Theory (GRT) is a powerful framework for characterizing independence and separability in the perception of multidimensional stimuli. Using a factorial paradigm and statistical tests of report independence and marginal response invariance, patterns of identification errors reveal violations of independence and separability. Traditionally, the levels for each stimulus dimension must be tediously determined through pilot-testing such that participants make errors but are still able to perform the task. This process is resource-intensive and subject to individual differences. To address this issue, we previously introduced an adaptive procedure for fitting a highly constrained GRT model to participant responses during the experiment. In the present work, we report the results of a human validation study and compare them with the results of previously reported simulation and control studies. As expected, accuracy was more variable in the control study than in the adaptive study, particularly for integral stimuli. In both the control and adaptive studies, tests of marginal response invariance indicated violations of perceptual separability at a rate of approximately twice greater for integral stimuli than separable stimuli. There was no difference in the number of participants who exhibited a violation of perceptual separability in the control study, regardless of stimulus type, whereas in the adaptive study, participants were approximately twice as likely to violate perceptual separability with integral stimuli than with separable stimuli. Collectively, these results indicate the adaptive method leads to more consistent outcomes than the traditional approach.

This is an in-person presentation on July 20, 2026 (15:40 ~ 16:00 EDT).