Several studies in rodents and humans have shown that category information (frequency in which pieces of evidence favor a response option; CI) can increase primacy and confirmation biases, whereas sensory information (stimulus strength towards the correct option; SI) can do the opposite. Guided by a new qualitative modular learning model, our behavioral experiment in humans aimed to show that CI and SI can modulate task accuracy through changes in integration biases and degree of caution. 59 healthy university students (86% females) performed two decision-making tasks. In Clouds-of-Dots trials, participants identified the dominant color of a sequence of 10 static dot clouds (red or blue) with a right-left bottom press. Then, they rated their response confidence (1–4 Likert scale). Correct responses earned virtual coins. In Preference trials, participants were shown five pairs of images (e.g., animals, fruits). The first four showed the preferences of an unknown player (e.g., orange > apple, …). In exchange for extra coins at the end of the experiment, participants had to infer the preferred stimulus at the fifth pair. Immediate feedback only included the true preference. As expected, higher SI increased accuracy (p<.001, r=.87). However, strikingly, low CI increased accuracy (p<.001, r=.87) and confidence (p=.01, d=.27), while decreasing reaction time (p=.019, d=.35). The accuracy improvement tended to be stronger in cautious participants (more negative metacognitive bias and that increased their reaction time in difficult trials; p=.07). Moreover,low CI tended to increase caution (p=.07, d=-.27), while low SI increased it (p<.001, d=2.16). On the other hand, 42% of participants showed integration bias patterns (e.g., U-shaped kernels) that have not been predicted by previous models. In conclusion, each information type seemed to have opposite effects in deliberation caution, and cause a higher diversity of integration bias patterns than previously found or predicted. A dual model of learning may be needed to account for these results.
Towards a dual learning system in humans: sensory and category information cause opposed bias in behavior
Alejandro Sospedra
2025
Abstract
Several studies in rodents and humans have shown that category information (frequency in which pieces of evidence favor a response option; CI) can increase primacy and confirmation biases, whereas sensory information (stimulus strength towards the correct option; SI) can do the opposite. Guided by a new qualitative modular learning model, our behavioral experiment in humans aimed to show that CI and SI can modulate task accuracy through changes in integration biases and degree of caution. 59 healthy university students (86% females) performed two decision-making tasks. In Clouds-of-Dots trials, participants identified the dominant color of a sequence of 10 static dot clouds (red or blue) with a right-left bottom press. Then, they rated their response confidence (1–4 Likert scale). Correct responses earned virtual coins. In Preference trials, participants were shown five pairs of images (e.g., animals, fruits). The first four showed the preferences of an unknown player (e.g., orange > apple, …). In exchange for extra coins at the end of the experiment, participants had to infer the preferred stimulus at the fifth pair. Immediate feedback only included the true preference. As expected, higher SI increased accuracy (p<.001, r=.87). However, strikingly, low CI increased accuracy (p<.001, r=.87) and confidence (p=.01, d=.27), while decreasing reaction time (p=.019, d=.35). The accuracy improvement tended to be stronger in cautious participants (more negative metacognitive bias and that increased their reaction time in difficult trials; p=.07). Moreover,low CI tended to increase caution (p=.07, d=-.27), while low SI increased it (p<.001, d=2.16). On the other hand, 42% of participants showed integration bias patterns (e.g., U-shaped kernels) that have not been predicted by previous models. In conclusion, each information type seemed to have opposite effects in deliberation caution, and cause a higher diversity of integration bias patterns than previously found or predicted. A dual model of learning may be needed to account for these results.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
