Periodic Reporting for period 1 - METACURIO (What makes us curious? The role of metacognition.)
Reporting period: 2019-09-01 to 2021-08-31
This project addressed the question of the determinants of human curiosity. Curiosity is a basic biological drive that motivates learning and guides decision-making. Despite its ubiquitous influence over our daily lives, we have a very limited understanding of its function, mechanisms and neural basis. In the current project, several aspects were investigated to better understand the causes and consequences of curiosity and information-seeking behavior.
The overall goal of this project was to determine how metacognitive processes modulate curiosity. To accomplish this, we planned to combine experimental, computational and neuroimaging techniques in order to determine how metacognitive processes shape curiosity.
First, we found that information-seeking behavior is spontaneously biased towards regularities in the environment. We further found that the attentional bias effect diminished with continued exposure to the regularities. This highlights a spontaneous bias of attention towards task-irrelevant regularities that is transient, and proposes that learning of regularities may play a key role in moderating information seeking behavior.
Second, we examined the relationship between risk, ambiguity and curiosity. We assessed this question by designing a novel experiment, in which people were asked to indicate their desire to know the outcome of an uncertain event, while manipulating these factors independently. The results showed that while curiosity was strongly related to risk, showing linear increases in curiosity with increasing entropy, there was no additional variance explained in curiosity by ambiguity. This is an important building block and finding for understanding the determinants of human curiosity.
Finally, we examined curiosity in a learning context, in which participants were provided with small samples of information and asked to form hypotheses about the underlying generative model that produces the samples.
The work carried out during this project has had (and is having) substantial impact on the field of curiosity and information-seeking, in elucidating the behavioral and computational mechanisms of curiosity, and how it is shaped by various external and internal (metacognitive) factors.
Our findings have important implications for diverse fields such as learning, decisionmaking, and robotics. A better understanding of curiosity and metacognition may also have important implications for the education sector, as both of these mechanisms have been implicated as important predictors of academic success.
The overall goal of this project was to determine how metacognitive processes modulate curiosity. To accomplish this, we planned to combine experimental, computational and neuroimaging techniques in order to determine how metacognitive processes shape curiosity.
First, we found that information-seeking behavior is spontaneously biased towards regularities in the environment. We further found that the attentional bias effect diminished with continued exposure to the regularities. This highlights a spontaneous bias of attention towards task-irrelevant regularities that is transient, and proposes that learning of regularities may play a key role in moderating information seeking behavior.
Second, we examined the relationship between risk, ambiguity and curiosity. We assessed this question by designing a novel experiment, in which people were asked to indicate their desire to know the outcome of an uncertain event, while manipulating these factors independently. The results showed that while curiosity was strongly related to risk, showing linear increases in curiosity with increasing entropy, there was no additional variance explained in curiosity by ambiguity. This is an important building block and finding for understanding the determinants of human curiosity.
Finally, we examined curiosity in a learning context, in which participants were provided with small samples of information and asked to form hypotheses about the underlying generative model that produces the samples.
The work carried out during this project has had (and is having) substantial impact on the field of curiosity and information-seeking, in elucidating the behavioral and computational mechanisms of curiosity, and how it is shaped by various external and internal (metacognitive) factors.
Our findings have important implications for diverse fields such as learning, decisionmaking, and robotics. A better understanding of curiosity and metacognition may also have important implications for the education sector, as both of these mechanisms have been implicated as important predictors of academic success.
Due to a combination of external circumstances (e.g. the covid pandemic) and internal circumstances (related to integration, housing, and mental health) several amendments to the plan had to be made. Specifically, the WP2 on altering metacognitive processing using fMRI neurofeedback turned out to be too ambitious given the set of formerly mentioned circumstances, and the projects instead focused on psychophysical and computational principles of curiosity and its relationship to metacognition (WP1).
Work Package 1:
First, we designed a behavioral study to test the relationship between information-seeking behavior and regularities in the environment. We reassessed the spontaneous bias of attention towards task-irrelevant statistical regularities reported by Zhao, Al-Aidroos, and Turk-Browne (Psychological Science, 24(5), 667–677, 2013) using a larger sample size (N=90). We additionally extended the study to investigate the potential role of learning, doubling the duration of the experiment to determine whether the attentional bias towards regularities is reduced with continued exposure. We found that performance on a visual search task was better in a location containing task-irrelevant statistical regularities, supporting the conclusion that information-seeking behavior is spontaneously biased towards regularities in the environment. We further found that the attentional bias effect diminished with continued exposure to the regularities. This highlights a spontaneous bias of attention towards task-irrelevant regularities that is transient, and proposes that learning of regularities may play a key role in moderating information seeking behavior. This work has been accepted for publication in the Open Access journal Royal Society Open Science (in press).
Second, we examined the relationship between risk, ambiguity and curiosity. Risk and ambiguity are both factors that influence the uncertainty of a situation, but they are to some extent distinct: risk refers to the unknowable, irreducible uncertainty, whereas ambiguity refers to knowable, reducible uncertainty. We felt it was important to first establish firmly which form of uncertainty has a stronger relationship with curiosity. It is well known that curiosity is monotonically related to uncertainty, but it is currently unknown whether the uncertainty relates to reducible or irreducible forms of uncertainty. We assessed this question by designing a novel experiment, in which people were asked to indicate their desire to know the outcome of a ball being picked out of a vase, while manipulating these factors independently (cf van Lieshout, de Lange, Cools, 2021). The results showed that while curiosity was strongly related to risk, showing linear increases in curiosity with increasing entropy, there was no additional variance explained in curiosity by ambiguity. This is an important building block and finding for understanding the determinants of human curiosity. The manuscript that describes these results is currently being prepared, and in its final stage. We intend to publish this article in the Open Access journal Royal Society Open Science.
Third, we examined curiosity in a learning context, in a large (N=120) online sample of participants. We designed a novel task, in which participants are provided with small samples of information and are asked to form hypotheses about the underlying generative model that produces the samples. We probed curiosity about upcoming samples, as a function of uncertainty, metacognitive confidence, and prediction error as participants were exposed to the sampling over time. We also manipulated the volatility of the environment, to examine whether participants incorporate this parameter in their metacognitive confidence. Results are currently being analyzed, in the context of a state-of-the-art computational framework (active inference), with help of a visiting PhD student.
Work Package 1:
First, we designed a behavioral study to test the relationship between information-seeking behavior and regularities in the environment. We reassessed the spontaneous bias of attention towards task-irrelevant statistical regularities reported by Zhao, Al-Aidroos, and Turk-Browne (Psychological Science, 24(5), 667–677, 2013) using a larger sample size (N=90). We additionally extended the study to investigate the potential role of learning, doubling the duration of the experiment to determine whether the attentional bias towards regularities is reduced with continued exposure. We found that performance on a visual search task was better in a location containing task-irrelevant statistical regularities, supporting the conclusion that information-seeking behavior is spontaneously biased towards regularities in the environment. We further found that the attentional bias effect diminished with continued exposure to the regularities. This highlights a spontaneous bias of attention towards task-irrelevant regularities that is transient, and proposes that learning of regularities may play a key role in moderating information seeking behavior. This work has been accepted for publication in the Open Access journal Royal Society Open Science (in press).
Second, we examined the relationship between risk, ambiguity and curiosity. Risk and ambiguity are both factors that influence the uncertainty of a situation, but they are to some extent distinct: risk refers to the unknowable, irreducible uncertainty, whereas ambiguity refers to knowable, reducible uncertainty. We felt it was important to first establish firmly which form of uncertainty has a stronger relationship with curiosity. It is well known that curiosity is monotonically related to uncertainty, but it is currently unknown whether the uncertainty relates to reducible or irreducible forms of uncertainty. We assessed this question by designing a novel experiment, in which people were asked to indicate their desire to know the outcome of a ball being picked out of a vase, while manipulating these factors independently (cf van Lieshout, de Lange, Cools, 2021). The results showed that while curiosity was strongly related to risk, showing linear increases in curiosity with increasing entropy, there was no additional variance explained in curiosity by ambiguity. This is an important building block and finding for understanding the determinants of human curiosity. The manuscript that describes these results is currently being prepared, and in its final stage. We intend to publish this article in the Open Access journal Royal Society Open Science.
Third, we examined curiosity in a learning context, in a large (N=120) online sample of participants. We designed a novel task, in which participants are provided with small samples of information and are asked to form hypotheses about the underlying generative model that produces the samples. We probed curiosity about upcoming samples, as a function of uncertainty, metacognitive confidence, and prediction error as participants were exposed to the sampling over time. We also manipulated the volatility of the environment, to examine whether participants incorporate this parameter in their metacognitive confidence. Results are currently being analyzed, in the context of a state-of-the-art computational framework (active inference), with help of a visiting PhD student.
The research has pushed the boundaries of psychological research into curiosity, by employing state-of-the-art computational models from the reinforcement learning and active inference field to better understand information seeking behavior. It is expected that it will lead to at least two more scientific publications, which are currently being prepared. While the research is of fundamental nature, it is expected that the findings will have societal relevance and impact for the educational sector, in informing about optimal strategies to encourage curiosity behavior and engagement in the classroom.