Engagement is a critical motivational factor that has broad effects on learning, productivity, performance, and even satis-faction and happiness. However, it can also be impacted by a myriad of factors which have made it difficult to model anddesign interventions. Here we address this problem by developing an integrated metacognitive framework for understand-ing task engagement. We treat engagement as resulting from a unified metacognitive decision process where the gradientof engagement results from a common priority calculation. Priority signals are computed relative to a set of availabletasks and updated across time and environmental changes. We propose a metacognitive controller makes decisions aboutboth task switching (when to quit, next task) and cognitive resourcing (working memory, attention, etc) using the gradedpriority signals. By simultaneously choosing the task and allocating resources using the same graded signals, we capturethe complex dependencies of engagement with task errors, performance, and time allocation.

Explanation generation is ubiquitous, and is shaped by our individual and collective knowledge about a phenomenon. Some phenomena are not easily explained by known explanations and require the creation of original or unknown explanations (How does scientific knowledge grow?), while others are strongly associated with known explanations (What caused the Battle of Dunkirk?). The processes underlying explanation generation when known/unknown have different characterizations, and broad implications for learning and innovation. However, this distinction is not studied in prior work. We hypothesize that unknown (compared to known) explanations will: require relatively more sense-making––manipulation of acquired information; exhibit self-directed information search not captured by simple information foraging models; and result in looser knowledge networks containing semantically dissimilar foraged-for information. We embed these hypotheses into a model framework, and show that it captures the behavior of participants asked to generate explanations in unknown/known conditions using information they forage for on Wikipedia.

Traditionally, theory and practice of Cognitive Control are linked via literature reviews by human domain experts. This approach, however, is inadequate to track the ever-growing literature. It may also be biased, and yield redundancies and confusion. Here we present an alternative approach. We performed automated text analyses on a large body of scientific texts to create a joint representation of tasks and constructs. More specifically, 385,705 scientific abstracts were first mapped into an embedding space using a transformers-based language model. Document embeddings were then used to identify a task-construct graph embedding that grounds constructs on tasks and supports nuanced meaning of the constructs by taking advantage of constrained random walks in the graph. This joint task-construct graph embedding, can be queried to generate task batteries targeting specific constructs, may reveal knowledge gaps in the literature, and inspire new tasks and novel hypotheses.

Prior research has distinguished between acquiring new/related knowledge (information foraging) and restructuring col-lected data (sensemaking), demonstrating that both substantially contribute to the comprehension of unknown information.These behaviors are critical cognitive abilities that can lead to scientific success and to innovation. Yet, little is knownabout whether there are individual differences in these behaviors. We provide a paradigm to study how these cognitiveabilities are utilized as participants attempt to understand the causal structure of a fictitious islands ecosystem (e.g., Whatis making the animals sick?). Some causal structures are directly posed by the environment, and can be discovered by in-formation foraging, whereas others can only be derived by sensemaking by merging or splitting the causes and/or effectsof already acquired information. We expect to see individual differences in information foraging and in sensemaking asreflected by the type of structures reported and time spent collecting or assessing the data.

Cognitive science is an archipelago of concepts and models,with cross-pollination between topics of interest often prohib-ited by incompatible approaches. Despite this, behavioral per-formance universally depends on information transmission be-tween brain regions and is limited by physical and biologicalconstraints. These constraints can be formalized as informa-tion theoretic constraints on transmission, which provide nor-mative predictions across a surprising range of cognitive do-mains. To illustrate this, we describe a simple variable-lengthrate coding model built with Poisson processes, Bayesian in-ference, and an entropy-based decision threshold. This modelreplicates features of human task performance and provides aprincipled connection between a high-level normative frame-work and neural rate codes. We thereby integrate several dis-joint ideas in cognitive science by translating plausible con-straints into information theoretic terms. Such efforts to trans-late concepts, paradigms and models into common theoreti-cal languages are essential for synthesizing our rich but frag-mented understanding of cognitive systems.

Curiosity has been suggested to reflect a drive for learning and to constitute a learning opportunity signal. If rational, curiosity should incorporate the reliability of the information source: more credible sources should instill more curiosity and learning. We tested these hypotheses in a lab experiment (n = 23) and online replication (n = 64), where we randomly assigned 100 zoology trivia questions and answers to one of three different sources claimed to be .99, .90 and .75 valid, respectively. Participants rated their curiosity for each source-indicated answer, read the answers, rated their credibility, and then took a retest on the questions. We found that indicated source credibility significantly affected curiosity ratings yielding an average of .56z,.22z and -.78z, respectively. Similarly, response update (learning) increased with 87% (exp 1) and 67% (exp 2) per z-score rated credibility (both p < 10^-18). Manipulated source credibility thus influenced both curiosity and learning.