We examine how people judge the probabilities of real-world events, such as natural disasters in different countries. We find that the associations between the words and phrases that constitute these events, as assessed by vector space semantic models, strongly correlate with the probabilities assigned to these events by participants. Thus, for example, the semantic proximity of “earthquake” and “Japan” accurately predicts judgments regarding the probability of an earthquake in Japan. Our results suggest that the mechanisms and representations at play in language are also active in high- level domains, such as judgment and decision making, and that existing insights regarding these representations can be used to make precise, quantitative, a priori predictions regarding the probability estimates of individuals.

We present a computational model of the representativeness heuristic. This model is trained on the entire English language Wikipedia corpus, and is able to use representativeness to answer questions spanning a very large domain of knowledge. Our trained model mimics human behavior by generating the probabilistic fallacies associated with the representativeness heuristic. It also, however, achieves a high rate of accuracy on unstructured judgment problems, obtained from large quiz databases and from the popular game show Who Wants to be a Millionaire?. Our results show how highly simplistic cognitive processes, known to be responsible for some of the most robust and pervasive judgment biases, can be used to generate the type of flexible, sophisticated, high-level cognition observed in human decision makers.

People learn new categories on a daily basis, and the study of category learning is a major topic of research in cognitive science. However, most prior work has focused on how people learn categories over abstracted, artificial (and usually perceptual) representations. Little is known about how new categories are learnt for natural objects, for which people have extensive prior knowledge. We examine this question in three pre-registered studies involving the learning of new categories for everyday foods. Our models use word vectors derived from large-scale natural language data to proxy mental representations for foods, and apply classical models of categorization over these vectorized representations to predict participant categorization judgments. This approach achieves high predictive accuracy rates, and can be used to identify the real-world settings in which category learning is impaired. In doing so, it shows how existing theories of categorization can be used to predict and improve everyday cognition and behavior.

Memory is a crucial component of everyday decision making, yet little is known about how memory and choice processesinteract, and whether or not established memory regularities persist during memory-based decision making. In this paper,we introduce a novel experimental paradigm to study the differences between memory processes at play in standard listrecall versus in preferential choice. Using computational memory models, fit to data from two pre-registered experiments,we find that some established memory regularities (primacy, recency, semantic clustering) emerge in preferential choice,whereas others (temporal clustering) are significantly weakened relative to standard list recall. Notably, decision-relevantfeatures, such as item desirability, play a stronger role in guiding retrieval in choice. Our results suggest memory processesdiffer across preferential choice and standard memory tasks, and that choice modulates memory by differentially activatingdecision-relevant features such as what we like.

We quantitatively modeled and compared two types of errorsin numerical estimation for naturalistic judgment targets: map-ping errors and knowledge errors. Mapping errors occur whenpeople make mistakes reporting their beliefs about a particularnumerical quantity (e.g. by inflating small numbers), whereasknowledge errors occur when people make mistakes usingtheir knowledge about the judgment target to form their be-liefs (e.g. by overweighting or underweighting cues). In twostudies, involving estimates of the calories of common fooditems and estimates of infant mortality rates in various coun-tries, we found that knowledge error models predicted partic-ipant estimates with very high out-of-sample accuracy rates,significantly outperforming the predictions of mapping errormodels. The knowledge error models were also able to iden-tify the objects and concepts most associated with incorrectestimates, shedding light on the psychological underpinningsof numerical judgment.

People’s desire to be patient or impatient can fluctuate from moment to moment, yet little is known about the effects of variability in time preference on intertemporal choice behavior. We examine this issue through the lens of an exponential discounting model with noisy discount factors. We show that such a model can generate decreasing patience over time, accounting for behavioral patterns typically attributed to hyperbolic discounting, while also making reasonable predictions regarding violations of intertemporal dominance. Additionally, two experiments reveal that many participants do display noise in their discount factors, and that a noisy discount factor model outperforms hyperbolic models in terms of quantitative fit. Ultimately the majority of participants are best described by some type of exponential discounting model (with or without noisy discount factors). These results indicate that it may not be necessary to assume alternate forms of non- exponential discounting, as long as the discount factors in an exponential model are permitted to vary at random. These results also highlight the importance of allowing for different sources of noise in choice modeling.People’s desire to be patient or impatient can fluctuate from moment to moment, yet little is known about the effects of variability in time preference on intertemporal choice behavior. We examine this issue through the lens of an exponential discounting model with noisy discount factors. We show that such a model can generate decreasing patience over time, accounting for behavioral patterns typically attributed to hyperbolic discounting, while also making reasonable predictions regarding violations of intertemporal dominance. Additionally, two experiments reveal that many participants do display noise in their discount factors, and that a noisy discount factor model outperforms hyperbolic models in terms of quantitative fit. Ultimately the majority of participants are best described by some type of exponential discounting model (with or without noisy discount factors). These results indicate that it may not be necessary to assume alternate forms of non- exponential discounting, as long as the discount factors in an exponential model are permitted to vary at random. These results also highlight the importance of allowing for different sources of noise in choice modeling.

We examine the effect of variability in model parameters on the predictions of expected utility theory and cumulative prospect theory, two of the most influential choice models in decision making research. We find that zero-mean and symmetrically distributed noise in the underlying parameters of these models can systematically distort choice probabilities, leading to false conclusions. Likewise, differences in choice proportions across decision makers might be due to differences in the amount of noise affecting underlying parameters rather than to differences in actual parameter values. Our results suggest that care and caution are needed when trying to infer the underlying preferences of decision makers, or the effects of psychological, biological, economic, and demographic variables on these preferences.