Search

Police investigators worldwide use lineups to test an eyewitness's memory of a perpetrator. A typical lineup consists of one suspect (who is innocent or guilty) plus five or more fillers who resemble the suspect and who are known to be innocent. Although eyewitness identification decisions were once biased by police pressure and poorly constructed lineups, decades of social science research led to the development of reformed lineup procedures that provide a more objective test memory. Under these improved testing conditions, cognitive models of memory can be used to better understand and ideally enhance eyewitness identification performance. In this regard, one question that has bedeviled the field for decades is how similar the lineup fillers should be to the suspect to optimize performance. Here, we model the effects of manipulating filler similarity to better understand why such manipulations have the intriguing effects they do. Our findings suggest that witnesses rely on a decision variable consisting of the degree to which the memory signal for a particular face in the lineup stands out relative to the crowd of memory signals generated by the set of faces in the lineup. The use of that decision variable helps to explain why discriminability is maximized by choosing fillers that match the suspect on basic facial features typically described by the eyewitness (e.g., age, race, gender) but who otherwise are maximally

The simultaneous six-pack photo lineup is a standard eyewitness identification procedure, consisting of one police suspect plus five physically similar fillers. The photo lineup is either a target-present array (the suspect is guilty) or a target-absent array (the suspect is innocent). The eyewitness is asked to search the six photos in the array with respect to a target template stored in memory (namely, the memory of the perpetrator's face). If the witness determines that the perpetrator is in fact in the lineup (detection), then the next step is to specify the position of the perpetrator's face in the lineup (localization). The witness may also determine that the perpetrator is not present and reject the lineup. In other words, a police lineup is a detection-plus-localization visual search task. Signal detection concepts that have long guided thinking about visual search have recently had a significant impact on our understanding of police lineups.

In many decision tasks, we have a set of alternative choices and are faced with the problem of how to use our latent beliefs and preferences about each alternative to make a single choice. Cognitive and decision models typically presume that beliefs and preferences are distilled to a scalar latent strength for each alternative, but it is also critical to model how people use these latent strengths to choose a single alternative. Most models follow one of two traditions toestablish this link. Modern psychophysics and memory researchers make use of signal detection theory, assuming that latent strengths are perturbed by noise, and the highest resulting signal is selected. By contrast, many modern decision theoretic modeling and machine learning approaches use the softmax function (which is based on Luce’s choice axiom; Luce, 1959) to give some weight to non-maximal-strength alternatives. Despite the prominence of these two theories of choice, current approaches rarely address the connection between them, and the choice of one or the other appears more motivated by the tradition in the relevant literature than by theoretical or empirical reasons to prefer one theory to the other. The goal of the current work is to revisit this topic by elucidating which of these two models provides a better characterization of latent processes in m-alternative decision tasks, with a particular focus on memory tasks. In a set of visual memory experiments, we show that, within the same experimental design, the softmax parameter β varies across m-alternatives, whereas the parameter d′ of the signal-detection model is stable. Together, our findings indicate that replacing softmax with signal-detection link models would yield more generalizable predictions across changes in task structure. More ambitiously, the invariance of signal detection model parameters acrossdifferent tasks suggests that the parametric assumptions of these models may be more than just a mathematical convenience, but reflect something real about human decision-making.

How do people detect lies from the content of messages, and design lies that go undetected? Lying requires strategic reasoning about how others think and respond. We propose a unified framework underlying lie design and detection, formalized as recursive social reasoning. Senders design lies by inferring the likelihood the receiver detects potential lies; receivers detect lies by inferring if and how the sender would lie. Under this framework, we can predict the rate and content of lies people produce, and which lies are detected. In Experiment 1, we show that people calibrate the extremeness of their lies and what lies they detect to beliefs about goals and the statistics of the world. In Experiment 2, we present stronger diagnostic evidence for the function of social reasoning in lying: people cater their lies to their audience, even when their audience's beliefs differ from their own. We conclude that recursive and rational social reasoning is a key cognitive process underlying how people communicate in adversarial settings. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Working memory is a reconstructive process that requires integrating multiple hierarchical representations of objects. This hierarchical reconstruction allows us to overcome perceptual uncertainty and limited cognitive capacity, but yields systematic biases in working memory as individual items are influenced by the ensemble statistics of the scene, or of their particular group. Given the importance of the hierarchical encoding of a display, we aim to characterize what structured priors people use to encode visual scenes using a nonparametric data-driven approach. In Experiment 1, we examine visuospatial memory’s priors for locations by asking participants to recall the locations of objects in a serial reproduction task. We show that people have priors that bias items toward a more compact structure and organize them into clustered spatial groups. In Experiment 2, we explicitly introduce discrete color groups, allowing us to test whether the color feature governs the spatial grouping. We find that the spatial structures were color-contingent. By analyzing color groups, we circumvent the grouping uncertainty in Experiment 1 and further reveal spatial priors that compress color groups into collinear structures with similar orientations and equidistant spacing.

Pietraszewski proposes four triadic “primitives” for representing social groups. We argue that, despite surface differences, these triads can all be reduced to similar underlying welfare trade-off ratios, which are a better candidate for social group primitives. Welfare trade-off ratios also have limitations, however, and we suggest there are multiple computational strategies by which people recognize and reason about social groups.

Network-based approaches are a popular representational scheme of semantic units and the relations about them. Recent network-based approaches use graph theoretic analyses to examine individual differences in semantic networks and how they relate to other higher-level cognitive processes. However, it remains ambiguous whether individual differences captured via network analyses reflect true differences in the content and structure of latent knowledge schemes, or differences in how people strategically approach semantic relatedness tasks. To test this, we examine the reliability of content- and structure- based metrics of semantic networks across different semantic relatedness tasks. In two experiments, we find that both weighted and unweighted graph theoretic representations reliably capture individual differences in content-based measures of semantic networks. We find that we can predict individual differences in connections between semantic units across different semantic relatedness tasks. Furthermore, a measure of node centrality, which captures the relative ‘importance’ of units within the network, is reliable across tasks. In contrast, we find that metrics of structural properties of semantic networks, such as the average clustering coefficient and shortest path length, are less robust across tasks and may not provide reliable individual difference measures. We discuss the implications of these results and offer recommendations for researchers who seek to apply graph theoretic analyses in the study of individual differences in semantic memory.

People are concerned with signalling their social status to others, and conspicuous consumption may be a prevalent means of signalling, such that purchasing decisions are motivated not only by the direct value of a product, but by the indirect value gained from what the product might communicate to others. Here we measure which products people might use as signals and ask how the signalling potential of products relates to the distribution of product offerings in the consumer goods market. In particular, we asked how the signalling potential of products influences the number, price, and dispersion of prices within and across department stores. Using data scraped from 11 department stores, we found that products with greater signalling potential are available in greater quantity, more expensive within a given store, and that more expensive stores stock more products with higher signalling potential, leading to greater global variance in prices for goods with greater signalling potential. Further, we use product visibility as an instrumental variable to estimate the causal effect of signalling potential on product offerings. Altogether, these results suggest that consumers demand to use visible goods as social signals, and being sensitive to this demand, suppliers of consumer goods position their product offerings to supply ample material for signalling via consumption.

Is cognitive science interdisciplinary or multidisciplinary? We contribute to this debate by examining the authorship structure and topic similarity of contributions to the Cognitive Science Society from 2000 to 2019. Our analysis focuses on graph theoretic features of the co-authorship network-edge density, transitivity, and maximum subgraph size-as well as clustering within the space of scientific topics. We also combine structural and semantic information with an analysis of how authors choose their collaborators based on their interests and prior collaborations. We compare findings from CogSci to abstracts from the Vision Science Society over the same time frame and validate our approach by predicting new collaborations in the 2020 CogSci proceedings. Our results suggest that collaboration across authors and topics within cognitive science has become increasingly integrated in the last 19 years. More broadly, we argue that a formal quantitative approach which combines structural co-authorship information and semantic topic analysis provides inroads to questions about the level of interdisciplinary collaboration in a scientific community.

Author(s): Holdaway, Cameron; Vul, Ed | Abstract: Humans are social beings, and most of our decisions are influenced by considerations of how others will respond. Whether in poker or political negotiations, the riskiness of a decision is often determined by the variance of the other party's possible responses. Such socially-contingent decisions can be framed in terms of adversarial games, which differ from other risky situations such as lotteries because the risk arises from uncertainty about the opponent's decisions, and not some independent stochasticity in the world. We use chess as a lens through which we can study human risk-taking behavior in adversarial decision making. We develop a novel algorithm for calculating the riskiness of each move in a chess game, and apply it to data from over 1 billion online chess games. We find that players not only exhibit state-dependent risk preferences, but also change their risk-taking strategy depending on their opponent, and that this effect differs in experts and novices.

People seem to infer each others’ beliefs and desires when navigating social interactions, perhaps because such a “theory of mind” can guide cooperation and coordination. However, such strategic, altruistic interactions fall naturally out of evolutionary game theory without invoking any theory of mind; so why is theory of mind useful? Here we show that the interactions studied in game theory have been too impoverished to require theory of mind, but when interacting in variable games, agents with theory of mind have a clear advantage. We use simulated tournaments to demonstrate that traditional action-level strategies such as tit-for-tat fare miserably in variable games, that goal-based agents can adapt to new games instantly, and that having a theory of mind is increasingly helpful for coping with a variety of opponents as the variability in games increases. Our work suggests that variable games merit further investigation in game theory and social sciences.

Existing knowledge shapes and distorts our memories, serving as a prior for newly encoded information. Here, we investigate the role of stable long-term priors (e.g. categorical knowledge) in conjunction with priors arising from recently encountered information (e.g. 'serial dependence') in visual working memory for color. We use an iterated reproduction paradigm to allow a model-free assessment of the role of such priors. In Experiment 1, we find that participants' reports reliably converge to certain areas of color space, but that this convergence is largely distinct for different individuals, suggesting responses are biased by more than just shared category knowledge. In Experiment 2, we explicitly manipulate trial n-1 and find recent history plays a major role in participants' reports. Thus, we find that both global prior knowledge and recent trial information have biasing influences on visual working memory, demonstrating an important role for both short- and long-term priors in actively maintained information.

Is cognitive science interdisciplinary or multidisciplinary? We contribute to this debate by examining the authorship structure and topic similarity of contributions to the Cognitive Science Society from 2000 to 2019. We compare findings from CogSci to abstracts from the Vision Science Society over the same time frame. Our analysis focuses on graph theoretic features of the co-authorship network—edge density, transitivity, and maximum subgraph size—as well as clustering within the topic space of CogSci contributions. We also combine structural and semantic information with an analysis of homophily. We validate this approach by predicting new collaborations in this year’s CogSci proceedings. Our results suggest that cognitive science has become increasingly interdisciplinary in the last 19 years. More broadly, we argue that a formal quantitative approach which combines structural co-authorship information and semantic topic analysis provides inroads to questions about the level of interdisciplinary collaboration in the cognitive science community.

In this study, we investigate people’s ability to predict and adapt to the behavior of others in order to make plans of their own, a cornerstone of cooperative and competitive behavior. Participants played 300 rounds of rock, paper, scissors against another human player. We investigate the degree to which participants are able to identify patterns in their opponent’s behavior in order to exploit them in subsequent rounds. We find strong evidence that participants exploit their opponents over the course of 300 rounds, suggesting that people identify dependencies in their opponent’s move choices during the game. Nonetheless, analysis of dependencies across participant move choices reveals that people exhibit a number of regularities in their own moves. Based on these dependencies, we argue that participants are far from optimal in their exploiting, suggesting that there are substantial constraints on people’s ability to identify and adapt to patterned opponent behavior across repeated interactions.

Whether estimating the size of a crowd or rating a restaurant on a five-star scale, humans frequently navigate between subjective sensory experiences and shared formal systems. Here we ask how people manage this in the case of estimating number. We present participants with arrays of dots and ask them to report how many dots there are. Our results produce two novel findings. First, people's estimates are best fit by a bilinear function in log space, rather than the traditional power law described in previous literature. Second, we find that people's estimates do not have a stable coefficient of variation at higher magnitudes, and that the likely cause of this is a "drift" in people's estimate calibration over many trials which has not previously been identified. Building on these results, we present a model of the mapping function from subjective numerosity to symbolic number that relies primarily on a constrained set of previous estimates and familiar numerosities, rather than the robust internal scale used in existing models. Our model is able to generate an accurate mapping with limited data and reproduce notable aspects of estimation seen in our experimental results. This suggests that human number estimation, and perhaps other domains in which we must navigate between subjective representations and formal systems, is governed by a relatively simple decision process that primarily seeks to maintain consistency with previous estimates. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Does human behavior exploit deep and accurate knowledge about how the world works, or does it rely on shallow and often inaccurate heuristics? This fundamental question is rooted in a classic dichotomy in psychology: human intuitions about even simple scenarios can be poor, yet their behaviors can exceed the capabilities of even the most advanced machines. One domain where such a dichotomy has classically been demonstrated is intuitive physics. Here we demonstrate that this dichotomy is rooted in how physical knowledge is measured: extrapolation of ballistic motion is idiosyncratic and erroneous when people draw the trajectories but consistent with accurate physical inferences under uncertainty when people use the same trajectories to catch a ball or release it to hit a target. Our results suggest that the contrast between rich and calibrated versus poor and inaccurate patterns of physical reasoning exists as a result of using different systems of knowledge across tasks, rather than being driven solely by a universal system of knowledge that is inconsistent across physical principles.

The world around us is filled with complex objects, full of color, motion, shape, and texture, and these features seem to be represented separately in the early visual system. Anne Treisman pointed out that binding these separate features together into coherent conscious percepts is a serious challenge, and she argued that selective attention plays a critical role in this process. Treisman also showed that, consistent with this view, outside the focus of attention we suffer from illusory conjunctions: misperceived pairings of features into objects. Here we used Treisman's logic to study the structure of pre-attentive representations of multipart, multicolor objects, by exploring the patterns of illusory conjunctions that arise outside the focus of attention. We found consistent evidence of some pre-attentive binding of colors to their parts, and weaker evidence of binding multiple colors of the same object. The extent to which such hierarchical binding occurs seems to depend on the geometric structure of multipart objects: Objects whose parts are easier to separate seem to exhibit greater pre-attentive binding. Together, these results suggest that representations outside the focus of attention are not entirely a "shapeless bundles of features," but preserve some meaningful object structure.

Modern society frequently requires that we express our subjective senses in objective, shared formal systems; this entails mapping multiple internal variables onto a common scale. Here we ask whether we accomplish this feat in the case of estimating number by learning a single mapping between explicit numbers and one integrated subjective estimate of numerosity, or if we separately map different perceptual features onto numbers. We present people with arrays of dots and ask them to report how many dots there are; we rely on the systematic under/overestimation of number at higher quantities to estimate error in the mapping function. By comparing how this error changes over time, as the mapping fluctuates for different visual cues to numerosity, we can evaluate whether these cues share a single mapping, or are mapped onto number individually. We find that area, size, and density all share a common mapping, indicating that people obtain a unified subjective estimate of numerosity before mapping it onto the formal number line.

Accounts of human physical reasoning based on simulation from a noisy physics engine have enjoyed considerable suc- cess in recent years. However, simulating complex physical dynamics can be a computationally expensive process, and it is possible that people use faster, cheaper shortcuts to make predictions and inferences in complicated physical scenarios. Here we asked people to predict the eventual destination of a ball on a 2D bumper table (in the style of Smith, de Peres, Vul, and Tenenbaum (2017)). We designed scenarios that we ex- pected would modulate the use of heuristics and simulation: the bumper table provided varying degrees of containment to constrain future outcomes and to make a containment heuristic more useful, and could have more or less internal structure to vary the reliability of noisy simulation. As the containment heuristic becomes more useful, and as simulation be- comes more expensive, we expected that people would switch from using simulation to rely more on rapid heuristic-based predictions and therefore respond faster. Instead, we found that even when containment was very predictive, people were progressively slower and less accurate as simulation complexity increased, indicating that they persisted in using simulation rather than containment heuristics.

The human ability to deceive others and detect deception has long been tied to theory of mind. We make a stronger argument: in order to be adept liars – to balance gain (i.e. maximizing their own reward) and plausibility (i.e. maintaining a realistic lie) – humans calibrate their lies under the assumption that their partner is a rational, utility-maximizing agent. We develop an adversarial recursive Bayesian model that aims to formalize the behaviors of liars and lie detectors. We compare this model to (1) a model that does not perform theory of mind computations and (2) a model that has perfect knowledge of the opponent’s behavior. To test these models, we introduce a novel dyadic, stochastic game, allowing for quantitative measures of lies and lie detection. In a second experiment, we vary the ground truth probability. We find that our rational models qualitatively predict human lying and lie detecting behavior better than the non-rational model. Our findings suggest that humans control for the extremeness of their lies in a manner reflective of rational social inference. These findings provide a new paradigm and formal framework for nuanced quantitative analysis of the role of rationality and theory of mind in lying and lie detecting behavior.

Multiple guesses from one individual, like guesses from a crowd, yield a better estimate when averaged. How far can such solipsistic polling take us in real, high-stakes settings? Now 1.2 million incentivized, real-world guesses show just how much people can improve their judgements by reconsidering their own estimates.

People commonly track objects moving in complex natural displays and their performance in the multiple object tracking paradigm has been used to study such visual attention for more than three decades. Given the theoretical and practical importance of object tracking, it is critical to understand how people solve the correspondence problem to track objects; however, it remains unclear what information people use to achieve this feat. In particular, although people can track multiple moving objects based on their positions, there is ambiguity about whether people can track objects via higher order kinematic information, such as velocity. We designed a paradigm in which position was rendered uninformative to directly examine whether people could use higher order kinematic information to track multiple objects. We find that people can track via velocity, but not acceleration, even though observers can reliably detect the acceleration cue that they cannot use for tracking. Furthermore, we show a capacity constraint on using higher order kinematic information-people perform worse when required to use velocity to resolve correspondence for multiple object pairs simultaneously. Together, our results suggest that, although people can use higher order kinematic information for object tracking, precise higher order kinematic information is not freely available from the early visual system.

© 2015 American Psychological Association. What happens to memories as we forget? They might gradually lose fidelity, lose their associations (and thus be retrieved in response to the incorrect cues), or be completely lost. Typical long-term memory studies assess memory as a binary outcome (correct/incorrect), and cannot distinguish these different kinds of forgetting. Here we assess long-term memory for scalar information, thus allowing us to quantify how different sources of error diminish as we learn, and accumulate as we forget. We trained subjects on visual and verbal continuous quantities (the locations of objects and the distances between major cities, respectively), tested subjects after extended delays, and estimated whether recall errors arose due to imprecise estimates, misassociations, or complete forgetting. Although subjects quickly formed precise memories and retained them for a long time, they were slow to learn correct associations and quick to forget them. These results suggest that long-term recall is especially limited in its ability to form and retain associations.

Does human behavior exploit deep and accurate knowledge about how the world works, or does it rely on shallow and often inaccurate heuristics? This fundamental question is rooted in a classic dichotomy in psychology: human intuitions about even simple scenarios can be poor, yet their behaviors can exceed the capabilities of even the most advanced machines. One domain where such a dichotomy has classically been demonstrated is intuitive physics. Here we demonstrate that this dichotomy is rooted in how physical knowledge is measured: extrapolation of ballistic motion is idiosyncratic and erroneous when people draw the trajectories, but consistent with accurate physical inferences under uncertainty when people use the same trajectories to catch a ball or release it to hit a target. Our results suggest that the contrast between rich and calibrated, versus poor and inaccurate patterns of physical reasoning exist as a result of using different systems of knowledge across tasks, rather than as a universal system of knowledge that is inconsistent across physical principles.

How do we search through our vast stores of knowledge to retrieve information demanded by the context and task at hand? Davelaar (this issue) finds that the words people report while searching for one word that is meaningfully associated with three cues (e.g., cues: comb, moon, dew; answer: honey; the Remote Associates Test, RAT) seem to be drawn from a superadditive combination of the semantic associations of the three cues, rather than “clustered” in pre-existing patches as expected by previous theories of semantic search (e.g., Troyer, Moscovitch, & Winocur, 1997). Although we found similar behavior in the same paradigm (Smith, Huber, & Vul, 2013), our interpretation hinged on a different property of search: There is sequential dependence between words reported during the search process. We believe that this sequential dependence is a fundamental feature of semantic search that gives rise to apparent clustering in previous work as well as providing further insights into Davelaar’s current results. When people list words that come to mind while solving RAT problems, responses that are closer in the sequence are more similar to one another than would be expected based solely on independent word selection, given the cue constraints (Smith et al., 2013). This holds both for words from the same semantic patch and those in different patches, indicating that semantic search is not simply sampling from sets of words or patches constrained by the cues, but that instead, the current state of the search process influences what will be retrieved next. This sequential dependence has recently provided new explanations of behavioral phenomena in semantic search with a single cue. For instance, it has been well established that in the process of free recall—producing as many words as possible that relate to a given category—peoples’ responses “cluster” (e.g., when listing animals, they might list several farm animals then some aquatic animals, etc.; Bousfield, 1953; Bousfield &

Face recognition memory is often tested by the police using a photo lineup, which consists of one suspect, who is either innocent or guilty, and five or more physically similar fillers, all of whom are known to be innocent. For many years, lineups were investigated in lab studies without guidance from standard models of recognition memory. More recently, signal detection theory has been used to conceptualize lineup memory and to motivate receiver operating characteristic (ROC) analysis of lineup performance. Here, we describe three competing signal-detection models of lineup memory, derive their likelihood functions, and fit them to empirical ROC data. We also introduce the notion that memory signals generated by the faces in a lineup are likely to be correlated because, by design, those faces share features. The models we investigate differ in their predictions about the effect that correlated memory signals should have on the ability to discriminate innocent from guilty suspects. A popular compound signal detection model known as the Integration model predicts that correlated memory signals should impair discriminability. Empirically, this model performed so poorly that, going forward, it should probably be abandoned. The best-fitting model incorporates a principle known as “ensemble coding,” which predicts that correlated memory signals should enhance discriminability. The ensemble model aligns with a previously proposed theory of eyewitness identification according to which the simultaneous presentation of faces in a lineup enhances discriminability compared to when faces are presented in isolation because it permits eyewitnesses to detect and discount non-diagnostic facial features.

The idea that people learn detailed probabilistic generative models of the environments they interact with is intuitively appealing, and has received support from recent studies of implicit knowledge acquired in daily life. The goal of this study was to see whether people efficiently induce a probability distribution based upon incidental exposure to an unknown generative process. Subjects played a ‘whack-a-mole’ game in which they attempted to click on objects appearing briefly, one at a time on the screen. Horizontal positions of the objects were generated from a bimodal distribution. After 180 plays of the game, subjects were unexpectedly asked to generate another 180 target positions of their own from the same distribution. Their responses did not even show a bimodal distribution, much less an accurate one (Experiment 1). The same was true for a pre-announced test (Experiment 2). On the other hand, a more extreme bimodality with zero density in a middle region did produce some distributional learning (Experiment 3), perhaps reflecting conscious hypothesis testing. We discuss the challenge this poses to the idea of efficient accurate distributional learning.

International audience; Functional neuroimaging techniques have transformed our ability to probe the neurobiological basis of behaviour and are increasingly being applied by the wider neuroscience community. However, concerns have recently been raised that the conclusions that are drawn from some human neuroimaging studies are either spurious or not generalizable. Problems such as low statistical power, flexibility in data analysis, software errors and a lack of direct replication apply to many fields, but perhaps particularly to functional MRI. Here, we discuss these problems, outline current and suggested best practices, and describe how we think the field should evolve to produce the most meaningful and reliable answers to neuroscientific questions.

In many learning or inference tasks human behavior approximates that of a Bayesian ideal observer, suggesting that, at some level, cognition can be described as Bayesian inference. However, a number of findings have highlighted an intriguing mismatch between human behavior and standard assumptions about optimality: People often appear to make decisions based on just one or a few samples from the appropriate posterior probability distribution, rather than using the full distribution. Although sampling-based approximations are a common way to implement Bayesian inference, the very limited numbers of samples often used by humans seem insufficient to approximate the required probability distributions very accurately. Here, we consider this discrepancy in the broader framework of statistical decision theory, and ask: If people are making decisions based on samples—but as samples are costly—how many samples should people use to optimize their total expected or worst-case reward over a large number of decisions? We find that under reasonable assumptions about the time costs of sampling, making many quick but locally suboptimal decisions based on very few samples may be the globally optimal strategy over long periods. These results help to reconcile a large body of work showing sampling-based or probability matching behavior with the hypothesis that human cognition can be understood in Bayesian terms, and they suggest promising future directions for studies of resource-constrained cognition., United States. Office of Naval Research. Multidisciplinary University Research Initiative (Grant N00014-07-1-0937), United States. Intelligence Advanced Research Projects Activity (United States. Dept. of Interior Contract D10PC20023)

Many important problems require consideration of multiple constraints, such as choosing a job based on salary, location, and responsibilities. We used the Remote Associates Test to study how people solve such multiply-constrained problems by asking participants to make guesses as they came to mind. We evaluated how people generated these guesses by using Latent Semantic Analysis to measure the similarity between the guesses, cues, and answers. We found that people use two systematic strategies to solve multiply-constrained problems: (a) people produce guesses primarily on the basis of just one of the three cues at a time; and (b) people adopt a local search strategy--they make new guesses based in part on their previous guesses. These results inform how people combine constraints to search through and retrieve semantic information from memory.