Low-level “adaptive” and higher-level “sophisticated” human reasoning processes have been proposed to play opposing roles in the emergence of unpredictable collective behaviors such as crowd panics, traffic jams, and market bubbles. While adaptive processes are widely recognized drivers of emergent social complexity, complementary theories of sophistication predict that incentives, education, and other inducements to rationality will suppress it. We show in a series of multiplayer laboratory experiments that, rather than suppressing complex social dynamics, sophisticated reasoning processes can drive them. Our experiments elicit an endogenous collective behavior and show that it is driven by the human ability to recursively anticipate the reasoning of others. We identify this behavior, “sophisticated flocking”, across three games, the Beauty Contest and the “Mod Game” and “Runway Game”. In supporting our argument, we also present evidence for mental models and social norms constraining how players express their higher-level reasoning abilities. By implicating sophisticated recursive reasoning in the kind of complex dynamic that it has been predicted to suppress, we support interdisciplinary perspectives that emergent complexity is typical of even the most intelligent populations and carefully designed social systems.
Best, R. M., & Goldstone, R. L. (in press). Bias to (and away from) the Extreme: Comparing Two Models of Categorical Perception Effects. Journal of Experimental Psychology: Learning, Memory, and Cognition.
Categorical Perception (CP) effects manifest as faster or more accurate discrimination between objects that come from different categories compared to objects that come from the same category, controlling for the physical differences between the objects. The most popular explanations of CP effects have relied on perceptual warping causing stimuli near a category boundary to appear more similar to stimuli within their own category and/ or less similar to stimuli from other categories. Hanley and Roberson (2011), on the basis of a pattern not previously noticed in CP experiments, proposed an explanation of CP effects that relies not on perceptual warping, but instead on inconsistent usage of category labels. Experiments 1 and 2 in this paper show a pattern opposite the one Hanley and Roberson pointed out. Experiment 3, using the same stimuli but with different choice statistics (i.e., different probabilities of each face being the target), obtains the same pattern as the one Hanley and Roberson showed. Simulations show that both category label and perceptual models are able to reproduce the patterns of results from both experiments, provided they include information about the choice statistics. This suggests two conclusions. First, the results described by Hanley and Roberson should not be taken as evidence in favor of a category label model. Second, given that participants did not receive feedback on their choices, there must be some mechanism by which participants monitor their own choices and adapt to the choice statistics present in the experiment.
Goldstone, R. L., Gopnik, A., Thagard, P., & Ullman, T. D. (2018). Models of human scientific discovery. Proceedings of the 40th Annual Conference of the Cognitive Science Society. (pp. 29-30). Madison, Wisconsin: Cognitive Science Society.
The scientific understanding of scientific understanding has been a long-standing goal of cognitive science. A satisfying formal model of human scientific discovery would be a major intellectual achievement, requiring solutions to core problems in cognitive science: the creation and use of apt mental models, the prediction of the behavior of complex systems involving interactions between multiple classes of elements, high-level perception of noisy and multiply interpretable environments, and the active interrogation of a system through strategic interventions on it – namely, via experiments. Over the past decades there have been numerous attempts to build formal models that capture what Perkins (1981) calls some of the “mind’s best work” – scientific explanations for how the natural world works by systematic observation, prediction, and testing. Early work by Hebert Simon and his colleagues (Langley, Simon, Bradshaw, & Zytkow, 1987) developed production rule systems employing heuristics to tame extremely large conjoint search spaces of experiments to run and hypotheses to test. Qualitative physics approaches seek to understand physical phenomena by building non-numeric, relational models of the phenomena (Forbus, 1984). Some early connectionist models interpreted scientific explanation in terms of emerging patterns of strongly activated hypotheses that mutually support one another (Thagard, 1992).
Bouhlel, I., Wu, C. M., Hanaki, N., & Goldstone, R. L. (2018). Sharing is not erring: Pseudo-reciprocity in collective search. Proceedings of the 40th Annual Conference of the Cognitive Science Society. (pp. 156-161). Madison, Wisconsin: Cognitive Science Society.
Information sharing in competitive environments may seem counterintuitive, yet it is widely observed in humans and other animals. For instance, the open-source software movement has led to new and valuable technologies being released publicly to facilitate broader collaboration and further innovation. What drives this behavior and under which conditions can it be beneficial for an individual? Using simulations in both static and dynamic environments, we show that sharing information can lead to individual benefits through the mechanisms of pseudoreciprocity, whereby shared information leads to by-product benefits for an individual without the need for explicit reciprocation. Crucially, imitation with a certain level of innovation is required to avoid a tragedy of the commons, while the mechanism of a local visibility radius allows for the coordination of self-organizing collectives of agents. When these two mechanisms are present, we find robust evidence for the benefits of sharing—even when others do not reciprocate.
Goldstone, R. L., Kersten, A., & Carvalho, P. F. (2017). Categorization and Concepts. In J. Wixted (Ed.) Stevens’ Handbook of Experimental Psychology and Cognitive neuroscience, Fourth Edition, Volume Three: Language & Thought. New Jersey: Wiley. (pp. 275-317).
Concepts are the building blocks of thought. They are critically involved when we reason, make inferences, and try to generalize our previous experiences to new situations. Behind every word in every language lies a concept, although there are concepts, like the small plastic tubes attached to the ends of shoelaces, that we are familiar with and can think about even if we do not know that they are called aglets . Concepts are indispensable to human cognition because they take the “blooming, buzzing confusion” (James, 1890, p. 488) of disorganized sensory experiences and establish order through mental categories. These mental categories allow us to make sense of the world and predict how worldly entities will behave. We see, hear, interpret, remember, understand, and talk about our world through our concepts, and so it is worthy of reflection time to establish where concepts come from, how they work, and how they can best be learned and deployed to suit our cognitive needs.
Lara-Dammer, F., Hofstadter, D. R., & Goldstone, R. L. (2017). A computer model of context dependent perception in a very simple world. Journal of Experimental & Theoretical Artificial Intelligence, 29:6, 1247-1282. DOI: 10.1080/0952813X.2017.1328463
We propose the foundations of a computer model of scientic discovery that takes into account certain psychological aspects of human observation of the world. To this end, we simulate two main components of such a system. The first is a dynamic microworld in which physical events take place, and the second is an observer that visually perceives entities and events in the microworld. For reason of space, this paper focuses only on the starting phase of discovery, which is the relatively simple visual inputs of objects and collisions.
We lay out a multiple, interacting levels of cognitive systems (MILCS) framework to account for the cognitive capacities of individuals and the groups to which they belong. The goal of MILCS is to explain the kinds of cognitive processes typically studied by cognitive scientists, such as perception, attention, memory, categorization, decision-making, problem solving, judgment, and flexible behavior. Two such systems are considered in some detail—lateral inhibition within a network for selecting the most attractive option from a candidate set and a diffusion process for accumulating evidence to reach a rapid and accurate decision. These system descriptions are aptly applied at multiple levels, including within and across people. These systems provide accounts that unify cognitive processes across multiple levels, can be expressed in a common vocabulary provided by network science, are inductively powerful yet appropriately constrained, and are applicable to a large number of superficially diverse cognitive systems. Given group identification processes, cognitively resourceful people will frequently form groups that effectively employ cognitive systems at higher levels than the individual. The impressive cognitive capacities of individual people do not eliminate the need to talk about group cognition. Instead, smart people can provide the interacting parts for smart groups
The very expertise with which psychologists wield their tools for achieving laboratory control may have had the unwelcome effect of blinding psychologists to the possibilities of discovering principles of behavior without conducting experiments. When creatively interrogated, a diverse range of large, real-world data sets provides powerful diagnostic tools for revealing principles of human judgment, perception, categorization, decision-making, language use, inference, problem solving, and representation. Examples of these data sets include patterns of website links, dictionaries, logs of group interactions, collections of images and image tags, text corpora, history of financial transactions, trends in twitter tag usage and propagation, patents, consumer product sales, performance in high-stakes sporting events, dialect maps, and scientific citations. The goal of this issue is to present some exemplary case studies of mining naturally existing data sets to reveal important principles and phenomena in cognitive science, and to discuss some of the underlying issues involved with conducting traditional experiments, analyses of naturally occurring data, computational modeling, and the synthesis of all three methods.This article serves as the introduction to a TopiCS topic with the same name. The rest of the downloadable papers in this Topic are:
Moat, H. S., Olivola, C. Y., Chater, N., & Preis, T. (2016). Searching choices: Quantifying decision making processes using search engine data. Topics in Cognitive Science, 8, 685–696. doi: 10.1111/tops.12207.
Below is an index of supplemental videos for the manuscript:
Lara-Dammer, F., Hofstadter, D. R., & Goldstone, R. L. (under review).An Integrated Computational Model of Perception and Scientific Discovery in a Very Simple World, Aiming at Psychological Realism
Free Space in a Circle (Tricycle)
Boyle’s Law Sophisticated A (Tricycle)
Boyle’s Law Sophisticated B (Tricycle)
Boyle’s Law Sophisticated C (Tricycle)
Non-ideal Gas A (non-success, Tricycle)
Non-ideal Gas B (Tricycle)
Non-ideal Gas C (Tricycle)
Understanding Noise (Tricycle)
Failed Discovery (Tricycle)
Thinking in Groups A (Tricycle)
Thinking in Groups B (Tricycle)
Learning abstract concepts through concrete examples may promote learning at the cost of inhibiting transfer. The present study investigated one approach to solving this problem: systematically varying superficial features of the examples. Participants learned to solve problems involving a mathematical concept by studying either superficially similar or varied examples. In Experiment 1, less knowledgeable participants learned better from similar examples,while more knowledgeable participants learned better from varied examples. In Experiment 2, prior to learning how to solve the problems, some participants received a pretraining aimed at increasing attention to the structural relations underlying the target concept. These participants, like the more knowledgeable participants in Experiment 1, learned better from varied examples. Thus, the utility of varied examples depends on prior knowledge and, in particular, ability to attend to relevant structure. Increasing this ability can prepare learners to learn more effectively from varied examples.
We consider a situation in which individuals search for accurate decisions without direct feedback on their accuracy, but with information about the decisions made by peers in their group. The “wisdom of crowds” hypothesis states that the average judgment of many individuals can give a good estimate of, for example, the outcomes of sporting events and the answers to trivia questions. Two conditions for the application of wisdom of crowds are that estimates should be independent and unbiased. Here, we study how individuals integrate social information when answering trivia questions with answers that range between 0% and 100% (e.g., “What percentage of Americans are left-handed?”). We find that, consistent with the wisdom of crowds hypothesis, average performance improves with group size. However, individuals show a consistent bias to produce estimates that are insufficiently extreme. We find that social information provides significant, albeit small, improvement to group performance. Outliers with answers far from the correct answer move toward the position of the group mean. Given that these outliers also tend to be nearer to 50% than do the answers of other group members, this move creates group polarization away from 50%. By looking at individual performance over different questions we find that some people are more likely to be affected by social influence than others. There is also evidence that people differ in their competence in answering questions, but lack of competence is not significantly correlated with willingness to change guesses. We develop a mathematical model based on these results that postulates a cognitive process in which people first decide whether to take into account peer guesses, and if so, to move in the direction of these guesses. The size of the move is proportional to the distance between their own guess and the average guess of the group. This model closely approximates the distribution of guess movements and shows how outlying incorrect opinions can be systematically removed from a group resulting, in some situations, in improved group performance. However, improvement is only predicted for cases in which the initial guesses of individuals in the group are biased.
de Leeuw, J. R., & Goldstone, R. L. (2015). Memory constraints affect statistical learning; statistical learning affects memory constraints. Proceedings of the Thirty-Seventh Annual Conference of the Cognitive Science Society. (pp. 530-535). Pasadena, CA: Cognitive Science Society.
We present evidence that successful chunk formation during a statistical learning task depends on how well the perceiver is able to parse the information that is presented between successive presentations of the to-be-learned chunk. First, we show that learners acquire a chunk better when the surrounding information is also chunk-able in a visual statistical learning task. We tested three process models of chunk formation, TRACX, PARSER, and MDLChunker, on our two different experimental conditions, and found that only PARSER and MDLChunker matched the observed result. These two models share the common principle of a memory capacity that is expanded as a result of learning. Though implemented in very different ways, both models effectively remember more individual items (the atomic components of a sequence) as additional chunks are formed. The ability to remember more information directly impacts learning in the models, suggesting that there is a positive-feedback loop in chunk learning.