ABSTRACT. To engage in conceptual change and learning, children revise beliefs about the world based on various forms of evidence. While descriptions of belief revision are ubiquitous in cognitive development, the mechanisms of children’s belief revision are largely unexplained. The present symposium explores the processes by which children revise their beliefs.

The first talk examines how different types of evidence influence the way children engage in information seeking behaviors. Inconsistent or ambiguous evidence motivates information seeking, while consistent evidence does not. Only ambiguous evidence promoted change in children’s beliefs. The second talk considers the role of the children’s self-generated explanations and the diversity of counterevidence in determining when children will change an uncertain belief. When shown counterevidence that a guess between two alternatives is incorrect, children shown diverse evidence to the contrary were likely to change their belief. When shown sparse evidence, children who explained their initial guess were more likely to revise their belief than children. These authors posit that children treat verbal information as critical for the malleability of belief.

The third talk examines a framework for thinking about small-scale conceptual change that integrates a well-studied neurocognitive mechanism (prediction-outcome mismatch signaling) with a computational framework that relies on Bayesian inference. The goal is to describe a neural account of how children use unmet expectations in order to gain information about the strength of a prediction. The fourth talk considers the kinds of probabilistic models young children learn. The author rejects the hypothesis that young children update beliefs about all sorts of probabilities given experience that is really (at least from an adult perspective) only informative about a single conditional (i.e., they try to learn a generative model when they should be learning a discriminative one). Instead, children’s belief updating is sensitive to task structure, and they may have the most success in learning when focused on simple discriminations.

The final talk considers an educational implication of belief revision in children – how children learn about evolutionary concepts. The authors presented children and adults a short tutorial designed to teach evolution by natural selection. Seven- to 12-year-olds could easily learn these concepts – no different from adult participants. But these age groups continued to give explanations of adaptation based on non-evolutionary mechanisms. These results suggest that older elementary-school-aged children can be taught evolutionary concepts, but learning such concepts does not lead to the revision of non-evolutionary views of biological adaptation.

Talk Titles, Authors and Affiliations

1) The development of reasoning about evidence Justin T.A. Busch and Cristine H. Legare, University of Texas at Austin

2) Evidence diversity and explanation influence how children resolve counterevidence David M. Sobel and Deanna Macris, Brown University

3) Developmental changes in the ability to change your mind: A neurobiological framework Mark A. Sabbagh and Kirsten Quistberg, Queens University, Ontario

4) Which beliefs do children revise with evidence? Learning discriminative and generative models Charles Kalish, University of Wisconsin

5) Children can learn evolutionary explanations for biological adaptation as well as adults can Andrew Shtulman, Cara Neal, and Gabrielle Lindquist, Occidental College

ABSTRACT. Making inductive predictions based on existing knowledge or evidence has long been recognized as a core component of human reasoning. The past few years has seen a marked expansion in research on induction. This has been accompanied by a diversification in the research questions studied and the methods used to measure and model inductive processes. This double-symposium brings together leading researchers from around the world to showcase these new frontiers of inductive research. One theme that runs through many of the talks is that methods and modelling approaches that have proven useful in other areas of cognitive science are leading to new insights into the way that induction works. Travers and Feeney, for example, show how tracking people’s movements while making inductive judgments can reveal the types of knowledge that drive those judgments. Several papers (Rotello, Heit & Kelly; Trippas; Stephens, Hayes & Dunn) highlight how signal detection approaches can be used to enrich our understanding of inductive reasoning. Hayes, Hawkins and Heit examine how evidence accumulation models, which have been used extensively in research on memory and decision-making, can be used to understand how inductive judgments are made in real time. Miser and Sloutsky show how measures of attention and memory processes can inform our understanding of inductive development. A second theme in this symposium is the increasing breadth of topics considered by induction researchers. Papers in the symposium will take us into new research territory such as the way that collaborating groups (rather than individuals) make inductive inferences (Heit, Bergmann, Bhat & Dale) and how the development of active enquiry strategies shapes children’s inferences (Rhodes, Kachergis & Gureckis). Another important new trend is the recognition that inductive inferences are influenced by people’s beliefs about where their data comes from, i.e. their assumptions about data sampling. The crucial role played by these assumptions in inductive inference will be highlighted by Voorspoels, Navarro, Perfors, Ransom & Storms. A final theme is the breaking down of traditional boundaries between research on induction and research on other kinds of reasoning. In this vein, Vasilyeva and Coley examine how inductive predictions relate to explanations of observed phenomena (abduction). Trippas examines the role of background knowledge in deduction. Stephens et al. take a fresh look at the question of whether a single or multiple latent processes are required to explain inductive and deductive reasoning.

TITLES AND LIST OF SPEAKERS New Approaches and Methods in Inductive Reasoning

Session 1 

1. Is there evidence for two processes in reasoning? A state-trace approach. Rachel Stephens. University of New South Wales, Australia Brett K. Hayes University of New South Wales, Australia John C. Dunn University of Adelaide

2. Do modals identify better models? A comparison of signal detection and probabilistic models of inductive reasoning. Caren M. Rotello University of Massachusetts, Amherst Evan Heit University of California, Merced Laura J. Kelly University of California, Merced

3. The SDT model of belief bias: Testing the equal variance assumption. Dries Trippas Max Planck Institute for Human Development

4. Negative observations in category-based induction: non-monotonic generalization and sampling assumptions. Wouter Voorspoels University of Leuven Daniel J. Navarro University of New South Wales, Australia Amy Perfors University of Adelaide Keith Ransom University of Adelaide Gert Storms University of Leuven

5. A bibliometric approach to studying group reasoning. Evan Heit University of California, Merced Till Bergmann University of California, Merced Harish S. Bhat University of California, Merced Rick Dale University of California, Merced

Session 2

1. Evidence accumulation models of inductive reasoning. Brett K. Hayes University of New South Wales, Australia Guy Hawkins University of Amsterdam Evan Heit University of California, Merced

2. What mouse tracking reveals about the roles of visual similarity and conceptual knowledge in inductive reasoning. Eoin Travers University of London Aidan Feeney Queen’s University Belfast

3. Explanation as a mechanism of selective inductive inference. Nadya Vasilyeva University of California, Berkeley John Coley Northeastern University

4. The implications of category knowledge for inductive reasoning across development. Marjorie Rhodes New York University George Kachergis New York University Todd Gureckis New York University

5. Attention, memory and the development of inductive generalization. Tracey Miser The Ohio State University Vladimir Sloutsky The Ohio State University

ABSTRACT. People are regularly presented with different types of evidence in order to persuade them towards a course of action such as buying a particular product or making a difficult choice. The effects of different kinds of evidence (e.g., statistical, causal, anecdotal) on decision making remain unclear. Four experiments are reported that examine the effects of different types of evidence on health related decision making. In Experiment 1 participants were presented with a health related scenario which contained either statistical or causal evidence for a particular course of action (e.g., for a particular drug to cure an illness). Experiments 2 and 3 investigated the impact of strong versus weak statistical and causal evidence on decision making while Experiment 4 presented different types of evidence for different choices. Overall, the results indicated that participants were somewhat sensitive to the different types and strengths of evidence as demonstrated by the reasons they gave for the choices they made. However, the effects of the different types of evidence on their choices were not clear cut suggesting other factors played a role too. The implications of the findings for theories of decision making are considered.

ABSTRACT. Children discriminate between reliable and unreliable sources from as young as 3 years (e.g. Koenig, Clément & Harris, 2004). However, it is unclear whether they are discriminating between reliable and unreliable informants because they have some epistemic awareness regarding the knowledge of their informants, or for some other reason. The current experiment manipulates source reliability in a more naturalistic way (science teacher vs. nursery child), and looks at the effects of age (6-7, 8-9, 10-11 years old) and schooling (two schools, one arts and the other science focused) on children’s’ predictions about the distance travelled on an inclined plane, where height, surface friction, starting point and weight can be changed. Children are either told unintuitive but true information regarding the non-effect of weight, coming from differentially reliable sources, or no information. Children in the high reliable condition were more likely to change their predictions regarding the effect of weight following information from the more reliable source, but only for the science focused school. It is concluded that schooling may mediate the interaction between source reliability and causal understanding.

ABSTRACT. Previous theoretical and empirical work suggests the possibility for observing action-based fluency effects on causal inference: The body specificity hypothesis (Casasanto, 2009) predicts an association between positive conceptual information and the side of space associated with the dominant hand. The causal force hypothesis (Wolff et al., 2014) states that individuals understand causation via their experience of haptic physical forces. Because fluency can induce intuitive versus analytical processing (Alter & Oppenheimer, 2009), we investigated whether embodied fluency affects causal learning. Right-handed participants learned about two possible causes of plant blooming in a trial-by-trial learning task. One of the possible causes was unrelated to plant blooming, and the other was weakly associated with it. We manipulated the spatial location of the causes (left, right) and the hand participants used to make responses (left, right). Consistent with expected fluency effects, when using their right hand to respond, participants judged the cause appearing on the right as more causal than that on the left, and they were more accurate in judging causes appearing on the left. RT and eye tracking data suggest that the effects observed in causal judgment were driven by changes in fluency rather than changes in the overt allocation of attention.

ABSTRACT. Some researchers have argued that mass perception, causal ascriptions, and predictions in simple billiard ball interactions can be modeled as inductive Bayesian inference over a (noisy) Newtonian representation of the world. However, there are phenomena, such as the asymmetrical ascription of forces to colliding objects, that are conceptually incompatible with the symmetry of Newtonian physics. We propose that human inference in physical scenarios operates over a pre-Newtonian physical representation that is based on impetus intuitions. Impetus theories assume that object movements are caused by an internal force, impetus, that is transferred and reflected when objects collide with each other. Moreover, impetus is inherently asymmetric. We present a mathematical model that implements a version of impetus theory and show that the theory is well suited to model perceived causal asymmetry. Moreover, the theory can also explain phenomena that so far have been presented as unique evidence for (noisy) Newtonian representations.

ABSTRACT. The aim of this research was to examine how much blame people attribute to different types of contributing events following a negative outcome. Specifically we explored the blame apportioned to events that caused a negative outcome and compared it to the blame apportioned to events that enabled a negative outcome. We conducted two experiments to examine the impact of event causality and event controllability on blame attribution. Participants were presented with a series of scenarios, each of which consisted of two events leading to a negative outcome, and were asked to rate how much blame they would attribute to each event. Results suggest that causality is more important than controllability in blame assignment, in that people attribute more blame to events that cause rather than enable a negative outcome. However, the relative controllability of the event does have an impact on this pattern, particularly for the enabling event, even after other factors such as social acceptability are controlled for. The implications of the findings are discussed in the context of previous research on blame attribution and for theories of blame assignment.

ABSTRACT. In this study we aimed to assess the impact of mental state information in blame attributions when people evaluate group structures. Specifically, we sought to establish how mental state information (Shaver, 1970; Weiner, 1985; Alicke,1992; 2000; 2011; Malle, Gulielmo & Monroe, 2014) interacts with information about causal structures (Latané & Darley, 1968; Chokler & Halpern, 2004) to guide the attribution process depending on the agents’ intentions and specific relationships between their actions. Based on the recently proposed criticality-pivotality model (Lagnado & Gerstenberg, 2013; Lagnado, Gerstenberg & Zultan, 2013) we explore how the two kinds of key information (structural and epistemic) allow an observer to gain a more complete view of the situation when engaging in blame attribution. This possibility had not yet been explored in the literature. We conducted two experiments to explore the possible relationship between causal information (pivotality and criticality) and epistemic state information (intention and foreknowledge) on blame attribution. A 2X2 ANOVA with epistemic state (present vs absent) and causal structure (high vs low) as factors were run for each experiment. For experiment 1 results indicate the main effect of epistemic state (1,95) =88.278 p< 0.01 and causal structure F(1,95) =16.633 p< 0,01 and, interestingly, an absence of interaction between the two factors F(1,95)=0.002 p=0.9. This shows that participants responded to the causal structure of the situation and assigned more blame when pivotality and criticality were higher. Also, information about epistemic states leads to a higher blame attribution when present compared to absent. The lack of interaction shows that both kinds of information were used independently, that is, in an additive fashion, for a higher blame attribution when causal structure were high and epistemic states were present. Results of experiment 2 replicate the findings of experiment 1, with a main effect of epistemic state F(1,95) =16.028 p< 0.01, causal structure F(1,95) =68.92 p< 0,01 and a lack of interaction between the two F(1,95)=3.590 p=0.6. The additive relationship indicates that when both types of information are available, blame attributions are higher, and it shows that both kinds of information can be used independently of the other or in a combined fashion. This link had not been established in the literature before and opens the possibility to explore the limits of the pivotality-criticality model as a general model of responsibility and blame attribution. References Alicke, M. & Davis, T.L. (1990). Capacity responsibility in social evaluation. Personality and Social Psychology Bulletin, 16, 465-474. Alicke, M. (1992). Culpable causation. Journal of personality and social psychology. 3, 368-378 Alicke, M. (2000). Culpable control and the psychology of blame. Psychological bulletin vol.126 (4), 556-574 Chockler, H. & Halpern, J. Y. (2004). Responsibility and blame: A structural-model approach. Journal of Artificial Intelligence Research, 22 (1), 93-115. Lagnado, D & Gerstenberg T. (2013). A difference-making framework for intuitive judgments of responsibility. Unpublished dissertation Lagnado, D., Gerstenberg, T. & Zultan R. (2013). Causal responsibility and counterfactuals. Cognitive science. 37, 1036-1073 Latané, B. & Darley, J. (1968). Group inhibition of bystander intervention in emergencies. Journal of personality and social psychology. 10(3) 215-221 Malle, B., Guglielmo, S. & Monroe, E. (2014). A theory of blame. Psychological inquiry: an international journal for the advancement of psychological theory. 25 (2), 147-186 Shaver, K. (1970). Defensive attribution: effects of severity and relevance on the responsibility assigned for an accident. Journal of personality and social psychology. 2, 101-113

ABSTRACT. A complex social system often has more norms than an individual can remember or invoke. How do individuals represent and navigate the rules that govern their social behavior? To answer this question, we track mental representations of social norms to learn how individuals perceive, explore, and utilize norms in a real world social system: the English-language Wikipedia. Wikipedia, one of the world’s largest online social systems, is a community-managed knowledge commons. Norms play a crucial role in its governance, by forming collective expectations for how to edit articles and resolve conflicts. Editors reference these norms in discussions and arguments that occur on an article’s “talk” page. We use Wikipedia’s detailed log of talk page comments to record the norms that individual users cite on a subset of Wikipedia’s articles. We report results on the inferred structure of user mental representations of the system’s social norms, the impact of environment (article type) on norm invocation, and the portability of norms by users across environments.

ABSTRACT. Knowledge of intention and outcome are integral to making judgments of responsibility, blame, and causality. Yet, little is known about the effect of conflicting intentions and outcomes on these judgments. In a series of four experiments, we combine good and bad intentions with positive and negative outcomes and present these through relatable, everyday moral scenarios. Results demonstrate an asymmetry in responsibility and causality judgments for the two incongruent conditions: well-intended agents are regarded as more morally and causally responsible for negative outcomes than agents with bad intentions are for positive outcomes. This novel effect of an intention-outcome asymmetry identifies an unexplored aspect of moral judgments and is partially explained by the perceived effort of the acting moral agent.

ABSTRACT. Moral psychology is an area that has recently seen a dramatic growth in research, and a number of prominent theories have been proposed in order to account for relevant behavioural phenomena. Many of these theories, however, have been relatively underspecified and are not typically discussed in formal terms, which has led to some confusion within the literature about their exact predictions – especially about the mechanisms of some of their hypothesised systems. More rigorously defined theories may thus benefit the field’s development. Based upon this suggestion, a computational model of moral decision making was implemented which employed influence diagrams to capture behaviour. The output of the model was compared to findings from multiple behavioural experiments, and was found to have achieved reasonably good predictive success. The implications of this for moral psychology will be discussed; a probabilistic model is compatible with positions such as a dual process theory of moral decision making, for instance, and may allow for one possible method of formalising such a framework.

ABSTRACT. We have designed and implemented an experiment so as to better understand the beliefs of the American population regarding the fairness of wealth inequality. More specifically, our experiment investigates the role that scientific evidence about the detrimental socioeconomic consequences of poverty plays in shaping people’s beliefs about economic justice, in comparison with the role played by empathetic engagement with personal stories. We ran an in-between participants experiment with two main conditions: scientific information versus empathetic engagement. Our findings have surprisingly revealed that neither scientific evidence about the negative effects of poverty on development, nor the triggering of empathetic feelings for those in extreme poverty, have any effects on people’s beliefs about distributive justice. However, motivated cognition appears to play a prominent role in the determination of beliefs about economic justice. We found that participants’ scores in the Economic System Justification (ESJ) scale predicted their desired levels of wealth distribution across the population under the empathy experimental condition. It was only under the condition in which we elicited empathy that ESJ predicted participants' desired levels of wealth distribution. When primed to feel empathy, the higher participants scored in ESJ, the more income they were willing to distribute to the top 40%, and the less income they were willing to distribute to the bottom 40%.

ABSTRACT. The Bayesian approach to reasoning and argumentation is currently very popular in psychology as well as in philosophy. But how good is the approach actually? What are its scope and limits? And how does it relate to other approaches to scientific and ordinary argumentation? These are some of the questions that we will explore in this symposium.

Abstracts of the talks:

Bayesian Argumentation Meets the Real World Ulrike Hahn

Bayesian models have had considerable success in elucidating fundamental intuitions about evidential strength within philosophy. At the same time, they have been applied to argumentation, in particular, they have been used to provide a formal account of so-called fallacies of argumentation. Ultimately, a complete theory of argumentation needs to have something to say about argument quality in any real world context that might arise. The talk evaluates Bayesian Argumentation with respect to that potential.

A Unified Theory of Bayesian Argumentation Stephan Hartmann

This talk motivates and sketches a unified Bayesian theory of argumentation. According to this theory, an agent has prior beliefs about some propositions A, B,… These beliefs are represented by a Bayesian network and a probability distribution P defined over it. The agent then learns some propositions (the premises of the argument) form some information source, which aims at convincing her from the truth of some other proposition (the conclusion of the argument). To accommodate the new information, the agent assumes that the “causal” structure of the propositions she entertains (i.e. the Bayesian network) does not change and that the new probability distribution P’ follows from minimizing some “distance measure” (such as the Kullback-Leibler divergence) between P’ and P. I will show how this works for modus ponens, modus tollens, affirming the consequent, and denying the antecedent. I then generalize the proposed theory to partially reliable information sources and show how disabling conditions (e.g. known exceptions to the conditional) can be taken into account. This suggests a unified theory of Bayesian argumentation according to which argumentation is learning premises and updating one’s beliefs accordingly, making sure that the “causal” structure of one’s beliefs is not changed. Finally, I explore alternative distance measures and suggest empirical tests that distinguish between them.

Causal Networks in Evidential Reasoning David Lagnado

How do people reason in the face of complex and contradictory evidence? Focusing on investigative and legal contexts, we present an idiom-based approach to evidential reasoning, in which people combine and reuse causal schemas to capture large bodies of interrelated evidence and hypotheses. We examine both the normative and descriptive status of this framework, illustrating with real legal cases and empirical studies. We also argue that it is qualitative casual reasoning, rather than fully Bayesian computation, that lies at the heart of human evidential reasoning.

Inference to the Best Explanation and the Kinematics of Belief Jonah Schupbach

In its simplest formulation, Inference to the Best Explanation (IBE) states that we ought to infer hypotheses that provide the best available, competing explanations of our evidence. As an unqualified inference rule, IBE seems to offer up a normative principle for how categorical beliefs should change over time—specifically, when we should adopt new categorical beliefs in hypotheses. But what should we make of IBE if we are interested in giving a kinematics of belief in terms of quantitative degrees of belief, or credences? Must IBE simply be dropped in favor of a credence-updating rule? Can IBE and such updating rules be part of a larger, coherent account of rational belief change? Or perhaps can these rules be developed (or reinterpreted) to preserve and validate IBE’s central insight that explanatory judgments carry legitimate normative weight? In this presentation, I will confront these questions, exploring IBE's potential bearing on normative and descriptive accounts of updating.

ABSTRACT. Making inductive predictions based on existing knowledge or evidence has long been recognized as a core component of human reasoning. The past few years has seen a marked expansion in research on induction. This has been accompanied by a diversification in the research questions studied and the methods used to measure and model inductive processes. This double-symposium brings together leading researchers from around the world to showcase these new frontiers of inductive research. One theme that runs through many of the talks is that methods and modelling approaches that have proven useful in other areas of cognitive science are leading to new insights into the way that induction works. Travers and Feeney, for example, show how tracking people’s movements while making inductive judgments can reveal the types of knowledge that drive those judgments. Several papers (Rotello, Heit & Kelly; Trippas; Stephens, Hayes & Dunn) highlight how signal detection approaches can be used to enrich our understanding of inductive reasoning. Hayes, Hawkins and Heit examine how evidence accumulation models, which have been used extensively in research on memory and decision-making, can be used to understand how inductive judgments are made in real time. Miser and Sloutsky show how measures of attention and memory processes can inform our understanding of inductive development. A second theme in this symposium is the increasing breadth of topics considered by induction researchers. Papers in the symposium will take us into new research territory such as the way that collaborating groups (rather than individuals) make inductive inferences (Heit, Bergmann, Bhat & Dale) and how the development of active enquiry strategies shapes children’s inferences (Rhodes, Kachergis & Gureckis). Another important new trend is the recognition that inductive inferences are influenced by people’s beliefs about where their data comes from, i.e. their assumptions about data sampling. The crucial role played by these assumptions in inductive inference will be highlighted by Voorspoels, Navarro, Perfors, Ransom & Storms. A final theme is the breaking down of traditional boundaries between research on induction and research on other kinds of reasoning. In this vein, Vasilyeva and Coley examine how inductive predictions relate to explanations of observed phenomena (abduction). Trippas examines the role of background knowledge in deduction. Stephens et al. take a fresh look at the question of whether a single or multiple latent processes are required to explain inductive and deductive reasoning.

TITLES AND LIST OF SPEAKERS New Approaches and Methods in Inductive Reasoning

Session 1 1.Is there evidence for two processes in reasoning? A state-trace approach. Rachel Stephens. University of New South Wales, Australia Brett K. Hayes University of New South Wales, Australia John C. Dunn University of Adelaide

2.Do modals identify better models? A comparison of signal detection and probabilistic models of inductive reasoning. Caren M. Rotello University of Massachusetts, Amherst Evan Heit University of California, Merced Laura J. Kelly University of California, Merced

3.The SDT model of belief bias: Testing the equal variance assumption. Dries Trippas Max Planck Institute for Human Development 4.Negative observations in category-based induction: non-monotonic generalization and sampling assumptions. Wouter Voorspoels University of Leuven Daniel J. Navarro University of New South Wales, Australia Amy Perfors University of Adelaide Keith Ransom University of Adelaide Gert Storms University of Leuven

5.A bibliometric approach to studying group reasoning. Evan Heit University of California, Merced Till Bergmann University of California, Merced Harish S. Bhat University of California, Merced Rick Dale University of California, Merced

Session 2

1.Evidence accumulation models of inductive reasoning. Brett K. Hayes University of New South Wales, Australia Guy Hawkins University of Amsterdam Evan Heit University of California, Merced

2.What mouse tracking reveals about the roles of visual similarity and conceptual knowledge in inductive reasoning. Eoin Travers University of London Aidan Feeney Queen’s University Belfast

3. Explanation as a mechanism of selective inductive inference. Nadya Vasilyeva University of California, Berkeley John Coley Northeastern University

4.The implications of category knowledge for inductive reasoning across development. Marjorie Rhodes New York University George Kachergis New York University Todd Gureckis New York University

5.Attention, memory and the development of inductive generalization. Tracey Miser The Ohio State University Vladimir Sloutsky The Ohio State University

ABSTRACT. Surprise is a ubiquitous phenomenon that has been implicated in many areas of cognition (from learning, to hindsight bias, to counterfactual thinking).

Folk theories of surprise see it as a response to low probability outcomes, and probabilistic theories often adopt a similar view. However, the evidence for the most part does not support such probabilistic accounts, and their generatively as theories seems weak. Foster & Keane (2015, Cognitive Psychology, 81, 74-116) have advanced an explanation-based theory of surprise; that sees surprise as a function of the amount of cognitive, explanatory work that needs to be done to “make sense” of an event or outcome. This account has been used to make a number of predictions confirmed in several experiments, which show that explanatory aspects of surprise are much better predictors of people’s surprise judgements than probabilistic aspects.

This paper presents an explanation-based model of these findings, that builds a directed graph of explanations to link the setting and outcome of a given scenario, and uses this graph to predict surprise ratings. Simulations are reported that demonstrate how the model’s performance corresponds closely to human ratings for different surprising scenarios. Finally, the relationship of this model to probabilistic accounts is discussed.

ABSTRACT. It has been proposed that the domain of the explanandum (e.g., artifact, animate being) is the main factor that determines which type of explanation people prefer, mechanistic or teleological. In this study, we proposed that the explanandum’s thematic relation, which is mostly determined by the predicate, i.e., action verb or state verb, can be crucial for the explanation type preference. To compare the two theories, we asked participants to generate an explanation after they read a sentence describing either the action of an actor/artifact or the state of an actor/artifact. The domain of the objects to be explained and the properties to be explained were factorially combined. In addition, participants generated the explanation either without time pressure or under time pressure, to figure out whether there is a default preferred explanation type. The patterns of generated explanations were almost identical for the two conditions. Most of the explanations generated for the state verbs were mechanistic explanations, whereas the two types of explanation were about equally generated for the action verbs. The effect of the domain was not significant. We will try to figure out the reasons why verb types affect the explanation type preference.

ABSTRACT. We often explain events by positing causes of those events. For instance, given a patient’s symptoms, a doctor must decide which disease has caused those symptoms. However it is also important for explanations to describe the mechanisms by which a cause leads to an effect. That is, how does a disease cause a set of symptoms? Understanding the mechanism underlying a cause enables us to better understand the causal system, and potentially leads to more accurate inferences.

Previous studies have shown that people performing diagnostic inference have a preference for constrained hypotheses. Typically, there is a preference for hypotheses that invoke fewer causes and have a narrow scope (do not explain unobserved effects). We tested whether the inclusion of mechanistic information in an explanation affects explanatory preference.

We found that when performing a causal inference task (determining the causes of an event), participants preferred explanations with fewer causes, consistent with prior findings. However when explanations contained mechanistic information, participants preferred explanations that appealed to multiple causes. This effect is observed in spite of the fact that both conditions contain the same probabilistic information regarding base rates and strengths of various causes.

ABSTRACT. A number of philosophers argue for the value of abstraction or idealization in explanation. According to these prescriptive theories, an explanation becomes superior when it leaves out details that make no difference to the occurrence of the event one is trying to explain (the explanandum). Abstract or idealized explanations are not frugal placeholders for improved, detailed future explanations, but are more valuable than their concrete counterparts because they highlight the factors that do the causal work, the factors without which the explanandum would not occur. We present several experiments that test whether people follow this prescription, i.e. whether people prefer abstract difference makers rather than concrete details and, furthermore, whether they prefer explanations that omit descriptively accurate but causally irrelevant information. Contrary to the prescription, our findings show that people have a general preference for concreteness and detail. Participants rated explanations with concrete details higher than their abstract counterparts and in many cases they did not penalize the presence of causally irrelevant details. Nevertheless, causality still constrained participants’ preferences: They downgraded concrete explanations that did not communicate the critical causal properties.

ABSTRACT. There is still little consensus about the nature of logical reasoning, and equally important, about how it develops. To address this question, we looked at the early origins of logical reasoning in preschool children. We examined the contribution of two factors to the reasoning ability of very young children: 1) inhibitory capacity; 2) the capacity to generate alternative ideas. In Study 1, we assessed idea generation, inhibition and logical reasoning. Logical reasoning was measured using knowledge-based premises such as “all dogs have legs”, and two different inferences: Modus Ponens and Affirmation of the Consequent. Results revealed that correctly reasoning with both inferences is not related to the measure of inhibition, but is rather related to the capacity to generate alternative ideas. Critically, this was estimated by the number of categories covered by the ideas given. In Study 2, we either gave the generation task or the inhibition task (as control) before the logical reasoning measure. Results showed that receiving the generation task beforehand significantly improved logical reasoning compared to the inhibition task given beforehand. Overall, these results provide evidence for the greater importance of idea generation in the early development of logical reasoning.

ABSTRACT. In Stanovichʼs tripartite model of rationality, thinking dispositions are thought to be indexes of reflective mind that regulate functioning of algorithmic mind (correlated with intelligence). The aim of the study was to examine how specific abilities (such as scientific reasoning) interact with thinking dispositions and fluid intelligence in their effect on rationality, which was conceptualized as having normative responses on cognitive biases task and restraining from holding epistemically suspect beliefs. 356 young adults participated in the study. Several measures of cognitive biases (CB), thinking dispositions (TD), intelligence (IQ), scientific reasoning (SR) and epistemically suspect beliefs (ESB) were administered. In hierarchical regression analysis Intelligence and scientific reasoning were significant independent predictors of CB responses, as were IQ x TD and IQ x SR interactions. IQ and SR were significant independent predictors also for the endorsement of ESB, but only IQ x TD interaction was a significant predictor for the endorsement of ESB. The results are discussed within the tripartite model of rationality and we further advocate giving special focus on acquiring specific skills, such as scientific reasoning, for promoting rational thinking characterized by resistance to cognitive biases and unsound beliefs.

ABSTRACT. Scientific evidence plays an important role in a range of decisions faced by nonscientists, yet little is known about what skills are needed in order to read and evaluate scientific evidence, and to what extent nonscientists possess these skills. We draw on research in cognitive developmental psychology, public understanding of science, and behavioral decision research to identify and measure the scientific reasoning skills that are needed to evaluate the quality of scientific evidence. We present three studies detailing the development and validation of an individual difference measure of these skills, the Scientific Reasoning Scale (SRS). Our results indicate that the SRS is internally consistent and distinct from existing measures of scientific literacy. Participants with higher SRS scores are more likely to hold beliefs consistent with the scientific consensus on contentious issues including climate change and vaccination, above and beyond political and religious beliefs, education, and scores on two existing measures of scientific literacy. Participants with higher SRS scores also displayed better performance on a task requiring them to analyze scientific information. Our results suggest that the SRS provides a theoretically informed contribution to decoding lay responses to scientific results and controversies.

ABSTRACT. Innovation is a social process. Individual innovators must both learn from, and differentiate themselves from, prior influences and their own contemporaries. To study this we require longitudinal data on both creative output and cultural input. In this investigation, we study the development of Charles Darwin's theory of natural selection through his reading and writing behavior from 1837 to 1860. While he reported his "Mathusian insight" in 1838 and wrote two outlines in 1842 and 1844, he hedged on publication of On the Origin of Species until 1859. This 21 year gap from insight to publication is known as "Darwin's delay". We use topic models trained over the full-text of books recorded in his reading notebooks to place his writings in a common semantic space. We find that the information-theoretic distance between his readings and writings increases over time, indicating growing differentiation from his influences. We also find that the model reflects Darwin's observation that Alfred Russell Wallace's contemporaneous theories were much closer to his earlier drafts than the published Origin, establishing priority. This multi-decade study of an exemplary scientist's behavior uses new, general methods to show how influence, differentiation, and maturation are part of the dynamics of innovation.

ABSTRACT. Whereas the relationship between numeracy and decision-making competence is well-documented, to date only a few studies investigated the potential link between mathematical anxiety (i.e., a feeling of anxiety and threat when performing numerical operations) and decision-making skills. In this talk we first present evidence from recent studies that show a link between math anxiety and decision competence in various contexts (e.g., performance on the Iowa Gambling Task, the ability to assess the potential risks of consuming genetically modified food, etc.) Then we present a series of studies that showed a link between math anxiety and performance on the cognitive reflection test. Indeed, this link was also present when the effects of numeracy and test anxiety were controlled for. Finally, we demonstrate that mathematical anxiety is related to people’s ability to interpret information about medical risks, as well as people’s confidence in these interpretations. Overall, these results show that the effect of mathematical anxiety extends beyond academic contexts, and that it makes an important contribution to explaining individual differences in decision making skills.

ABSTRACT. A theoretical link between reasoning and mathematical ability has been supported by some recent empirical evidence. We argue that some of this evidence is indirect and measure selection may have influenced this relationship. We report two studies in which mathematical ability was measured using a fluency and a calculation measure (Wood-cock Johnson-III, 2001) and reasoning ability was measured using an extended cognitive reflection test. In Study 1, reasoning was also measured using a belief bias conditional reasoning task, and in Study 2, we included Ravens Progressive Matrices and a maths anxiety measure. Results from 68 students (Study 1) and 80 students (Study 2) suggest that mathematical ability is predicted by performance on the cognitive reflection test but not conditional reasoning or Ravens Matrices performance when mathematical fluency and maths anxiety are taken into consideration. We discuss the implications of these findings for research on the link between mathematical ability and reasoning skills.

ABSTRACT. Successful social interaction involves coordinating thoughts and behaviour between people. But how such coordination achieved? If each person attempts to second-guess the thoughts and behaviour of the other, there is a danger of an infinite regress. A tries to infer what B will do; and knows that B will try to infer what A will do; so A needs to figure out what B thinks that A will do; but what A will do in turn depends on what B thinks A thinks that B will do, and so on, forever. We introduce a different approach: that people can reason jointly about what they would agree to think or do, were they able to negotiate. That is, they reason not about “What will you do?” and “What should I do?, but rather “What should we agree to do?” Where it is “obvious” what a resultant such negotiation would be, no actual communication is required: we can coordinate our thoughts and actions through a simulation of the bargaining process. Virtual bargaining provides a new foundation for understanding the reasoning that underpins social behaviour, including communication itself.

ABSTRACT. (POSTER) Based on a discrepancy in conditional reasoning performances between reasoning about possibilities and reasoning about truth values, Gauffroy & Barrouillet (2011) claimed that there are two kinds of propositional reasoning that do not require the same abilities. Reasoning about possibilities seems to be psychologically basic whereas reasoning about truth values involves meta-ability. However, this discrepancy between the two forms of reasoning could be specific to the suppositional and hypothetical nature of the “if… then…” connective. In this experiment, we compared the development of both kinds of reasoning when second, fourth and sixth graders are asked to understand disjunctions (e.g. “the circle is red or the star is yellow”). The results showed the predicted developmental lag between reasoning about possibilities and reasoning about truth value. In the former, all the age groups produced an inclusive interpretation of disjunctive statements, whereas when reasoning about truth values, second graders produced a conjunctive interpretation (“or” being interpreted as “and”), fourth graders mainly exhibited exclusive interpretations and sixth graders an inclusive interpretation.

ABSTRACT. -Poster-

We propose a conception of mind bounded by the qualitative constraint of relevance at conscious and unconscious levels. The core of this conception is an interpretative function, as adaptive characteristic of the human cognitive system. This perspective is supported by evidence from our recent research on insight problem solving, which we consider a privileged route to understanding what kind of special unconscious thought produces the solution. During incubation, in the absence of conscious control, relevance constraint allows multilayered thinking to discover a new interpretation of the data that finally offers an exit from the impasse. We speculate that the creative act of restructuring implies a form of high-level unconscious thought, the unconscious analytic thought.

ABSTRACT. When individuals judge alternative choices, presenting alternatives concurrently improves decision making outcomes. The joint decision making advantage has been proven in the Western world, yet generalizations for other cultures are missing. This paper explores the applicability of joint decision making for global public policy decisions in the intergenerational equity domain. Presenting the viewpoints of two generations with outcomes now or later concurrently worked towards intergenerationally equitable choices when surveying 223 Chinese individuals (Study 1) and 374 online recruited respondents (Study 2). Joint decision making is a powerful, previously untested means to overcome hyperbolic discounting biases in decisions on global common goods dilemmas. We also find policy bundling decreases social responsibility. The joint alternative presentation thus leads to future-orientation versus social responsibility trade-off predicaments in intergenerational decisions. Policy makers are advised to consider a multi-faceted decision schema and age-differentiated consortia may help implement intergenerational equity.

ABSTRACT. Common retirement investment advice encourages younger people to hold more risky assets and decrease portfolio riskiness as the retirement age approaches. In the present research we tested the effectiveness of this advice on retirement asset allocation decisions. In a between subject experiment, participants (N=456, age M=30, SD=4) were making retirement asset allocations with/without reading the advice. Additionally to making an allocation decision at this moment, participants were asked how they would invest at different ages, to assess their intentions to reallocate retirement savings. The participants who received the investment advice chose to invest slightly more into risky assets at their present age t(598)=-1.776, p=.038, d=0.15. Furthermore, they were more likely to decrease portfolio riskiness with age after receiving the advice b=0.754, χ2 (1)=14.463, p<.001, R2=.044. The degree of shifting to safer assets among those who chose decrease their exposure to risky assets with age did not differ between experimental groups t(251)=-0.191, p=n.s. The investment advice seems to both increase exposure to risky assets and promote intentions to reallocate into safer assets with the retirement age approaching. These effects, however, seem to be rather small. Alternative communication of the advice should be considered, if larger effects are to be expected.

This paper could be presented as a poster.

ABSTRACT. One challenge of promoting behavior change is that individuals must sacrifice the comfort and convenience of their usual habits. In three studies, we investigate the most effective ways to promote energy-efficient household behaviors. We find that individuals are more likely to adopt energy-efficient behaviors when financial consequences are presented on a monthly basis, and maintain status quo behaviors when the financial consequences are presented as daily or annually. A second study suggests that there may be an effect of cognitive fluency – individuals may be more amenable to energy efficiency in the context of their typical monthly energy bills. Results of a third study suggest that the timeframe of their typical household energy expenses provides context for how individuals respond to costs. Specifically, we find that individuals are most likely to engage in energy efficiency when given annual associated costs and typical annual household energy expenses. In general, we recommend that policy-makers present costs of energy inefficient decisions in the context of typical energy expenses for that same timeframe to increase fluency and maximize energy savings.

ABSTRACT. Dispositional theories held that people interprete interactions of two objects including an asymmetric impact of forces as causal (Mayrhofr & Waldmann, 2014; Talmy, 1988; White, 2009; Wolff, 2007). This intuitive conception of causality appears to emanate from specific bodily experiences aquired in and reinforced throughout development. This reinforcement occurs whenever bodily force is used to change something physically by direct contact. Two studies examine the developmental origins of causal thinking and intent to compare the adoption of an agent-patient relationship when judging physical and biological phenomena. Children (5-6-year-old, 7-8-year-old, 11-12-year-old) and adults judged a collision and a stinging event with a sentence verification task covering the central aspects of dispositional causality, e.g., assumption of asymmetric forces, agent-patient role distribution, antagonistic interaction, goal-directed production of effect. Adults additionally experienced time pressure. Results indicate that participants cross-domainly rely on a causal core concept when interpreting interactions between two objects. Moreover, the tendency to adopt a disposition of causality appears to increase with age, particularly noticeable in biology. Since the child-oriented setting of the present study could have primed the intuitive thinking, in a revised study, adults repeat the task in a scientific setting. So far, first results slightly support the prior findings.

ABSTRACT. Executive control modulates the expression of automatic stereotyping. We aimed at examining whether cognitive reflection – the ability to exert cognitive effort, suppress intuitive responses and engage in analytical thinking – also predicts stereotypic and prejudiced responses, and at identifying corresponding moderators of the effects. In two experiments with 220 undergraduates, we administered the Cognitive reflection test, the Stroop test, the Actively open-minded scale (AOT) and the Cultural cognition worldview scales, either before or after the critical experimental tasks. In Study I, we employed the Word evaluation tasks as an implicit attitude measure. In Study II, various scenarios were used as an explicit method of assessing expression of ethnic prejudice. Both executive control and cognitive reflection predicted a lower prevalence of stereotype- and prejudice-based responses, although not exclusively. Cognitive control reduced automatic bias only among people who prefer hierarchical values. Further, cognitive reflection attenuated stereotyping and expression of prejudice at low or moderate levels of AOT. Thus, we confirmed that cognitive reflection plays a crucial role in inhibition and expression of stereotypes and prejudice. Moreover, our findings corroborate the hypothesis of a nonlinear relationship between executive functioning and stereotyping, and highlight the need for promoting critical thinking to counter prejudice.

ABSTRACT. Nanoscience and HPV vaccination are controversial topics which have not yet been subjected to public debate in Slovakia. We were therefore interested whether exposure to balanced arguments, relative distance from opposing argument advocates, participants’ cultural worldview and topic familiarity determine evaluation of respective risks. In contrast to earlier findings (Kahan el al., 2009), nanoscience was viewed as a progressive technology, which was endorsed by those with an egalitarian worldview and feared by less liberal hierarchists. In the no-arguments condition, participants familiar with nanoscience considered it less risky. Moreover, risk perception of people with little initial knowledge was higher when their cultural values resembled those of the con-advocate and differed from those of the pro-advocate. Assessment of risks inherent to HPV vaccination was also influenced by both familiarity and cultural cognition. But this time, gender was crucially implicated. Overall, hierarchism amplified risk perception only among men and, conversely, familiarity reduced it only among women. In addition, attenuation of perceived risk due to topic familiarity occurred only among people whose worldview was much closer to the pro-advocate. Taken together, it seems that for novel domains, risk perception is influenced by a complex interplay of cultural cognition, familiarity and personal relevance.

ABSTRACT. There are several explanations for the relationship between inattention and CRT performance. In particular, working memory has been considered a strong predictor of CRT performance and essential to success on the task. This executive function has also recently been put forward as the core deficit in ADHD. The current study investigated the relationship between the ADHD symptom of inattention in a non-clinical population and performance on the CRT, taking into account gender too. Participants were 169 high school students (67% Male; mean age= 16.27; SD = 1.52). The Inattention subscale (AN- DSM-IV-TR ADHD, Conners 3, 2008) and the CRT-Long (Primi et al., 2015) were administered. A 2x2 ANOVA with inattention (low/high) and gender on cognitive reflection correct responses showed a significant effect of inattention (F(1,103)=3.68, p<.05, ηp2=.04). Students with lower inattention have a better performance on CRT. There was no effect of gender but a significant interaction between inattention and gender (F(1,103)=4.29, p<.05, ηp2=.04). Female students have a better performance than male when they have a low inattention. Overall, these results show that the symptom of inattention affects the success on the Cognitive Reflection Test, and that it could make an important contribution to explain gender differences in the CRT.

ABSTRACT. “Weak” teleosemantics (e.g. Dretske) holds that representation is the informational process between the producer and consumer where the producer is evolutionarily ‘drafted’ by the consumer in virtue of its descriptive function–the function of correlating with a state of affairs. On the other hand “Strong” teleosemantics (e.g. Millikan), in terms of the coevolutionary thesis between the producer and consumer, rejects that producers generate contents that are ‘pre’-anchored to the things in the world. The disjunction problem is the problem of explaining the capacity for misrepresentation–e.g. why the representational state of a frog when it flicks its tongue to catch a fly is about a fly rather than a B.B. or the retinal image? While the weak and strong versions agree that the evolutionary function of representation is the key to the solution, they disagree with respect to the nature of function. Dretske argues that the frog’s representation is not about the retinal image because the frog’s mechanism of learning to discriminate a fly from other things requires the frog’s producer systems (e.g. the gustatory system) to pick up the properties of a fly. This sort of solution by weak teleosemantics is problematic because it generates another problem, namely the binding problem. I analyze biological studies in animal communication and show that it is empirically more plausible that natural selection solves the binding problem by the coevolution between the producer and consumer. On the basis of this analysis, I explain how strong teleosemantics ‘resolves’ the disjunction problem.

ABSTRACT. Although DeepBlue's triumph in chess and AlphaGo's recent triumph in Go are milestones in artificial intelligence, it is unclear whether their algorithms resemble human thinking. Laboratory tasks that are common in decision-making psychology bring no relief, as they rarely involve long sequences of decisions with many options at each step. To investigate how people explore such decision trees, we turned a challenging variant of tic-tac-toe into 4 experimental paradigms: people play against each other, against AI agents of different strengths, choose between two moves, or evaluate positions.

We model human thinking with an AI-inspired heuristic search algorithm. The algorithm explores a decision tree using best-first search with selective pruning, guided by a heuristic function that evaluates boards with a weighted sum of features. To capture characteristic human mistakes, we incorporate three types of noise. We estimate model parameters for individual subjects, and show that this model predicts subjects' choices better than a dozen alternative models. Moreover, the model can generalize from subjects' choices during games to predict their choices in the 2AFC task, board evaluations, response times and eye movements. Finally, the parameters inferred for different subjects suggest that stronger players build larger decision trees and make less variable decisions.

ABSTRACT. Both first-order and second-order false belief mastery are viewed as important in acquisition of Theory of Mind, the capacity to ascribe mental states. Our logical analysis of second-order false-belief tasks shows that this sort of reasoning involves recursion. Language involves recursion as well; recursive possessive and complements clauses are clear examples.

The development of second-order social reasoning is complex and depends on both individual cognitive resources and immersion in a wide range of interactive contexts. But since the ‘usual’ interactive contexts do not make the same sense to children with Autism Spectrum Disorder (ASD), it has been proposed that they may use language as scaffolding in false-belief understanding.

We hypothesize that competency in linguistic recursion predicts second-order false belief mastery for children with ASD. In order to investigate our hypothesis, we intend to train children with ASD to better comprehend and produce recursive possessive and complement clauses. We have developed and validated a tool to measure the recursion competency in the Danish language, and we will apply this tool in a randomized controlled training study.

ABSTRACT. This symposium will take a retrospective and prospective look at Pearl’s (2000) causal Bayes net (CBN) framework.  CBN theories have helped clarify the computational problems involved in causal cognition as well as highlight potential ways the mind might solve those problems. Causal cognition requires learners represent their beliefs about the structure of the world in ways that let them make useful inferences and CBNs provide a promising solution to this kind of problem.  
Pearl’s work sparked a wave of research into causal cognition, with CBNs used to model people’s causal beliefs, active (interventional) learning, and model-based judgments.  However, this has revealed that CBN theories both over- and under-estimate human abilities.  Here we attempt to understand these differences and use them to better understand causal cognition. 

1.    James Woodward -  Normative theory and descriptive psychology in understanding causal reasoning:  The role of invariance and specificity (20 mins)
CBNs provide an attractive framework for capturing aspects of reasoning to do with the distinctions people draw between causal and correlational relationships. James will focus on the distinct problem of understanding distinctions people draw among causal relationships, including their apparent preference for relationships that are more stable (or invariant) and more specific, and will raise the question of how, if at all, these distinctions can be connected to the CBN framework.

2.    Anna Coenen - Beyond controlling variables: strategies of of causal experimentation (20 mins)
Anna will discuss some recent work that investigates what intervention strategies people use to figure out which variables cause events to happen

3.    Neil Bramley - Learning and representing causal structure in time (20 mins)
Neil will discuss experiments on causal learning from temporal (rather than contingency) information, and augmenting CBNs to model representations that support both temporal and statistical inference.

4.    Elias Bareinboim - Causal inference and the data-fusion problem (20 mins)
Elias will discuss his recent work with Pearl on the data-fusion problem. This problem concerns two basic processes in human learning. First, agents (e.g., babies) learn by interacting both passively (observing) and actively (experimenting) and are able to combine these two data-collection modes in a cohesive and efficient fashion. Second, agents learn to extrapolate (transport) their experiences from where the data was gathered to where usage is needed. Overall, this work unveils the conditions and types of invariances required to perform extrapolation across heterogeneous sources of information.

5.    Steven Sloman - Causal Bayes nets theory in psychology: Invaluable and deeply flawed (10 mins)
Steven will tie things off with a discussion of what we have learned and prospects for the future of modelling causal cognition.

ABSTRACT. Symposium on ‘Counterfactual thoughts about alternatives to reality’ Symposium convener and chair: Ruth Byrne

The symposium reports recent empirical studies on counterfactual ‘if only’ thoughts about how things could have been different. People often create counterfactual alternatives to reality and imagine how aspects of past events could have turned out better or worse. Counterfactual reasoning develops throughout childhood, as illustrated by the first talk, presented by Eva Rafetseder. Counterfactual thoughts and emotions such as regret develop and change throughout middle age, as discussed in the second talk, presented by Sarah Beck. They can have both positive and negative effects in many social judgments, including moral judgments and health decisions. For example, the tendency to imagine alternatives to the past is associated with the ability and tendency to engage in deception, as shown by experiments reported in the third talk presented by Raluca Briazu. Thoughts about how a good outcome that resulted from a morally uplifting act could have turned out differently ‘if only…’ or could have turned out the same ‘even if…’ differentially affect judgments about the appropriateness of morally good acts, as discussed in the fourth talk presented by Ruth Byrne. Counterfactual thoughts can also have positive and negative effects on risk taking, for example in decisions about health matters, as considered in the fifth talk presented by Kai Epstude. The symposium illustrates the consequences of imagining counterfactual alternatives to reality on many everyday judgments.

1. Eva Rafetseder and Josef Perner University of Stirling, UK and University of Salzburg, Austria ‘Understanding counterfactuality: evidence from developmental studies.’

2. Sarah Beck, Lily FitzGibbon, Suzanne Higgs, Ian Apperly, and Jane Raymond University of Birmingham, UK ‘Do regret and counterfactual thinking develop in middle-age?’

3. Raluca Briazu, Catherine Deeprose, Giorgio Ganis, and Clare Walsh University of Plymouth, UK ‘Thinking counterfactually, acting immorally – the link between counterfactual thought and lies.’

4. Ruth Byrne and Tiago da Silva Almeida Trinity College Dublin, University of Dublin, Ireland ‘Counterfactuals, semifactuals, and judgments about morally good acts.’

ABSTRACT. In just a few years, research on moral psychology has known tremendous growth, largely outpacing traditional research on reasoning (the publication and citation ratio between the two fields has gone from 1 to more than 3). The rapid growth of moral psychology has generated both a challenge and an opportunity for the psychology of reasoning. The challenge is to make reasoning research relevant for moral cognition theories, which have shown a tendency to make token mentions of reasoning, or to ignore theories of reasoning even when they could have been informative. With that challenge comes a great opportunity though, for specialists of reasoning to leverage the current impact of moral research into reaching a new and vast audience.

It is at this juncture that we undertook the publication (planned in 2016) of a collection entitled "Moral Inferences", meant to serve as a checkpoint between the psychology of reasoning and the psychology of morality. The book gathers contributions from reasoning scholars who have transitioned to the study of morality, and contributions from specialists of morality whose work resonates with the psychology of reasoning. The contributors have been enthusiastic about organizing a symposium in which they would each present their chapter, and jointly discuss the future of research on moral reasoning. This is the symposium idea we are submitting. Even though not all authors were available, we are still looking at 10 talks in total. Accordingly, we would very much love to be allowed to use two back-to-back symposia for this event.

Hugo Mercier, Thomas Castelain, Nafees Hamid, Bradly Marín Picado: The power of moral arguments

Bastien Trémolière, Wim De Neys, Jean-François Bonnefon: From reasoning to moral reasoning: a common dual-process toolbox

Jonas Nagel, Michael Waldmann, Alex Wiegmann: Causal models mediate moral inferences

Jonathan Ellis, Eric Schwitzgebel: Rationalization in moral and philosophical thought

Philipp Koralus and Mark Alfano: Reasons-based moral judgment and the erotetic theory 

ABSTRACT. SYMPOSIUM CO-ORGANIZERS:

Jeff Zemla (Brown Univ.) & Claire Cook (McGraw-Hill Education)

---------

Explanations play an important role in learning. Educators use explanations to impart knowledge, to encourage critical thinking, and to help students connect disparate ideas. In addition to the epistemological benefit of explanations, numerous studies have shown that engaging in explanation plays an influential mechanistic role in learning. Generating explanations leads us to be more aware of the boundaries of our own understanding, and the more nuanced entailment relationships we assume are in play, which can guide our inquiry and information-seeking behavior. Instructor-provided explanations offer us a scaffold to build knowledge and draw novel inferences.

How can we leverage explanations as a tool to enhance learning outcomes? Empirical studies can help us understand the impact of different pedagogical methods. What are the benefits of having students self-generate explanations, as opposed to being provided explanations from a teacher? In addition to traditional learning environments, we can use modern tools (such as online learning platforms) that enable us to explore new methods for using explanations, such as crowdsourcing explanations, adaptive tracking/selection, and iterative techniques. We can also get better traction on what features broadly make some explanations more helpful than others, when we evaluate them at scale.

Although it is clear that we can use explanation as a tool to aid learning, explanations can also lead us astray. We use explanations to build models of the world and when these models are imperfect, we may draw invalid inferences or fail to explore (and exploit) more broadly in a way that would benefit learning. In addition, we sometimes rely on simple heuristics to evaluate explanations. For instance, we judge explanations to be true based on their appeal to inherent features, leading us to endorse explanations that may be factually incorrect. Understanding the aspects that make an explanation appealing may help us craft better explanations and avoid bias in learning contexts.

This symposium brings together a number of perspectives on the risks and benefits of explanation: When does explanation boost learning, and when does it interfere or provide no benefit? How do these potential benefits interact with other factors such as the nature of a task, the age of the learner, the domain of study, and the type of learning context? What do findings from empirical work on explanation suggest about how to deploy explanation most effectively as a tool in the classroom, as well as in informal learning contexts and everyday thinking?

---------

SYMPOSIUM TALKS

1) Caren Walker* (UCSD) & Tania Lombrozo (Berkeley) TITLE: Explaining the moral of the story: Generating explanations facilitates children’s learning of abstract themes

2) Andrei Cimpian* & Daniel Storage (Univ. of Illinois) TITLE: Spontaneous explanations exhibit an inherence bias: Evidence from a false-recognition paradigm

3) Joseph Jay Williams* (Harvard University) TITLE: Discovering which explanations help learning, for whom: Scalable crowdsourcing and machine learning methods via MOOClets

4) Susan Letourneau* (Brown Univ. & Providence Children’s Museum), David Sobel (Brown Univ.), & Robin Meisner (Providence Children’s Museum) TITLE: Scientific reasoning & parent-child interactions: Learning through exploration and explanation in children's museum environments

5) Christine Gouveia* & Claire Cook (McGraw-Hill Education) TITLE: The balancing act: Supporting K-12 learners with content that drives instruction without stifling inquiry

*indicates speaker

ABSTRACT. Four talks present fresh new looks on classic perspectives on thought, emphasizing social experiences. Two talks focus on the self. Knobe discusses the ‘true self’ and Grossmann the difficulties of self-transcendence. Two talks address broad social phenomena. Vohs details new findings about mental time travel, and Baumeister argues that consciousness is for enabling talking and thus communication.

What is the essence of the self? Knobe’s prior work showed that people see their true self (meaning, authentic) as morally good. Knobe’s recent work shows that this effect is due to a general fact about the nature of psychological essentialism — the assumption that entities have good essences. Four studies showing moderation, mediation, and cross-cultural robustness (US, Singapore, Russia, Columbia) support this hypothesis. Shifting to the challenge of going beyond one’s own viewpoint is Grossmann’s work on wise reasoning. Wisdom involves humility, seeing the world as in flux, and considering others’ views. Grossmann’s new experiments and daily dairy studies demonstrate that the ability to reason wisely varies dramatically, and is particularly difficult for personal problems. One key antidote is cueing people to adopt a self-distanced perspective, which enhances wise reasoning. What occupies the mind has long been studied, but much less with respect to time. Vohs’s two large-scale experience sampling studies (total reports=9,542) had respondents report whether their last thought was about the past, present, future, or had no time aspect. Thoughts in time (past, present, future) were social in nature (about the self and others), suggesting that time is highly social. No-time thoughts were distinctly asocial. Stark and perhaps surprising results include a 4:1 ratio of future- vs past-focused thoughts, mental time travel (both directions) predicting more negative concurrent feelings, and the highly meaningful nature of thoughts combining times (e.g., present and future). Thinking in a social context also is Baumeister’s focus on consciousness's function, which he contends is communication via speech. Benefits of thought can be duplicated by unconscious, automatic processes, whereas talking seems a powerful and profound exception (i.e., consciousness is indispensable to speech). Consciousness used for talking enables social correction, which Baumeister argues suggests that humans evolved to discover the truth collectively, not individually. These talks match well with the Thinking conference, resonating with work by the keynotes: essence and identity (Nichols, Weber); reasoning (Chater, Lance); mental time travel (Tomasello, Sunstein); and communication and consciousness (Tomasello, Weber).

ABSTRACT. Some dual process theories suggest that reasoning occurs via two types of information processing: Type 1 and Type 2. Expertise has received little investigation from a dual process perspective. That is, how can a process which is initially experienced as Type 2, over time, take on the characteristics of a Type 1? A recently published theory of reasoning, the Complex Emergent Modularity model (CEM; Wastell, 2014), in conjunction with a leading theory of consciousness, the Global Workspace theory (GW; Baars et al., 2013), offers an elegant solution to this problem but has not been empirically tested. We examined whether mathematical expertise predicts performance differences on the Cognitive Reflections Test (CRT; Frederick, 2005). In the present study, participants with higher mathematical expertise outperformed participants with lower expertise. Furthermore, a three way interaction was observed, wherein an additional memory task negatively affected participants with greater expertise but did not affect those with lower expertise. Theoretical limitations in dual process theories render an interpretation from that perspective neither helpful for understanding this phenomenon, nor for guiding future research. We suggest that future research investigating the role of expertise in reasoning may be more effective if guided by the CEM/GWT theories of reasoning.

ABSTRACT. A great deal of modern reasoning and decision-making research has established that human judgment is often biased by intuitive heuristics. Contemporary “error” or bias detection studies have focused on reasoners’ abilities to detect whether their heuristic answer is erroneous and conflicts with logical or probabilistic principles. One crucial remaining question is whether there are individual differences in this bias detection efficiency and how this affects reasoning performance. Here we present four studies in which co-registration of different error detection measures allowed us to assess bias detection sensitivity at the individual participant level in a range of reasoning tasks. Our results indicate that most individuals show robust bias detection (as indexed by increased latencies and decreased confidence associated with erroneous answers) even in divergent reasoning contexts. However, there are subgroups of reasoners who consistently fail to do so. We discuss implications for the debate on human rationality and popular dual process theories.

ABSTRACT. A number of researchers apply logical modelling to problems from cognitive psychology. Our work on Theory of Mind belongs to this tradition, and is distinguished by (1) its use of hybrid modal logic, which takes local shifts of perspectives as fundamental (2) its focus on second-order reasoning about mental states.

The literature on first-order false belief is extensive, but less is known about the second-order case. We analyze the first-order case by combining belief formation sequencing with perspectival reasoning, and lift this analysis to the second-order level by adding a recursive belief manipulation rule. Technically, the second-order formalization is obtained by embedding Braüner’s first-order formalization (Conference on Theoretical Aspects of Rationality and Knowledge 2013, Journal of Logic, Language and Information 2014, Annual Conference of the Cognitive Science Society 2015) into a higher-level proof which leaves the original layer of reasoning unchanged.

We argue that our formalization supports an existing position on the development of second-order reasoning, the “conceptual-change” position. Accordingly, second-order mastery is a fundamental cognitive achievement that requires new conceptual apparatus, not merely increased information processing capacity. The application of the belief manipulation rule suggests that second-order reasoning involves a richer understanding of belief that reflects a conceptual shift.

ABSTRACT. It is well known that spontaneous reasoners tend to make fallacies and to suppress valid inférences (Byrne 1989) and that they do so in descriptive contexts and avoid them in normative contexts (Cosmides 1989). This paper will hold that descriptive inferences are such because they are part of our cognitive activity and should be explained by their cognitive functions. It will be held that AC and DA fallacies in conditional reasoning reduce causality to a single cause and that the MPT fallacy on disjunctions and the MTP fallacy on incompatibilities make us categorize via simple dichotomic categories. This way, learning to avoid these fallacies is cognitively relevant, allowing us to consider multiple alternative possible causes to an effect and to categorize in hierarchical structures. Suppressions should also be explained from their cognitive functions. The suppression of MPP and MTT on conditionals will be presented as allowing the lengthening of causal chains; while the suppression of valid MTP on disjunctions or of the valid MPT on incompatibilities will be presented as an effect of the discovery of new categories. Thus, our relation to logic is modelled by cognitive causes.

ABSTRACT. Unhealthy dietary choices are an increasing problem, with 24% of men and 27% of women in England already classified as obese (Health Survey for England, 2014). A targeted approach to healthy eating interventions is needed which segments the population according to their dietary preferences and tackles the cognitive biases that underlie these preferences.

We report the results of a large UK-wide survey which aims to identify the relationship between cognitive biases, food choices and health outcomes. Data was collected on participants’ susceptibility to a range of classic cognitive biases covering: framing effects; sunk costs; anchoring and adjustment; discounting; outcome bias and mental accounting. Food and drink purchasing was recorded over a 2-week period, and other associated data on demographics, Body Mass Index (BMI), cognitive capacity, shopping habits and thinking styles was also collected.

Initial high level results show that cognitive biases contribute to a significant amount of variance in BMI, even with the effects of traditional demographic predictors taken into account. Further trends in the data will be explored, focussing on links between biases and particular patterns of unhealthy food purchasing behaviour. Implications for targeted food choice interventions will be discussed.

ABSTRACT. The efficacy of thought experiments (TEs) raises a fundamental paradox: How can a process involving no new data about the world contribute to advances in knowledge about the world? Several resolutions have been offered, including: a) TEs involve imagistic simulations that produce “new data;” b) TEs are arguments; c) TEs highlight contradictions among beliefs. We asked if a TE can help young children to advance their understanding of matter, and if yes, then how. Like Aristotle, young children believe weight to be a property that some physical entities lack. Twenty-four children who maintained that a single grain of rice weighs nothing at all were assigned to a real experiment and a thought experiment condition. In the real experiment, children received evidence that a single grain of rice can topple a card placed on a fulcrum. The TE was structurally equivalent, but it was simulated in the head. Both the real and the TE had large and equivalent effects on posttest judgments about the weight of a grain of rice. Details of the data confirm imagistic simulation can drive belief change, and at least in this case, there is no evidence for the other two resolutions of the fundamental paradox.

ABSTRACT. Opacity is viewed as the hallmark of the mind. In the psychological discourse, this kind of philosophical statement is connected with a belief that children’s understanding of opacity is tied to their understanding of the mind as a representational medium (Kamawar et al. 2011, Apperly et al. 2003; Rakoczy 2016). For many years, it was accepted that if children pass FBT their conceptual development is sufficiently advanced to deal with opacity. However, some newer findings undermine this view (Apperly et al. 2003; Kamawar et al. 2011). Aiming at contributing to this debate, we run a complex, multidimensional study involving 195 children in 4 age groups. Children were asked to solve FBT and a series of intensionality tests. Our study revealed that there was a substantial sample of participants who passed FBT, but did not passed intensionality tests. The new design of our tasks allowed us to reject explanations attributing the discrepancy between FBT and intensionality task only to various syntactic or pragmatic factors. We show that at a certain stage the development of ToM unfolds as continuous conceptual enrichment which is interpreted as a situation in which the coarse-grained content of a concept is substituted with the fine-grained content.

ABSTRACT. Recent studies have demonstrated that young children can evaluate some of the arguments people offer to them. However, experimental studies of sensitivity to argument quality have not yet targeted children younger than 3 years of age. The present study aims at testing the ability of 2 years old (23 to 30 months) children to weight arguments. Children are asked to name ambiguous items, then an informant disagrees with them in one of three ways: either provides no argument, a weak (circular) argument, or a strong argument. Preliminary results suggest that 2–year-olds are more likely to adopt the testimony of informants who provide arguments, whether they are weak or strong. We are currently acquiring more data that will answer two more questions: is there in fact a difference between weak and strong arguments, and are informants who provide different types of testimony (no argument, weak argument, strong argument) more or less likely to be believed in a follow-up task in which the informants do not give any argument.

ABSTRACT. Imagine a scientist who falsifies data; she is a scientist in a descriptive sense, but perhaps not a true scientist in a normative sense. The current study tested whether preschoolers, like adults, can hold these dual character concepts (Knobe, Prasada, and Newman, 2013), meaning they dissociate normative and descriptive criteria when determining membership in a category (e.g., scientist). Children heard vignettes in which characters either were or were not labeled as members of social role categories (e.g., “Joe is a scientist”) and were described as either instantiating or not instantiating the normative values associated with the social role. Children were then asked whether the character was really a category member “deep down.” Results revealed that children assigned category membership more often to individuals who had the appropriate category label (M = .53, CI =.45-.61) than those who did not (M= .25, CI =.19-.32), (Wald χ² (1) = 51.78, p < .001), and also more to people who instantiated normative values (M = .58, CI=.50-.64) than those who did not (M = .22, CI =.16-.30), (Wald χ² (1) = 53.91, p < .001). These results indicate that preschoolers dissociate descriptive information (i.e., labels) from normative information when determining category membership.

ABSTRACT. How can education optimize transmission of knowledge while also fostering further learning? We focus on children at the cusp of formal schooling and investigating learning in three contexts: after instruction by a knowledgeable teacher, after pedagogical questioning by a knowledgeable teacher ---Pedagogical questioning---and after questioning by a naive teacher. Consistent with the predictions of a formal model of pedagogical learning, we find that instruction from a knowledgeable teacher allows effective information transfer but at the cost of exploration and further learning, while naive questioning fosters exploration and learning at the cost of transmitting knowledge. In contrast, pedagogical questioning affords the joint benefits of both effective transmission of knowledge and fosters exploration leading to further learning. The results suggest that by the time they arrive at formal schooling, children are already reasoning about the knowledge of others and their choices of instructional methods, with predictable implications of these choices for learning.

ABSTRACT. Although an increasing amount of research has approached the study of creativity from a dual-process perspective, further work is required to understand the way that intuitive and analytic modes of thought interact in creative reasoning more precisely. To explore this issue, we applied a two-response methodology developed by Thompson et al. (2011) within an analogical reasoning task. Participants saw four word sets and judged if an analogical relation existed between the word pairs, which varied as a function of validity (valid vs. invalid analogy) and semantic distance (close vs. distant). Importantly, participants made several responses on each trial, first making an intuitive judgment as quickly as possible, followed by a more reasoned response in which they could take as long as necessary to make an accurate judgment. Participants rated how right their first response felt and how confident they were in their final judgment. The results provided insight into the metacognitive processes supporting analogical reasoning. Specifically, lower feeling of rightness ratings on the first judgment led to longer RTs and more answer changes on second judgments for creative analogies. More generally, the results show that analytic thought is important for creative analogical reasoning.

Speaker: Nathaniel Barr

ABSTRACT. We call "stumpers" riddles which have the following properties: No (satisfying) solution comes to mind - respondents feel "stumped". But once provided, the solution is immediately recognized as correct, eliciting the reaction: "Of course! How could I've missed that?". A familiar example involves a man who is killed in a car accident. His injured son is rushed to hospital, where the surgeon exclaims: "OMG, it's my son!". How can this be? The simple solution, which nonetheless often eludes many, is that the surgeon is the boy's mother. The reason it does not come to mind is gender stereotyping – at the mention of a surgeon, stumped respondents only envision a male. Based on the model of this stumper, we assembled a diverse set of unfamiliar stumpers, and explore why each one is a stumper. We identify and study a diverse set of psychological principles (some novel) that prevent respondents from seeing the solutions that seem so obvious ex post. We provide independent evidence for the principles, and for the stumping dynamics.

ABSTRACT. How much control are people perceived to have over their minds? Are different degrees of control ascribed to different aspects of our mental functioning (thoughts, beliefs, emotions, etc.)? And how do perceptions of mental control relate to judgments of responsibility and character? Surprisingly, no comprehensive studies have addressed these questions. We conducted several experiments to examine ordinary judgments of agent control across a diverse range of mental phenomena. Results show that people reliably attributed increasing controllability across the following mental state categories: emotions (lowest control), desires, beliefs, evaluations, imaginings, deliberations, and intentions (highest control). Increases in perceived control across these categories corresponded with higher judgments of responsibility as well as judgments that the mental states were more revealing of the agent’s character. This pattern was also observed within each of the mental state categories: when particular instances of beliefs, desires, emotions (and so on) were judged as more controllable, they were rated higher in responsibility and more character revealing relative to their less controllable counterparts. These results replicated across different subpopulations, abstract and concrete stimuli, and across naturalistic and tailored scenarios. Together, these experiments form a foundation for studying the social regulation of mental behavior.

ABSTRACT. The ‘Aha!’ experience, the sudden burst of insight, has often been explained through an internal cognitive framework. However, external actions may facilitate insight. The role of transfer and material interaction in insight problem-solving was investigated using the Cheap Necklace Problem. In Experiment 1, participants completed the same problem twice after a two-week gap, either using a standard paper-and-pencil questionnaire or using physical artefacts. Performance, measured as successful completion of the task, was substantially better with increased interactivity. All participants demonstrated transfer as they significantly improved in performance and were faster to reach a solution after the two-week gap. Experiment 2 measured the role of transfer and material interaction when the problem presentation changes after the two-week gap (low-interactivity to high-interactivity or high-interactivity to low-interactivity). Transfer was evident with improved performance after the two-week gap and improved latencies to solution. Transfer was better facilitated for problem-solvers who were presented with the low-interactivity version first, then experienced the interactive version. These quantitative findings were extended through a detailed qualitative video analysis of problem solving strategies in order to examine the exhaustive interactions between a problem-solver and his/her environment. The role of interactivity will be discussed with reference to the distributed cognition framework.

ABSTRACT. Since Newell and Simon, thinking has been analogized to computation, but the comparison between abstract symbol manipulators and living brains has commonly provided more fog than clarity. The comparison seems to slight brains, but a real materialist might object that computers actually manipulate electrons, not symbols. Using a novel application of fundamental concepts of information theory and thermodynamics, one can show that the differences between computers and brains are less a matter of principle than of organization and precedent. The lack of standards committees for brains has consequences both for the transmission of meaning, and its ontogeny.

As in a computer, error correction in living systems is organized into multiple levels. Bits make bytes, then blocks, then entire emails, and sounds beget words that are combined into sentences. Unlike computers, living systems embody many more levels, and the error correction at each level is inadequate by usual information theoretic standards. As with many software systems, this is not a bug but a feature, and has consequences for the efficient use of energy, the immediacy of sensory perception, and the embodiment of thought. It further provides a materialist explanation for creativity that does not depend on randomness, quantum fluctuations, or magic.

ABSTRACT. Philosophical observation and psychological studies indicate that people draw subtle distinctions in the normative domain.  But it remains unclear exactly what gives rise to such distinctions. On one prominent approach, emotion systems trigger non-utilitarian judgments. The main alternative, inspired by Chomskyan linguistics, suggests that moral distinctions derive from an innate moral grammar. We draw on recent developments in learning theory to argue that several aspects of moral judgment can be explained in terms of rational inference.