Inferring interventional predictions from observational learning data

Previous research has shown that people are capable of deriving correct predictions for previously unseen actions from passive observations of causal systems (Waldmann & Hagmayer, 2005). However, these studies were limited, since learning data were presented as tabulated data only, which may have turned the task more into a reasoning rather than a learning task. In two experiments, we therefore presented learners with trial-by-trial observational learning input referring to a complex causal model consisting of four events. To test the robustness of the capacity to derive correct observational and interventional inferences, we pitted causal order against the temporal order of learning events. The results show that people are, in principle, capable of deriving correct predictions after purely observational trial-by-trial learning, even with relatively complex causal models. However, conflicting temporal information can impair performance, particularly when the inferences require taking alternative causal pathways into account.     

Inferring action structure and causal relationships in continuous sequences of human action

In the real world, causal variables do not come pre-identified or occur in isolation, but instead are embedded within a continuous temporal stream of events. A challenge faced by both human learners and machine learning algorithms is identifying subsequences that correspond to the appropriate variables for causal inference. A specific instance of this problem is action segmentation: dividing a sequence of observed behavior into meaningful actions, and determining which of those actions lead to effects in the world. Here we present a Bayesian analysis of how statistical and causal cues to segmentation should optimally be combined, as well as four experiments investigating human action segmentation and causal inference. We find that both people and our model are sensitive to statistical regularities and causal structure in continuous action, and are able to combine these sources of information in order to correctly infer both causal relationships and segmentation boundaries.     

Inferring causal networks from observations and interventions

Information about the structure of a causal system can come in the form of observational data—random samples of the system's autonomous behavior—or interventional data—samples conditioned on the particular values of one or more variables that have been experimentally manipulated. Here we study people's ability to infer causal structure from both observation and intervention, and to choose informative interventions on the basis of observational data. In three causal inference tasks, participants were to some degree capable of distinguishing between competing causal hypotheses on the basis of purely observational data. Performance improved substantially when participants were allowed to observe the effects of interventions that they performed on the systems. We develop computational models of how people infer causal structure from data and how they plan intervention experiments, based on the representational framework of causal graphical models and the inferential principles of optimal Bayesian decision‐making and maximizing expected information gain. These analyses suggest that people can make rational causal inferences, subject to psychologically reasonable representational assumptions and computationally reasonable processing constraints.     

How well is causal structure inferred from cooccurrence information?

Individual causal relations tend to form parts of larger causal systems. In five experiments the ability of participants to infer the structures of two systems involving five entities from patterns of cooccurrence was investigated. Although the systems were fully deterministic, there were no indirect causal relations, and participants were given guidance on how to infer causal structure from cooccurrence information, low rates of success were observed. Judgements were based more on information about temporal relations, even though no guidance on the use of temporal relation information for causal inference was provided. When no guidance was provided, no participants succeeded in inferring the structure of both systems. The results indicate that temporal relations may be preferred as cues to causal structure over patterns of cooccurrence.     

The advantage of timely intervention

Can people learn causal structure more effectively through intervention rather than observation? Four studies used a trial-based learning paradigm in which participants obtained probabilistic data about a causal chain through either observation or intervention and then selected the causal model most likely to have generated the data. Experiment 1 demonstrated that interveners made more correct model choices than did observers, and Experiments 2 and 3 ruled out explanations for this advantage in terms of informational differences between the 2 conditions. Experiment 4 tested the hypothesis that the advantage was driven by a temporal signal; interveners may exploit the cue that their interventions are the most likely causes of any subsequent changes. Results supported this temporal cue hypothesis.     

因果模型在类比推理中的作用

通过操纵因果模型的特征维度及推理方向, 探讨因果模型在类比推理中的作用。实验一探讨了当结果特征未知时进行类比推理的情况, 发现在一果多因时, 被试采用因果模型进行类比推理, 而在一因多果时, 被试同时采用因果模型和计算模型进行类比推理。实验二探讨当原因特征未知时进行类比推理的情况, 发现在一果多因和一因多果时, 被试均采用因果模型进行类比推理。结果表明：(1)当结果特征未知时, 人们会建构因果模型进行类比推理。且当因果模型和计算模型处于冲突情境时, 人们会采用因果模型进行类比推理; 但当因果模型和计算模型处于非冲突情境时, 人们会同时采用因果模型和计算模型。(2)当原因特征未知时, 即按照因果模型推理的难度增加时, 人们仍会建构因果模型进行类比推理。     

Reconciling Compensatory and Noncompensatory Strategies of Cue Weighting: A Causal Model Approach

When forming a judgment about any unknown item, people must draw inferences from information that is already known. This paper examines causal relationships between cues as a relevant factor influencing how people determine the amount of weight to place on each piece of available evidence. We propose that people draw from their beliefs about specific causal relationships between cues when determining how much weight to place on those cues, and that understanding this process can help reconcile differences between predictions of compensatory and lexicographic heuristic strategies. As causal relationships change, different cues become more or less important. Across three experiments, we find support for the use of causal models in determining cue weights, but leave open the possibility that they work in concert with other strategies as well. We conclude by discussing relative strengths and weaknesses of the causal model approach relative to existing models, and suggest areas for future research. Copyright © 2016 John Wiley & Sons, Ltd.     

Attention to Explicit and Implicit Contrast in Verb Learning

Contrast information could be useful for verb learning, but few studies have examined children's ability to use this type of information. Contrast may be useful when children are told explicitly that different verbs apply, or when they hear two different verbs in a single context. Three studies examine children's attention to different types of contrast as they learn new verbs. Study 1 shows that 3.5-year-olds can use both implicit contrast (“I'm meeking it. I'm koobing it.”) and explicit contrast (“I'm meeking it. I'm not meeking it.”) when learning a new verb, while a control group's responses did not differ from chance. Study 2 shows that even though children at this age who hear explicit contrast statements differ from a control group, they do not reliably extend a newly learned verb to events with new objects. In Study 3, children in three age groups were given both comparison and contrast information, not in blocks of trials as in past studies, but in a procedure that interleaved both cues. Results show that while 2.5-year-olds were unable to use these cues when asked to compare and contrast, by 3.5 years old, children are beginning to be able to process these cues and use them to influence their verb extensions, and by 4.5 years, children are proficient at integrating multiple cues when learning and extending new verbs. Together these studies examine children's use of contrast in verb learning, a potentially important source of information that has been rarely studied.     

A Bayesian Theory of Sequential Causal Learning and Abstract Transfer

Two key research issues in the field of causal learning are how people acquire causal knowledge when observing data that are presented sequentially, and the level of abstraction at which learning takes place. Does sequential causal learning solely involve the acquisition of specific cause‐effect links, or do learners also acquire knowledge about abstract causal constraints? Recent empirical studies have revealed that experience with one set of causal cues can dramatically alter subsequent learning and performance with entirely different cues, suggesting that learning involves abstract transfer, and such transfer effects involve sequential presentation of distinct sets of causal cues. It has been demonstrated that pre‐training (or even post‐training) can modulate classic causal learning phenomena such as forward and backward blocking. To account for these effects, we propose a Bayesian theory of sequential causal learning. The theory assumes that humans are able to consider and use several alternative causal generative models, each instantiating a different causal integration rule. Model selection is used to decide which integration rule to use in a given learning environment in order to infer causal knowledge from sequential data. Detailed computer simulations demonstrate that humans rely on the abstract characteristics of outcome variables (e.g., binary vs. continuous) to select a causal integration rule, which in turn alters causal learning in a variety of blocking and overshadowing paradigms. When the nature of the outcome variable is ambiguous, humans select the model that yields the best fit with the recent environment, and then apply it to subsequent learning tasks. Based on sequential patterns of cue‐outcome co‐occurrence, the theory can account for a range of phenomena in sequential causal learning, including various blocking effects, primacy effects in some experimental conditions, and apparently abstract transfer of causal knowledge.     

Preschool children learn about causal structure from conditional interventions

The conditional intervention principle is a formal principle that relates patterns of interventions and outcomes to causal structure. It is a central assumption of experimental design and the causal Bayes net formalism. Two studies suggest that preschoolers can use the conditional intervention principle to distinguish causal chains, common cause and interactive causal structures even in the absence of differential spatiotemporal cues and specific mechanism knowledge. Children were also able to use knowledge of causal structure to predict the patterns of evidence that would result from interventions. A third study suggests that children's spontaneous play can generate evidence that would support such accurate causal learning.     

Interaction between previous beliefs and cue predictive value in covariation-based causal induction

The main aim of this work was to show the impact of preexisting causal beliefs on causal induction from cause-effect co-occurrence information, when several cues compete with each other for predicting the same effect. Two different causal scenarios -- one social (a), the other medical (b) -- were used to check the generality of the effects. In Experiments 1a and 1b, participants were provided information on the co-occurrence of a two-cause compound and an effect, but not about the potential relationship between each cause by its own and the effect. As expected, prior beliefs -- induced by means of instructions -- strongly modulated the causal strength assigned to each element of the compound. In Experiments 2a and 2b, covariation evidence was provided, not only about the predictive value of the two-cause compound, but also about one of the elements of the compound. When this evidence was available, prior beliefs had less impact on judgments, and these were mostly guided by the relative predictive value of the cue. These results demonstrate the involvement of inferential integrative mechanisms in the generation of causal knowledge and show that single covariation detection mechanisms -- either rule-based or associative -- are insufficient to account for human causal judgment. At the same time, the fact that the power of new covariational evidence to change prior beliefs depended on the availability of information on the relative (conditional) predictive value of the target candidate cause suggests that causal knowledge derived from information on causal mechanisms and from covariation probably share a common representational basis.     

Does causal knowledge help us be faster and more frugal in our decisions?

One challenge that has to be addressed by the fast and frugal heuristics program is how people manage to select, from the abundance of cues that exist in the environment, those to rely on when making decisions. We hypothesize that causal knowledge helps people target particular cues and estimate their validities. This hypothesis was tested in three experiments. Results show that when causal information about some cues was available (Experiment 1), participants preferred to search for these cues first and to base their decisions on them. When allowed to learn cue validities in addition to causal information (Experiment 2), participants also became more frugal (i.e., they searched fewer of the available cues), made more accurate decisions, and were more precise in estimating cue validities than was a control group that did not receive causal information. These results can be attributed to the causal relation between the cues and the criterion, rather than to greater saliency of the causal cues (Experiment 3). Overall, our results support the hypothesis that causal knowledge aids in the learning of cue validities and is treated as a meta-cue for identifying highly valid cues.     

A probabilistic contrast model of causal induction

Deviations from the predictions of covariational models of causal attribution have often been reported in the literature. These include a bias against using consensus information, a bias toward attributing effects to a person, and a tendency to make a variety of unpredicted conjunctive attributions. It is contended that these deviations, rather than representing irrational biases, could be due to (a) unspecified information over which causal inferences are computed and (b) the questionable normativeness of the models against which these deviations have been measured. A probabilistic extension of Kelley's analysis-of-variance analogy is proposed. An experiment was performed to assess the above biases and evaluate the proposed model against competing ones. The results indicate that the inference process is unbiased.     

Temporal information and children's and adults' causal inferences

Three experiments examined whether children and adults would use temporal information as a cue to the causal structure of a three-variable system, and also whether their judgements about the effects of interventions on the system would be affected by the temporal properties of the event sequence. Participants were shown a system in which two events B and C occurred either simultaneously (synchronous condition) or in a temporal sequence (sequential condition) following an initial event A. The causal judgements of adults and 6–7-year-olds differed between the conditions, but this was not the case for 4-year-olds' judgements. However, unlike those of adults, 6–7-year-olds' intervention judgements were not affected by condition, and causal and intervention judgements were not reliably consistent in this age group. The findings support the claim that temporal information provides an important cue to causal structure, at least in older children. However, they raise important issues about the relationship between causal and intervention judgements.     

The crossed-hands deficit in tactile temporal-order judgments: the effect of training

Several recent studies have shown that judgments of temporal order for tactile stimuli presented to the two hands are greatly affected by crossing the hands. The size of the threshold for judging temporal order may be up to four times larger with the hands crossed as compared to the hands uncrossed. The results from these recent studies suggest that with crossed hands, contrary to many situations involving the integration of tactile and proprioceptive information, subjects have difficulty in adjusting their perception of tactile inputs to correspond with the spatial positions of the hands. In the present study we examined the effect of training in judging temporal order on the size of this crossed-hands deficit--the difference in the thresholds for temporal-order judgments when the hands are crossed and uncrossed. All training procedures produced significant declines in the size of the deficit. With training, the difference between crossed-hands and uncrossed-hands temporal-order thresholds dropped from several hundred milliseconds to as little as 19 ms. A group of percussionists with experience in playing with crossed hands showed the same crossed-hands effects as non-musicians. The results were consistent in showing that the crossed-hands deficit was never completely eliminated but was greatly reduced with training. The implication is that subjects are able to adjust to the crossed-hands posture with modest amounts of training. The results are discussed in terms of the explanations that have been offered for the crossed-hands deficit.     

On the discrepant results in synchrony judgment and temporal-order judgment tasks: a quantitative model

Research on the perception of temporal order uses either temporal-order judgment (TOJ) tasks or synchrony judgment (SJ) tasks, in both of which two stimuli are presented with some temporal delay and observers must judge the order of presentation. Results generally differ across tasks, raising concerns about whether they measure the same processes. We present a model including sensory and decisional parameters that places these tasks in a common framework that allows studying their implications on observed performance. TOJ tasks imply specific decisional components that explain the discrepancy of results obtained with TOJ and SJ tasks. The model is also tested against published data on audiovisual temporal-order judgments, and the fit is satisfactory, although model parameters are more accurately estimated with SJ tasks. Measures of latent point of subjective simultaneity and latent sensitivity are defined that are invariant across tasks by isolating the sensory parameters governing observed performance, whereas decisional parameters vary across tasks and account for observed differences across them. Our analyses concur with other evidence advising against the use of TOJ tasks in research on perception of temporal order.     

The importance of decision making in causal learning from interventions

Recent research has focused on how interventions benefit causal learning. This research suggests that the main benefit of interventions is in the temporal and conditional probability information that interventions provide a learner. But when one generates interventions, one must also decide what interventions to generate. In three experiments, we investigated the importance of these decision demands to causal learning. Experiment 1 demonstrated that learners were better at learning causal models when they observed intervention data that they had generated, as opposed to observing data generated by another learner. Experiment 2 demonstrated the same effect between self-generated interventions and interventions learners were forced to make. Experiment 3 demonstrated that when learners observed a sequence of interventions such that the decision-making process that generated those interventions was more readily available, learning was less impaired. These data suggest that decision making may be an important part of causal learning from interventions.     

Judgment dissociation theory: an analysis of differences in causal,counterfactual, and covariational reasoning

Research suggests that causal judgment is influenced primarily by counterfactual or covariational reasoning. In contrast, the author of this article develops judgment dissociation theory (JDT), which predicts that these types of reasoning differ in function and can lead to divergent judgments. The actuality principle proposes that causal selections focus on antecedents that are sufficient to generate the actual outcome. The substitution principle proposes that ad hoc categorization plays a key role in counterfactual and covariational reasoning such that counterfactual selections focus on antecedents that would have been sufficient to prevent the outcome or something like it and covariational selections focus on antecedents that yield the largest increase in the probability of the outcome or something like it. The findings of 4 experiments support JDT but not the competing counterfactual and covariational accounts.     

Strategy selection in causal reasoning: when beliefs and covariation collide.

The present study investigated how people combine covariation information (Cheng & Novick, 1990, 1992) with pre-existing beliefs (White, 1989) when evaluating causal hypotheses. Three experiments, using both within- and between-subjects designs, found that the use of covariation information and beliefs interacted, such that the effects of covariation were larger when people assessed hypotheses about believable than about unbelievable causal candidates. In Experiment 2, this interaction was observed when participants made judgments in stages (e.g., first evaluating covariation information about a causal candidate and then evaluating the believability of a candidate), as well as when the information was presented simultaneously. Experiment 3 demonstrated that this pattern was also reflected in participants' metacognitive judgments: Participants indicated that they weighed covariation information more heavily for believable than unbelievable candidates. Finally, Experiments 1 and 2 demonstrated the presence of individual differences in the use of covariation- and belief-based cues. That is, individuals who tended to base their causality judgments primarily on belief were less likely to make use of covariation information and vice versa. The findings were most consistent with White's (1989) causal power theory, which suggests that covariation information is more likely to be considered relevant to believable than unbelievable causes.     

The role of learning data in causal reasoning about observations and interventions

Recent studies have shown that people have the capacity to derive interventional predictions for previously unseen actions from observational knowledge, a finding that challenges associative theories of causal learning and reasoning (e.g., Meder, Hagmayer, & Waldmann, 2008). Although some researchers have claimed that such inferences are based mainly on qualitative reasoning about the structure of a causal system (e.g., Sloman, 2005), we propose that people use both the causal structure and its parameters for their inferences. We here employ an observational trial-by-trial learning paradigm to test this prediction. In Experiment 1, the causal strength of the links within a given causal model was varied, whereas in Experiment 2, base rate information was manipulated while keeping the structure of the model constant. The results show that learners’ causal judgments were strongly affected by the observed learning data despite being presented with identical hypotheses about causal structure. The findings show furthermore that participants correctly distinguished between observations and hypothetical interventions. However, they did not adequately differentiate between hypothetical and counterfactual interventions.