Michael R. Waldmann

Predictive and Diagnostic Learning Within Causal Models: Asymmetries in Cue Competition

Several researchers have recently claimed that higher order types of learning, such as categorization and causal induction, can be reduced to lower order associative learning. These claims are based in part on reports of cue competition in higher order learning, apparently analogous to blocking in classical conditioning. Three experiments are reported in which subjects had to learn to respond on the basis of cues that were defined either as possible causes of a common effect (predictive learning) or as possible effects of a common cause (diagnostic learning). The results indicate that diagnostic and predictive reasoning, far from being identical as predicted by associationistic models, are not even symmetrical. Although cue competition occurs among multiple possible causes during predictive learning, multiple possible effects need not compete during diagnostic learning. The results favor a causal-model theory.

Causal Reasoning in Rats

Empirical research with nonhuman primates appears to support the view that causal reasoning is a key cognitive faculty that divides humans from animals. The claim is that animals approximate causal learning using associative processes. The present results cast doubt on that conclusion. Rats made causal inferences in a basic task that taps into core features of causal reasoning without requiring complex physical knowledge. They derived predictions of the outcomes of interventions after passive observational learning of different kinds of causal models. These competencies cannot be explained by current associative theories but are consistent with causal Bayes net theories.

Competition among causes but not effects in predictive and diagnostic learning.

Causal asymmetry is one of the most fundamental features of the physical world: Causes produce effects, but not vice versa. This article is part of a debate between the view that, in principle, people are sensitive to causal directionality during learning (causal-model theory) and the view that learning primarily involves acquiring associations between cues and outcomes irrespective of their causal role (associative theories). Four experiments are presented that use asymmetries of cue competition to discriminate between these views. These experiments show that, contrary to associative accounts, cue competition interacts with causal status and that people are capable of differentiating between predictive and diagnostic inferences. Additional implications of causal-model theory are elaborated and empirically tested against alternative accounts. The results uniformly favor causal-model theory.

KNOWLEDGE-BASED CAUSAL INDUCTION

This chapter reviews that the comparison between causal-model theory and associative accounts of causal induction highlighted a number of important differences between these two approaches. Causal-model theory postulates a rigorous separation between the learning input and mental representations. This characteristic allows for the flexible assignment of the learning input to elements of the resulting mental models. By contrast, most associative learning theories work in the tradition of stimulus response theories in which learning cues play the double causal role of representing events and eliciting responses. It discusses that this inflexibility leads to clear misrepresentations of objective causal relations. Associative theories code the learning cues as CS and the outcomes as US are unable to capture the structural characteristics of diagnostic learning situations in which effects are presented as cues. The Rescorla-Wagner theory correctly captures the asymmetry between causes and effects only when the learning situation is fortuitously presented in a way that corresponds to the implicit structural characteristics of this theory. A second major tenet of causal-model theory postulates the necessity of an interaction between top-down assumptions and the processing of the learning input. Causal-model theory represents reconciliation between theories focusing on statistical covariation learning and theories focusing on causal, mechanical processes. The chapter also explores that causal directionality is one of the most important features of causal relations that determine the way statistical relations are interpreted. It is a physical fact that multiple causes of a common effect potentially interact, whereas multiple effects of a common cause are rendered conditionally independent when the common cause is held constant. The chapter reviews that without prior knowledge that is already available at the outset of the induction process new causal knowledge cannot properly be acquired.

Causal models and the acquisition of category structure.

This article proposes that learning of categories based on cause-effect relations is guided by causal models. In addition to incorporating domain-specific knowledge, causal models can be based on knowledge of such general structural properties as the direction of the causal arrow and the variability of causal variables. Five experiments tested the influence of commoncause models and common-effect models on the ease of learning linearly separable and nonlinearly separable categories. The results show that causal models guide the interpretation of otherwise identical learning inputs, and that learning difficulty is determined by the fit between the structural implications of the causal models and the structure of the learning domain. These influences of the general properties of causal models were obtained across several different content domains, including domains for which subjects lacked prior knowledge. Tasks as apparently diverse as classical conditioning, category learning, and causal induction often require the learner to combine multiple cues in order to elicit a response. The cues may be conditioned stimuli (in conditioning), features of category instances (in category learning), or possible causes (in causal induction). Numerous learning models have been proposed in each of these areas, and a great deal of theoretical interest has focused on the extent to which common learning mechanisms may operate across these formally similar tasks. Most of these theories model learning as a domain-general process, bottom-up and basically associative in nature, that applies across diverse domains. Recently, more top-down or theory-based approaches have been proposed, which view learning as guided by domain-specific theories. In the present article we outline a position that is intermediate between these two views. We claim that a major subset of learning situations—

Throwing a Bomb on a Person Versus Throwing a Person on a Bomb Intervention Myopia in Moral Intuitions

Most people consider it morally acceptable to redirect a trolley that is about to kill five people to a track where the trolley would kill only one person. In this situation, people seem to follow the guidelines of utilitarianism by preferring to minimize the number of victims. However, most people would not consider it moral to have a visitor in a hospital killed to save the lives of five patients who were otherwise going to die. We conducted two experiments in which we pinpointed a novel factor behind these conflicting intuitions. We show that moral intuitions are influenced by the locus of the intervention in the underlying causal model. In moral dilemmas, judgments conforming to the prescriptions of utilitarianism are more likely when the intervention influences the path of the agent of harm (e.g., the trolley) than when the intervention influences the path of the potential patient (i.e., victim).

Predictive versus diagnostic causal learning: evidence from an overshadowing paradigm.

Causal directionality belongs to one of the most fundamental aspects of causality that cannot be reduced to mere covariation. This paper is part of a debate between proponents of associative theories, which claim that learners are insensitive to the causal status of cues and outcomes, and proponents of causal-model theory, which postulates an interaction of assumptions about causal directionality and learning. Some researchers endorsing the associationist view have argued that evidence for the interaction between cue competition and causal directionality may be restricted to two-phase blocking designs. Furthermore, from the viewpoint of causal-model theory, blocking designs carry the potential problem that the predicted asymmetries of cue competition are partly dependent on asymmetries of retrospective inferences. The present experiments use a one-phase overshadowing paradigm that does not allow for retrospective inferences and therefore represents a more unambiguous test of sensitivity to causal directionality. The results strengthen causal-model theory by clearly demonstrating the influence of causal directionality on learning. However, they also provide evidence for boundary conditions for this effect by highlighting the role of the semantics of the learning task.

Estimating causal strength: the role of structural knowledge and processing effort

The strength of causal relations typically must be inferred on the basis of statistical relations between observable events. This article focuses on the problem that there are multiple ways of extracting statistical information from a set of events. In causal structures involving a potential cause, an effect and a third related event, the assumed causal role of this third event crucially determines whether it is appropriate to control for this event when making causal assessments between the potential cause and the effect. Three experiments show that prior assumptions about the causal roles of the learning events affect the way contingencies are assessed with otherwise identical learning input. However, prior assumptions about causal roles is only one factor influencing contingency estimation. The experiments also demonstrate that processing effort affects the way statistical information is processed. These findings provide further evidence for the interaction between bottom-up and top-down influences in the acquisition of causal knowledge. They show that, apart from covariation information or knowledge about mechanisms, abstract assumptions about causal structures also may affect the learning process.

Beyond the Information Given: Causal Models in Learning and Reasoning

The philosopher David Humes conclusion that causal induction is solely based on observed associations still presents a puzzle to psychology. If we only acquired knowledge about statistical covariations between observed events without accessing deeper information about causality, we would be unable to understand the differences between causal and spurious relations, between prediction and diagnosis, and between observational and interventional inferences. All these distinctions require a deep understanding of causality that goes beyond the information given. We report a number of recent studies that demonstrate that people and rats do not stick to the superficial level of event covariations but reason and learn on the basis of deeper causal representations. Causal-model theory provides a unified account of this remarkable competence.

How temporal assumptions influence causal judgments

Causal learning typically entails the problem of being confronted with a large number of potentially relevant statistical relations. One type of constraint that may guide the choice of appropriate statistical indicators of causality are assumptions about temporal delays between causes and effects. There have been a few previous studies in which the role of temporal relations in the learning of events that are experienced in real time have been investigated. However, human causal reasoning may also be based on verbally described events, rather than on direct experiences of the events to which the descriptions refer. The aim of this paper is to investigate whether assumptions about the temporal characteristics of the events that are being described also affect causal judgment. Three experiments are presented that demonstrate that different temporal assumptions about causal delays may lead to dramatically different causal judgments, despite identical learning inputs. In particular, the experiments show that temporal assumptions guide the choice of appropriate statistical indicators of causality by structuring the event stream (Experiment 1), by selecting the potential causes among a set of competing candidates (Experiment 2), and by influencing the level of aggregation of events (Experiment 3).