Learning to search: From weak methods to domain-specific heuristics

Learning from experience involves three distinct components—generating behavior, assigning credit, and modifying behavior. We discuss these components in the context of learning search heuristics, along with the types of learning that can occur. We then focus on SAGE, a system that improves its search strategies with practice. The program is implemented as a production system, and learns by creating and strengthening rules for proposing moves. SAGE incorporates five different heuristics for assigning credit and blame, and employs a discrimination process to direct its search through the space of rules. The system has shown its generality by learning heuristics for directing search in six different task domains. In addition to improving its search behavior on practice problems, SAGE is able to transfer its expertise to scaled-up versions of a task, and in one case, transfers its acquired search strategy to problems with different initial and goal states.     

Mechanisms of theory formation in young children

Research suggests that by the age of five, children have extensive causal knowledge, in the form of intuitive theories. The crucial question for developmental cognitive science is how young children are able to learn causal structure from evidence. Recently, researchers in computer science and statistics have developed representations (causal Bayes nets) and learning algorithms to infer causal structure from evidence. Here we explore evidence suggesting that infants and children have the prerequisites for making causal inferences consistent with causal Bayes net learning algorithms. Specifically, we look at infants and children's ability to learn from evidence in the form of conditional probabilities, interventions and combinations of the two.     

Human causal induction: A glimpse at the whole picture

In the present work, most relevant evidence in causal learning literature is reviewed and a general cognitive architecture based on the available corpus of experimental data is proposed. However, contrary to algorithms formulated in the Bayesian nets framework, such architecture is not assumed to optimise the usefulness of the available information in order to induce the underlying causal structure as a whole. Instead, human reasoners seem to rely heavily on local clues and previous knowledge to discriminate between spurious and truly causal covariations, and piece those relations together only when they are demanded to do so. Bayesian networks and AI algorithms for causal inference are nonetheless considered valuable tools to identify the main computational goals of causal induction processes and to define the problems any intelligent causal inference system must solve.     

Identifying expectations about the strength of causal relationships

When we try to identify causal relationships, how strong do we expect that relationship to be? Bayesian models of causal induction rely on assumptions regarding people’s a priori beliefs about causal systems, with recent research focusing on people’s expectations about the strength of causes. These expectations are expressed in terms of prior probability distributions. While proposals about the form of such prior distributions have been made previously, many different distributions are possible, making it difficult to test such proposals exhaustively. In Experiment 1 we used iterated learning—a method in which participants make inferences about data generated based on their own responses in previous trials—to estimate participants’ prior beliefs about the strengths of causes. This method produced estimated prior distributions that were quite different from those previously proposed in the literature. Experiment 2 collected a large set of human judgments on the strength of causal relationships to be used as a benchmark for evaluating different models, using stimuli that cover a wider and more systematic set of contingencies than previous research. Using these judgments, we evaluated the predictions of various Bayesian models. The Bayesian model with priors estimated via iterated learning compared favorably against the others. Experiment 3 estimated participants’ prior beliefs concerning different causal systems, revealing key similarities in their expectations across diverse scenarios.     

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.     

Chunk hierarchies and retrieval structures: Comments on Saariluoma and Laine

The empirical results of Saariluoma and Laine are discussed and their computer simulations are compared with CHREST, a computational model of perception, memory and learning in chess. Mathematical functions such as power functions and logarithmic functions account for Saariluoma and Laine's correlation heuristic and for CHREST very well. However, these functions fit human data well only with game positions, not with random positions. As CHREST, which learns using spatial proximity, accounts for the human data as well as Saariluoma and Laine's correlation heuristic, their conclusion that frequency-based heuristics match the data better than proximity-based heuristics is questioned. The idea of flat chunk organisation and its relation to retrieval structures is discussed. In the conclusion, emphasis is given to the need for detailed empirical data, including information about chunk structure and types of errors, for discriminating between various learning algorithms.     

Proactive yet reflective? Materializing proactive personality into creativity through job reflective learning and activated positive affective states

Studies linking proactive personality to creativity have primarily taken a future-oriented perspective, describing a process where individuals assess future opportunities and risks of creative endeavors. Complementing this approach, we draw on an attribution theory perspective to delineate how proactive personality relates to employee creativity through the serial mediating effects of job reflective learning—a backward-looking cognitive process—and activated positive affective states. Job reflective learning captures backward-looking self-assessments and the underlying internal causal attributions, and it is differentiated into two valences: job reflective learning from successes and from failures. Based on two separate multi-wave, multi-source field studies, our findings consistently show a serial mediation process linking proactive personality to creativity through both valences of job reflective learning and activated positive affective states. Job reflective learning from successes breeds joviality, whereas job reflective learning from failures arouses attentiveness. Joviality and attentiveness—both types of activated positive affective states—in turn promote creativity. We discuss the theoretical and practical implications of how proactive employees manifest their proactivity trait into actual creativity through backward-looking cognitive and affective processes.     

SIMPLIFYING HEURISTICS VERSUS CAREFUL THINKING: SCIENTIFIC ANALYSIS OF MILLENNIAL SPIRITUAL ISSUES

There is ample evidence that humans (and other primates) possess a knowledge instinct—a biologically driven impulse to make coherent sense of the world at the highest level possible. Yet behavioral decision‐making data suggest a contrary biological drive to minimize cognitive effort by solving problems using simplifying heuristics. Individuals differ, and the same person varies over time, in the strength of the knowledge instinct. Neuroimaging studies suggest which brain regions might mediate the balance between knowledge expansion and heuristic simplification. One region implicated in primary emotional experience is more activated in individuals who use primitive heuristics, whereas two areas of the cortex are more activated in individuals with a strong knowledge drive: one region implicated in detecting risk or conflict and another implicated in generating creative ideas. Knowledge maximization and effort minimization are both evolutionary adaptations, and both are valuable in different contexts. Effort minimization helps us make minor and routine decisions efficiently, whereas knowledge maximization connects us to the beautiful, to the sublime, and to our highest aspirations. We relate the opposition between the knowledge instinct and heuristics to the biblical story of the fall, and argue that the causal scientific worldview is mathematically equivalent to teleological arguments from final causes. Elements of a scientific program are formulated to address unresolved issues.     

Singular Clues to Causality and Their Use in Human Causal Judgment

It is argued that causal understanding originates in experiences of acting on objects. Such experiences have consistent features that can be used as clues to causal identification and judgment. These are singular clues, meaning that they can be detected in single instances. A catalog of 14 singular clues is proposed. The clues function as heuristics for generating causal judgments under uncertainty and are a pervasive source of bias in causal judgment. More sophisticated clues such as mechanism clues and repeated interventions are derived from the 14. Research on the use of empirical information and conditional probabilities to identify causes has used scenarios in which several of the clues are present, and the use of empirical association information for causal judgment depends on the presence of singular clues. It is the singular clues and their origin that are basic to causal understanding, not multiple instance clues such as empirical association, contingency, and conditional probabilities.     

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.     

Category transfer in sequential causal learning: the unbroken mechanism hypothesis

The goal of the present set of studies is to explore the boundary conditions of category transfer in causal learning. Previous research has shown that people are capable of inducing categories based on causal learning input, and they often transfer these categories to new causal learning tasks. However, occasionally learners abandon the learned categories and induce new ones. Whereas previously it has been argued that transfer is only observed with essentialist categories in which the hidden properties are causally relevant for the target effect in the transfer relation, we here propose an alternative explanation, the unbroken mechanism hypothesis. This hypothesis claims that categories are transferred from a previously learned causal relation to a new causal relation when learners assume a causal mechanism linking the two relations that is continuous and unbroken. The findings of two causal learning experiments support the unbroken mechanism hypothesis.     

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.     

Generative connectionist networks and constructivist cognitive development

This article presents a novel computational framework for modeling cognitive development. The new modeling paradigm provides a language with which to compare and contrast radically different facets of children's knowledge. Concepts from the study of machine learning are used to explore the power of connectionist networks that construct their own architectures during learning. These so-called generative algorithms are shown to escape from Fodor's (1980) critique of Constructivist development. We describe one generative connectionist algorithm (cascade-correlation) in detail. We report on the successful use of the algorithm to model cognitive development on balance scale phenomena; seriation; the integration of velocity, time, and distance cues; prediction of effect sizes from magnitudes of causal potencies and effect resistances; and the acquisition of English personal pronouns. The article demonstrates that computer models are invaluable for illuminating otherwise obscure discussions.     

Domain-Specific Principles Affect Learning and Transfer in Children

In this paper I discuss the curious lack of contact between developmental psychologists studying the principles of early learning and those concentrating an later learning in children, where predispositions to learn certain types of concepts are less readily discussed. Instead, there is tacit agreement that learning and transfer mechanisms are content-independent and age-dependent. I argue here that one cannot study learning and transfer in a vacuum and that children's ability to learn is intimately dependent on what they are required to learn and the context in which they must learn it. Specifically, I argue that children learn and transfer readily, even in traditional laboratory settings, if they are required to extend their knowledge about causal mechanisms that they already understand. This point is illustrated in a series of studies with children from 1 to 3 years of age learning about simple mechanisms of physical causality (pushing-pulling, wetting, cutting, etc.). In addition, I document children's difficulty learning about causally impossible events, such as pulling with strings that do not appear to make contact with the object they are pulling. Even young children transfer on the basis of deep structural principles rather than perceptual features when they have access to the requisite domain-specific knowledge. I argue that a search for causal explanations is the basis of broad understanding, of wide patterns of generalization, and of flexible transfer and creative inferential projections—in sum, the essential elements of meaningful learning.     

Fast and frugal heuristics are plausible models of cognition: reply to Dougherty, Franco-Watkins, and Thomas (2008)

M. R. Dougherty, A. M. Franco-Watkins, and R. Thomas (2008) conjectured that fast and frugal heuristics need an automatic frequency counter for ordering cues. In fact, only a few heuristics order cues, and these orderings can arise from evolutionary, social, or individual learning, none of which requires automatic frequency counting. The idea that cue validities cannot be computed because memory does not encode missing information is misinformed; it implies that measures of co-occurrence are incomputable and would invalidate most theories of cue learning. They also questioned the recognition heuristic's psychological plausibility on the basis of their belief that it has not been implemented in a memory model, although it actually has been implemented in ACT-R (L. J. Schooler & R. Hertwig, 2005). On the positive side, M. R. Dougherty et al. discovered a new mechanism for a less-is-more effect. The authors of the present article specify minimal criteria for psychological plausibility, describe some genuine challenges in the study of heuristics, and conclude that fast and frugal heuristics are psychologically plausible: They use limited search and are tractable and robust.     

Learning geographical information from hypothetical maps

People show biases or distortions in their geographical judgments, such as mistakenly judging Rome to be south of Chicago (the Chicago-Rome illusion). These errors may derive from either perceptual heuristics or categorical organization. However, previous work on geographic knowledge has generally examined people's judgments of real-world locations for which learning history is unknown. This article reports experiments on the learning of hypothetical geographical spaces, in which participants acquired information in a fashion designed to control real-world factors, such as variable travel experiences or stereotypes about other countries, as well as to mimic initial encounters with locations through reading or conventional school-based geography education. Five experiments combine to suggest that biases in judgment based on learning of this kind are different in key regards from those seen with real-world geography and may be based more on the use of perceptual heuristics than on categorical organization.