Abstract Introduction Lumosity's Memory Match (LMM) is an online game requiring visual working memory. Change in LMM scores may be associated with individual differences in age‐related changes in working memory. Methods Effects of age and time on LMM learning and forgetting rates were estimated using data from 1890 game sessions for users aged 40 to 79 years. Results There were significant effects of age on baseline LMM scores (β = −.31, standard error or SE = .02, P < .0001) and lower learning rates (β = −.0066, SE = .0008, P < .0001). A sample size of 202 subjects/arm was estimated for a 1‐year study for subjects in the lower quartile of game performance. Discussion Online memory games have the potential to identify age‐related decline in cognition and to identify subjects at risk for cognitive decline with smaller sample sizes and lower cost than traditional recruitment methods.

Background A variety of studies have demonstrated gains in cognitive ability following cognitive training interventions. However, other studies have not shown such gains, and questions remain regarding the efficacy of specific cognitive training interventions. Cognitive training research often involves programs made up of just one or a few exercises, targeting limited and specific cognitive endpoints. In addition, cognitive training studies typically involve small samples that may be insufficient for reliable measurement of change. Other studies have utilized training periods that were too short to generate reliable gains in cognitive performance. Methods The present study evaluated an online cognitive training program comprised of 49 exercises targeting a variety of cognitive capacities. The cognitive training program was compared to an active control condition in which participants completed crossword puzzles. All participants were recruited, trained, and tested online (N = 4,715 fully evaluable participants). Participants in both groups were instructed to complete one approximately 15-minute session at least 5 days per week for 10 weeks. Results Participants randomly assigned to the treatment group improved significantly more on the primary outcome measure, an aggregate measure of neuropsychological performance, than did the active control group (Cohen’s d effect size = 0.255; 95% confidence interval = [0.198, 0.312]). Treatment participants showed greater improvements than controls on speed of processing, short-term memory, working memory, problem solving, and fluid reasoning assessments. Participants in the treatment group also showed greater improvements on self-reported measures of cognitive functioning, particularly on those items related to concentration compared to the control group (Cohen’s d = 0.249; 95% confidence interval = [0.191, 0.306]). Conclusion Taken together, these results indicate that a varied training program composed of a number of tasks targeted to different cognitive functions can show transfer to a wide range of untrained measures of cognitive performance. Trial Registration ClinicalTrials.gov NCT-02367898

A significant obstacle to developing effective treatments for Alzheimer's disease (AD) is the cost of clinical trials. Valid internet-based neuropsychological tests may reduce the costs to longitudinally assess cognitively normal subjects who are at risk for cognitive decline and to recruit these subjects into AD prevention trials. Memory Match (LMM) is an online working memory game developed by Lumos Labs. Although the validity of LMM scores as measures of working memory has not yet been evaluated, longitudinal trends in LMM scores may indicate declining performance due to aging or disease. LMM learning rates, forgetting rates, and changes in the learning rates over time can be estimated before subjects are enrolled in randomized studies and used to predict decline, thus increasing statistical power, reducing sample sizes, and lowering costs. With data provided by Lumos Labs, the effects of age and time on learning and forgetting rates were estimated with a mixed effects linear regression model. 2,212 Lumosity users (ages, 40 - 79) played forty LMM game sessions following ten run-in sessions for > 1 year. Sample sizes were calculated for 80% power to detect slowing the rate of decline in the learning rate by 25% in 1 year trials for subjects selected from the lowest quartile of learning rate change estimates (decliners). There were significant effects of age on lower initial LMM scores (β = -.39, P < .0001), lower initial learning rates (β = -.0031, P < .0001) and greater declines in learning rates over time (β = -8.00E-06, P < .001). Sample sizes as small as 136 subjects/arm were estimated for 1-year trials using subjects in the lower quartile of learning rate decline. The data suggest that declining learning rates are associated with older ages and that recruiting subjects in the lower quartile of learning rate decline significantly increases the statistical power to detect a treatment effect in clinical trials. As such, our data support the potential use of online memory games to identify subjects at risk for cognitive decline with smaller sample sizes and lower cost than traditional recruitment methods.

In recent years, a growing number of researchers have proposed that analogy is a core component of human cognition. According to the dominant theoretical viewpoint, analogical reasoning requires a specific suite of cognitive machinery, including explicitly coded symbolic representations and a mapping or binding mechanism that operates over these representations. Here we offer an alternative approach: we find that analogical inference can emerge naturally and spontaneously from a relatively simple, error-driven learning mechanism without the need to posit any additional analogy-specific machinery. The results also parallel findings from the developmental literature on analogy, demonstrating a shift from an initial reliance on surface feature similarity to the use of relational similarity later in training. Variants of the model allow us to consider and rule out alternative accounts of its performance. We conclude by discussing how these findings can potentially refine our understanding of the processes that are required to perform analogical inference.

Making new breakthroughs in understanding the processes underlying human cognition may depend on the availability of very large datasets that have not historically existed in psychology and neuroscience. Lumosity is a web-based cognitive training platform that has grown to include over 600 million cognitive training task results from over 35 million individuals, comprising the largest existing dataset of human cognitive performance. As part of the Human Cognition Project, Lumosity's collaborative research program to understand the human mind, Lumos Labs researchers and external research collaborators have begun to explore this dataset in order uncover novel insights about the correlates of cognitive performance. This paper presents two preliminary demonstrations of some of the kinds of questions that can be examined with the dataset. The first example focuses on replicating known findings relating lifestyle factors to baseline cognitive performance in a demographically diverse, healthy population at a much larger scale than has previously been available. The second example examines a question that would likely be very difficult to study in laboratory-based and existing online experimental research approaches at a large scale: specifically, how learning ability for different types of cognitive tasks changes with age. We hope that these examples will provoke the imagination of researchers who are interested in collaborating to answer fundamental questions about human cognitive performance.

How do humans learn contingencies between events? Both pathway-strengthening and inference-based process models have been proposed to explain contingency learning. We propose that each of these processes is used in different conditions. Participants viewed displays that contained single or paired objects and learned which displays were usually followed by the appearance of a dot. Some participants predicted whether the dot would appear before seeing the outcome, whereas other participants were required to respond quickly if the dot appeared shortly after the display. In the prediction task, instructions guiding participants to infer which objects caused the dot to appear were necessary in order for contingencies associated with one object to influence participants' predictions about the object with which it had been paired. In the response task, contingencies associated with one object affected responses to its pair mate irrespective of whether or not participants were given causal instructions. Our results challenge single-mechanism accounts of contingency learning and suggest that the mechanisms underlying performance in the two tasks are distinct.

On the Emergence of Analogical Inference Paul H. Thibodeau (pthibod1@stanford.edu) Stephen J. Flusberg (sflus@stanford.edu) Jeremy J. Glick (jjglick@stanford.edu) Daniel A. Sternberg (sternberg@stanford.edu) Department of Psychology, 450 Serra Mall, Bldg. 420 Stanford, CA 94305 USA Abstract What processes and mechanisms underlie analogical reasoning? In recent years, several computational models of analogy have been implemented to explore this question. One feature of many of these models is the assumption that humans possess dedicated analogy-specific cognitive machinery – for instance, a mapping or binding engine. In this paper, we question whether it is necessary to assume the existence of such machinery. We find that at least for some types of analogy, it is not. Instead, some forms of analogical processing emerge naturally and spontaneously from relatively simple, low-level learning mechanisms. We argue that this perspective is consistent with empirical findings from the developmental literature and with recent advances in cognitive neuroscience. Keywords: Analogy; metaphor; relational reasoning; development; connectionism; computational model. Introduction In the past three decades, there has been a growing appreciation for the possibility that analogy lies at the core of human cognition (Gentner, 1983; Hofstadter, 2001; Holyoak, Gentner, & Kokinov, 2001; Penn, Holyoak, & Povinelli, 2008). On this view, it is our ability to understand, produce, and reason with analogies that allows us to create the wonderfully rich and sophisticated intellectual and cultural worlds we inhabit. In an attempt to illuminate the cognitive mechanisms that underlie analogical processing, several detailed computational models have been developed that capture key components of the analogical reasoning process (see French, 2002 for a review). Among the most influential of these models are the Structure Mapping Engine (SME: Falkenhainer, Forbus, & Gentner, 1989), and Learning and Inference with Schemas and Analogies (LISA: Hummel & Holyoak, 1997). These models vary drastically in many ways; however, they share a fundamental commitment to explicitly structured symbolic or hybrid representations (e.g. of objects and relations), together with the existence of a dedicated analogical mapping or binding mechanism that operates over these representations. Indeed, proponents of these approaches argue that analogical inference is beyond the reach of models that lack these properties, including fully distributed connectionist models (e.g. Gentner & Markman, 1993; Holyoak & Hummel, 2000). While the structured approach has successfully captured adult behavior in numerous analogical reasoning tasks (e.g. Markman & Gentner, 1997; Hummel & Holyoak, 1997), it is unclear how this analogy-specific machinery comes to exist in the brain over the course of development. Even developmentally-oriented models such as DORA (Doumas, Hummel, & Sandhofer, 2008), which attempts to learn the structure used by LISA, assume a great deal of analogy- specific cognitive machinery without specifying how this machinery comes to exist in the first place. Here, we address this issue by proposing that some forms of analogical processing may emerge gradually over the course of development through the operation of low-level domain general learning mechanisms (Flusberg, Thibodeau, Sternberg, & Glick, 2010; Leech, Mareschal, & Cooper, 2008). In support of this view we describe a set of simulations carried out using the Rumelhart network (Rumelhart, 1990), a neurally inspired model that has succeeded in capturing many results from the literature on semantic development in children (e.g. Rogers & McClelland, 2004) and whose variants have been used to understand the deterioration of conceptual knowledge in semantic dementia (e.g. Dilkina, McClelland, & Plaut, Simulations Our learning task is inspired by Hinton’s (1986) family tree model, one of the first attempts to address relational learning in a connectionist network. The task of the model is to learn “statements” that are true about the various members of a family, including identity information, perceptual features, and relations between family members. Input to the model consists of activating a Subject unit, corresponding to a particular family member, and a Relation unit. The Relation units correspond to the different kinds of relationships that can hold between subjects and objects (e.g. “is_named”, “parent_of”). The network is wired up in a strictly feed-forward fashion, as shown in Figure 1, such that the input propagates forward through the internal layers, resulting in a set of predictions over the Object layer. Over the course of training, the network’s weights change (via backpropagation of the cross-entropy error on the output units) in order to better predict which Object outputs correspond to each combination of Subject and Relation inputs. As the model also contains intervening layers of units between the input and output layers, it is forced to re-represent the inputs as a distributed pattern of activation over these internal layers.

How do humans learn contingencies between events? Several types of process models have been proposed, including pathway strengthening and inference-based models. We propose that each of these processes is used in different task conditions. Human participants viewed displays containing single or paired objects and learned which displays were usually followed by a dot. Some participants predicted whether the dot would appear and then saw the outcome, while others were required to respond quickly if the dot appeared shortly after the objects. For predict participants, instructions guiding participants to infer which objects had the power to cause the outcome determined whether contingencies associated with one object affected predictions about its pair mate. For respond participants, contingencies associated with one object affected responses to the mate, whether or not independent these instructions were provided. The results challenge single-mechanism accounts and support the proposal that the mechanisms underlying performance in the two tasks are distinct. TWO MECHANISMS OF CONTINGENCY LEARNING 3" Two Mechanisms of Human Contingency Learning Understanding how people learn contingencies between events has been a focus of research for many years (Krechevsky, 1932; Pavlov, 1927; Tolman, 1948, 1949). In standard contingency learning tasks, participants view situations in which cues are followed by outcomes, and are later asked to predict outcomes for test cases. Two kinds of accounts have been offered to describe the process underlying performance in such tasks. One type of account is based on strengthening of pathways linking representations of cues to representations of outcomes or responses (Rescorla & Wagner, 1972; Pearce & Hall, 1980). The other is based on an explicit reasoning process that leads to inferences about the causal relations between the cues and outcomes in light of evidence (De Houwer, 2009; Mitchell, De Houwer & Lovibond, 2009). Pathway strengthening has been proposed as a mechanism for gradually learning contingent response tendencies. Stronger pathways promote fast, automatic responding (Cohen, Dunbar & McClelland, 1990), and pathway strengthening models can account for the gradual speeding of contingency-sensitive responding in fast-paced sequence learning tasks (Cleeremans & McClelland, 1991). On the other hand, considerable evidence now supports accounts of contingency learning that rely on a resource-intensive process of making explicit inferences in many situations, leading some to propose that a complete account of contingency learning is possible based only on explicit inference-based process (De Houwer, 2009, Mitchell, De Houwer & Lovibond, 2009). We suggest that both processes may be at work, depending on the task situation. In support of our view, we rely on a difference in the predictions the two accounts make about how instructions should influence what we call indirect effects in contingency

A growing body of data has been gathered in support of the view that the mind is embodied and that cognition is grounded in sensory-motor processes. Some researchers have gone so far as to claim that this paradigm poses a serious challenge to central tenets of cognitive science, including the widely held view that the mind can be analyzed in terms of abstract computational principles. On the other hand, computational approaches to the study of mind have led to the development of specific models that help researchers understand complex cognitive processes at a level of detail that theories of embodied cognition (EC) have sometimes lacked. Here we make the case that connectionist architectures in particular can illuminate many surprising results from the EC literature. These models can learn the statistical structure in their environments, providing an ideal framework for understanding how simple sensory-motor mechanisms could give rise to higher-level cognitive behavior over the course of learning. Crucially, they form overlapping, distributed representations, which have exactly the properties required by many embodied accounts of cognition. We illustrate this idea by extending an existing connectionist model of semantic cognition in order to simulate findings from the embodied conceptual metaphor literature. Specifically, we explore how the abstract domain of time may be structured by concrete experience with space (including experience with culturally-specific spatial and linguistic cues). We suggest that both EC researchers and connectionist modelers can benefit from an integrated approach to understanding these models and the empirical findings they seek to explain.

How do we learn causal relations between events from experience? Many have argued for an associative account inspired by animal conditioning models, but there is a growing literature arguing that indirect effects in contingency learning depend on explicit cognitive processes. Our experiments explore the basis of two such effects: blocking and screening off. In Experiment 1, we gave participants an untimed explicit prediction task to replicate standard findings in the contingency learning literature in a novel domain. We obtained robust indirect effects when participants had a causal framework to constrain their reasoning. In Experiment 2, we reduced the time available for explicit recollection by reconstructing the task as a fast-paced RT task. Participants continued to show robust learning of direct relationships, as measured by response times, but there were no indirect effects. Experiment 3 followed up on whether participants in our RT task would produce indirect effects through explicit processes when given an opportunity to make a more deliberative prediction at test.

When Should We Expect Indirect Effects in Human Contingency Learning? Daniel A. Sternberg (sternberg@stanford.edu) and James L. McClelland (mcclelland@stanford.edu) Department of Psychology, Stanford University Stanford, CA 94305 USA Abstract experiments. In these experiments, participants see a number of pairings of cues and outcomes during training. At test, they are asked to rate the various cues’ causal strengths or to make predictions about the likely outcomes for each cue. Early contingency learning researchers such as Alloy and Abramson (1979) and Dickinson and colleagues (1984) compared their findings to models of animal conditioning that automatically generate indirect effects (e.g., Rescorla & Wagner, 1972; Pearce & Hall, 1980). Indeed, a large class of error-correcting learning algorithms predicts these effects (Rosenblatt, 1958; Rumelhart et al., 1986; Sutton, 1988). Recent dual process models of implicit and explicit learning have employed error-correcting learning algorithms in the implicit component of the models (e.g., Ashby et al., 1998; Sun et al., 2005) – suggesting that indirect effects should be a basic outcome of an implicit learning system. How do we learn causal relations between events from experience? Many have argued for an associative account inspired by animal conditioning models, but there is a growing literature arguing that indirect effects in contingency learning depend on explicit cognitive processes. Our experiments explore the basis of two such effects: blocking and screening off. In Experiment 1, we gave participants an untimed explicit prediction task to replicate standard findings in the contingency learning literature in a novel domain. We obtained robust indirect effects when participants had a causal framework to constrain their reasoning. In Experiment 2, we reduced the time available for explicit recollection by reconstructing the task as a fast-paced RT task. Participants continued to show robust learning of direct relationships, as measured by response times, but there were no indirect effects. Experiment 3 followed up on whether participants in our RT task would produce indirect effects through explicit processes when given an opportunity to make a more deliberative prediction at test. Table 1: An example of direct and indirect effects in a contingency learning paradigm. Keywords: Learning; causal reasoning; implicit learning Training Blocking pair Screening pair Direct effect Indirect effect Introduction A child goes out to dinner with his family and at the end of the meal experiences a strong allergic reaction. Upon discussion with the restaurant manager, the child’s parents learn that the sauce for his entree contained shrimp, and peanuts were used in his dessert. Suppose the child has never had shrimp before. If he has had a history of peanut allergies, one may be inclined to attribute the allergy to the peanuts; if he had never had a peanut allergy before, one may be more inclined to suspect an allergy to the shrimp. We can consider the child’s previous experience with peanuts as the direct evidence about whether peanuts cause an allergic reaction. This evidence, together with the shrimp-and-peanuts event, provides indirect evidence about whether shrimp causes one. If peanuts had previously caused an allergy, this tends to block the inference that shrimp causes one; if peanuts had not previously caused an allergy, this tends to screen off the shrimp – increasing the likelihood of this inference. Comparing the two cases, the scenario above describes a direct effect whereby the strength of the perceived causal relation between peanuts and allergy should be higher for the blocking pair compared to the screening pair. It also describes an indirect effect whereby the strength for shrimp will be higher in the screening pair compared to the blocking pair. Table 1 encapsulates this information. Effects similar to the indirect effect described above have often been demonstrated in contingency learning Single item Pair B 1 + B 1 B 2 + S 1 - S 1 S 2 + B 1 > S 1 S 2 > B 2 Complicating the error-driven account have been findings of retrospective effects like backward blocking (Shanks, 1985), where the order of compound and single item events are reversed (e.g., shrimp and peanuts before peanuts alone). These models do not directly predict retrospective effects. Various modifications to the error- correcting learning algorithm have been proposed to accommodate retrospective effects (Van Hamme & Wasserman, 1994; Dickinson & Burke, 1996). These models continue to predict indirect effects as a basic outcome of the learning process. Another approach has been to argue that retrospective effects are instead driven by the explicit retrieval of memories for previously experienced events (McClelland & Thompson, 2008). More troubling are recent findings that suggest indirect effects are often quite fragile in contingency learning tasks. De Houwer and Beckers (2003) found that blocking was attenuated when participants were given a relatively difficult secondary task (discriminating between a high and low tone) during training and test phases. “High- level” constraints such as assumptions that cues are additive in their effects also appear to modulate the size of indirect effects (Lovibond et al, 2003; Beckers et al., 2005; cf. Livesey & Boakes, 2004). These findings have led some to argue that an explicit propositional reasoning

Mitchell et al. describe many fascinating studies, and in the process, propose what they consider to be a unified framework for human learning in which effortful, controlled learning results in propositional knowledge. However, it is unclear how any of their findings privilege a propositional account, and we remain concerned that embedding all knowledge in propositional representations obscures the tight interdependence between learning from experiences and the use of the results of learning as a basis for action.

Simultaneous acquisition of multiple languages to a native level of fluency is common in many areas of the world. This ability must be represented in any cognitive mechanisms used for language. Potential explanations of the evolution of language must also account for the bilingual case. Surprisingly, this fact has not been widely considered in the literature on language origins and evolution. We consider any array of potential accounts for this phenomenon, including arguments by selectionists on the basis for language variation. We find scant evidence for specific selection of the multilingual ability prior to language origins. Thus it seems more parsimonious that bilingualism came for free along with whatever mechanisms did evolve. Sequential learning mechanisms may be able to accomplish multilingual acquisition without specific adaptations. In support of this perspective, we present a simple recurrent network model that is capable of learning two idealized grammars simultaneously. These results are compared with recent studies of bilingual processing using eyetracking and fMRI showing vast overlap in the areas in the brain used in processing two different languages.