The feature-matching approach pioneered by Amos Tversky remains a groundwork for psychological models of similarity and categorization but is rarely explicitly justified considering recent advances in thinking about cognition. While psychologists often view similarity as an unproblematic foundational concept that explains generalization and conceptual thought, long-standing philosophical problems challenging this assumption suggest that similarity derives from processes of higher-level cognition, including inference and conceptual thought. This paper addresses three specific challenges to Tversky’s approach: (i) the feature-selection problem, (ii) the problem (...) of cognitive implausibility, and (iii) the problem of unprincipled tweaking. It subsequently supports key insights from Tversky’s account based on recent developments in Bayesian modeling of cognition. A novel computational view of similarity as inference is proposed that addresses each challenge by considering the contrast class as constitutive of similarity and selecting for highly informative features. In so doing, this view illustrates the ongoing promise of the feature-matching approach in explaining perception, generalization and conceptual thought by grounding them in principles of probabilistic inference. (shrink)

This paper analyzes the rapid and unexpected rise of deep learning within Artificial Intelligence and its applications. It tackles the possible reasons for this remarkable success, providing candidate paths towards a satisfactory explanation of why it works so well, at least in some domains. A historical account is given for the ups and downs, which have characterized neural networks research and its evolution from “shallow” to “deep” learning architectures. A precise account of “success” is given, in order to sieve out (...) aspects pertaining to marketing or sociology of research, and the remaining aspects seem to certify a genuine value of deep learning, calling for explanation. The alleged two main propelling factors for deep learning, namely computing hardware performance and neuroscience findings, are scrutinized, and evaluated as relevant but insufficient for a comprehensive explanation. We review various attempts that have been made to provide mathematical foundations able to justify the efficiency of deep learning, and we deem this is the most promising road to follow, even if the current achievements are too scattered and relevant for very limited classes of deep neural models. The authors’ take is that most of what can explain the very nature of why deep learning works at all and even very well across so many domains of application is still to be understood and further research, which addresses the theoretical foundation of artificial learning, is still very much needed. (shrink)

In recent years, the family of algorithms collected under the term ``deep learning'' has revolutionized artificial intelligence, enabling machines to reach human-like performances in many complex cognitive tasks. Although deep learning models are grounded in the connectionist paradigm, their recent advances were basically developed with engineering goals in mind. Despite of their applied focus, deep learning models eventually seem fruitful for cognitive purposes. This can be thought as a kind of biological exaptation, where a physiological structure becomes applicable for a (...) function different from that for which it was selected. In this paper, it will be argued that it is time for cognitive science to seriously come to terms with deep learning, and we try to spell out the reasons why this is the case. First, the path of the evolution of deep learning from the connectionist project is traced, demonstrating the remarkable continuity, and the differences as well. Then, it will be considered how deep learning models can be useful for many cognitive topics, especially those where it has achieved performance comparable to humans, from perception to language. It will be maintained that deep learning poses questions that cognitive sciences should try to answer. One of such questions is the reasons why deep convolutional models that are disembodied, inactive, unaware of context, and static, are by far the closest to the patterns of activation in the brain visual system. (shrink)

How much would you pay to play a lottery with an “infinite expected payoff?” In the case of the century old, St. Petersburg Paradox, the answer is that the vast majority of people would only pay a small amount. The authors seek to understand this paradox by providing an explanation consistent with a broad, process‐level model of human decision‐making under risk.

We seem to learn and use concepts in a variety of heterogenous “formats”, including exemplars, prototypes, and theories. Different strategies have been proposed to account for this diversity. Hybridists consider instances in different formats to be instances of a single concept. Pluralists think that each instance in a different format is a different concept. Eliminativists deny that the different instances in different formats pertain to a scientifically fruitful kind and recommend eliminating the notion of a “concept” entirely. In recent years, (...) hybridism has received the most attention and support. However, we are still lacking a cognitive-computational model for concept representation and processing that would underpin hybridism. The aim of this paper is to advance the understanding of concepts by grounding hybridism in a neuroscientific model within the Predictive Processing framework. In the suggested view, the different formats are not distinct parts of a concept but arise from different ways of processing a functionally unified representational structure. (shrink)

By the time they reach early adulthood, English speakers are familiar with the meaning of thousands of words. In the last decades, computational simulations known as distributional semantic models have demonstrated that it is possible to induce word meaning representations solely from word co-occurrence statistics extracted from a large amount of text. However, while these models learn in batch mode from large corpora, human word learning proceeds incrementally after minimal exposure to new words. In this study, we run a set (...) of experiments investigating whether minimal distributional evidence from very short passages suffices to trigger successful word learning in subjects, testing their linguistic and visual intuitions about the concepts associated with new words. After confirming that subjects are indeed very efficient distributional learners even from small amounts of evidence, we test a DSM on the same multimodal task, finding that it behaves in a remarkable human-like way. We conclude that DSMs provide a convincing computational account of word learning even at the early stages in which a word is first encountered, and the way they build meaning representations can offer new insights into human language acquisition. (shrink)

In this paper, we explore interaction history as a particular source of pressure for achieving emergent compositional communication in multi-agent systems. We propose a training regime implementing template transfer, the idea of carrying over learned biases across contexts. In the presented method, a sender-receiver dyad is first trained with a disentangled pair of objectives, and then the receiver is transferred to train a new sender with a standard objective. Unlike other methods, the template transfer approach does not require imposing inductive (...) biases on the architecture of the agents. We experimentally show the emergence of compositional communication using topographical similarity, zero-shot generalization and context-independence as evaluation metrics. The presented approach is connected to an important line of work in semiotics and developmental psycholinguistics: it supports a conjecture that compositional communication is scaffolded on simpler communication protocols. (shrink)

Conviction Narrative Theory (CNT) is a theory of choice underradical uncertainty– situations where outcomes cannot be enumerated and probabilities cannot be assigned. Whereas most theories of choice assume that people rely on (potentially biased) probabilistic judgments, such theories cannot account for adaptive decision-making when probabilities cannot be assigned. CNT proposes that people usenarratives– structured representations of causal, temporal, analogical, and valence relationships – rather than probabilities, as the currency of thought that unifies our sense-making and decision-making faculties. According to CNT, (...) narratives arise from the interplay between individual cognition and the social environment, with reasoners adopting a narrative that feels “right” to explain the available data; using that narrative to imagine plausible futures; and affectively evaluating those imagined futures to make a choice. Evidence from many areas of the cognitive, behavioral, and social sciences supports this basic model, including lab experiments, interview studies, and econometric analyses. We identify 12 propositions to explain how the mental representations (narratives) interact with four inter-related processes (explanation, simulation, affective evaluation, and communication), examining the theoretical and empirical basis for each. We conclude by discussing how CNT can provide a common vocabulary for researchers studying everyday choices across areas of the decision sciences. (shrink)

In March 2016, DeepMind's computer programme AlphaGo surprised the world by defeating the world‐champion Go player, Lee Sedol. AlphaGo exhibits a novel, surprising and valuable style of play and has been recognised as “creative” by the artificial intelligence (AI) and Go communities. This article examines whether AlphaGo engages in creative problem solving according to the standards of comparative psychology. I argue that AlphaGo displays one important aspect of creative problem solving (namely mental scenario building in the form of Monte Carlo (...) tree search), while lacking another (domain generality). This analysis has consequences for how we think about creativity in humans and AI. (shrink)

Recent work in the cognitive sciences has argued that beliefs sometimes acquire signaling functions in virtue of their ability to reveal information that manipulates “mindreaders.” This paper sketches some of the evolutionary and design considerations that could take agents from solipsistic goal pursuit to beliefs that serve as social signals. Such beliefs will be governed by norms besides just the traditional norms of epistemology. As agents become better at detecting the agency of others, either through evolutionary history or individual learning, (...) the candidate pool for signaling expands. This logic holds for natural and artificial agents that find themselves in recurring social situations that reward the sharing of one’s thoughts. (shrink)

How are new Bayesian hypotheses generated within the framework of predictive processing? This explanatory framework purports to provide a unified, systematic explanation of cognition by appealing to Bayes rule and hierarchical Bayesian machinery alone. Given that the generation of new hypotheses is fundamental to Bayesian inference, the predictive processing framework faces an important challenge in this regard. By examining several cognitive‐level and neurobiological architecture‐inspired models of hypothesis generation, we argue that there is an essential difference between the two types of (...) models. Cognitive‐level models do not specify how they can be implemented in brains and include structures and assumptions that are external to the predictive processing framework. By contrast, neurobiological architecture‐inspired models, which aim to better resemble brain processes, fail to explain important capacities of cognition, such as categorization and few‐shot learning. The “scaling‐up” challenge for proponents of predictive processing is to explain the relationship between these two types of models using only the theoretical and conceptual machinery of Bayesian inference. (shrink)

Contemporary ethical analysis of Artificial Intelligence is growing rapidly. One of its most recognizable outcomes is the publication of a number of ethics guidelines that, intended to guide governmental policy, address issues raised by AI design, development, and implementation and generally present a set of recommendations. Here we propose two things: first, regarding content, since some of the applied issues raised by AI are related to fundamental questions about topics like intelligence, consciousness, and the ontological and ethical status of humans, (...) among others, the treatment of these issues would benefit from interfacing with neuroethics that has been addressing those same issues in the context of brain research. Second, the identification and management of some of the practical ethical challenges raised by AI would be enriched by embracing the methodological resources used in neuroethics. In particular, we focus on the methodological distinction between conceptual and action-oriented neuroethical approaches. We argue that the normative discussion about AI will benefit from further integration of conceptual analysis, including analysis of some operative assumptions, their meaning in different contexts, and their mutual relevance in order to avoid misplaced or disproportionate concerns and achieve a more realistic and useful approach to identifying and managing the emerging ethical issues. (shrink)

This paper attempts to answer a central question in unsupervised learning: what does it mean to “make sense” of a sensory sequence? In our formalization, making sense involves constructing a symbolic causal theory that both explains the sensory sequence and also satisfies a set of unity conditions. The unity conditions insist that the constituents of the causal theory – objects, properties, and laws – must be integrated into a coherent whole. On our account, making sense of sensory input is a (...) type of program synthesis, but it is unsupervised program synthesis. Our second contribution is a computer implementation, the Apperception Engine, that was designed to satisfy the above requirements. Our system is able to produce interpretable human-readable causal theories from very small amounts of data, because of the strong inductive bias provided by the unity conditions. A causal theory produced by our system is able to predict future sensor readings, as well as retrodict earlier readings, and impute (fill in the blanks of) missing sensory readings, in any combination. In fact, it is able to do all three tasks simultaneously. We tested the engine in a diverse variety of domains, including cellular automata, rhythms and simple nursery tunes, multi-modal binding problems, occlusion tasks, and sequence induction intelligence tests. In each domain, we test our engine's ability to predict future sensor values, retrodict earlier sensor values, and impute missing sensory data. The Apperception Engine performs well in all these domains, significantly out-performing neural net baselines. We note in particular that in the sequence induction intelligence tests, our system achieved human-level performance. This is notable because our system is not a bespoke system designed specifically to solve intelligence tests, but a general-purpose system that was designed to make sense of any sensory sequence. (shrink)

In the 1940s, Gilbert Ryle argued that knowing how to do something is not just a matter of being well-regulated but also a matter of self-regulation. Ryle appears to have thought that know-how requires self-regulation in both a backward-looking and forward-looking sense, but both ideas run counter to ordinary intuitions about know-how. The basic idea behind self-regulation, undertaking trials and adjusting to feedback, is captured by the “law of effect.” Daniel Dennett has argued that the “law of effect will not (...) go away”. After updating Dennett’s “Tower of Generate-and-Test” through a broad survey of the empirical literature on learning, I identify a Rylean intuition about intelligence and argue that self-regulation in a forward-looking sense is necessary for know-how. (shrink)

This paper is about guessing: how people respond to a question when they aren’t certain of the answer. Guesses show surprising and systematic patterns that the most obvious theories don’t explain. We argue that these patterns reveal that people aim to optimize a tradeoff between accuracy and informativity when forming their guess. After spelling out our theory, we use it to argue that guessing plays a central role in our cognitive lives. In particular, our account of guessing yields new theories (...) of belief, assertion, and the conjunction fallacy—the psychological finding that people sometimes rank a conjunction as more probable than one of its conjuncts. More generally, we suggest that guessing helps explain how boundedly rational agents like us navigate a complex, uncertain world. (shrink)

Generalization does not come from repeatedly observing phenomena in numerous settings, but from theories explaining what is general in those phenomena. Expecting future behavior to look like past observations is especially problematic in psychology, where behaviors change when people's knowledge changes. Psychology should thus focus on theories of people's capacity to create and apply new representations of their environments.

The topic of this dissertation is the nature of linguistic competence, the capacity to understand and produce sentences of natural language. I defend the empiricist account of linguistic competence embedded in the connectionist cognitive science. This strand of cognitive science has been opposed to the traditional symbolic cognitive science, coupled with transformational-generative grammar, which was committed to nativism due to the view that human cognition, including language capacity, should be construed in terms of symbolic representations and hardwired rules. Similarly, linguistic (...) competence in this framework was regarded as being innate, rule-governed, domain-specific, and fundamentally different from performance, i.e., idiosyncrasies and factors governing linguistic behavior. I analyze state-of-the-art connectionist, deep learning models of natural language processing, most notably large language models, to see what they can tell us about linguistic competence. Deep learning is a statistical technique for the classification of patterns through which artificial intelligence researchers train artificial neural networks containing multiple layers that crunch a gargantuan amount of textual and/or visual data. I argue that these models suggest that linguistic competence should be construed as stochastic, pattern-based, and stemming from domain-general mechanisms. Moreover, I distinguish syntactic from semantic competence, and I show for each the ramifications of the endorsement of a connectionist research program as opposed to the traditional symbolic cognitive science and transformational-generative grammar. I provide a unifying front, consisting of usage-based theories, a construction grammar approach, and an embodied approach to cognition to show that the more multimodal and diverse models are in terms of architectural features and training data, the stronger the case is for the connectionist linguistic competence. I also propose to discard the competence vs. performance distinction as theoretically inferior so that a novel and integrative account of linguistic competence originating in connectionism and empiricism that I propose and defend in the dissertation could be put forward in scientific and philosophical literature. (shrink)

Resource rationality may explain suboptimal patterns of reasoning; but what of “anti-Bayesian” effects where the mind updates in a direction opposite the one it should? We present two phenomena — belief polarization and the size-weight illusion — that are not obviously explained by performance- or resource-based constraints, nor by the authors’ brief discussion of reference repulsion. Can resource rationality accommodate them?

Lake et al. propose three criteria which, they argue, will bring artificial intelligence (AI) systems closer to human cognitive abilities. In this paper, we explore the application of these criteria to a particular domain of human cognition: our capacity for moral reasoning. In doing so, we explore a set of considerations relevant to the development of AI moral decision-making. Our main focus is on the relation between dual-process accounts of moral reasoning and model-free/model-based forms of machine learning. We also discuss (...) how work in embodied and situated cognition could provide a valu- able perspective on future research. (shrink)

In today’s society, it becomes increasingly important to assess which non-human and non-verbal beings possess consciousness. This review article aims to delineate criteria for consciousness especially in animals, while also taking into account intelligent artifacts. First, we circumscribe what we mean with “consciousness” and describe key features of subjective experience: qualitative richness, situatedness, intentionality and interpretation, integration and the combination of dynamic and stabilizing properties. We argue that consciousness has a biological function, which is to present the subject with a (...) multimodal, situational survey of the surrounding world and body, subserving complex decision-making and goal-directed behavior. This survey reflects the brain’s capacity for internal modeling of external events underlying changes in sensory state. Next, we follow an inside-out approach: how can the features of conscious experience, correlating to mechanisms inside the brain, be logically coupled to externally observable properties? Instead of proposing criteria that would each define a “hard” threshold for consciousness, we outline six indicators: goal-directed behavior and model-based learning; anatomic and physiological substrates for generating integrative multimodal representations; psychometrics and meta-cognition; episodic memory; susceptibility to illusions and multistable perception; and specific visuospatial behaviors. Rather than emphasizing a particular indicator as being decisive, we propose that the consistency amongst these indicators can serve to assess consciousness in particular species. The integration of scores on the various indicators yields an overall, graded criterion for consciousness, somewhat comparable to the Glasgow Coma Scale for unresponsive patients. When considering theoretically derived measures of consciousness, it is argued that their validity should not be assessed on the basis of a single quantifiable measure, but requires cross-examination across multiple pieces of evidence, including the indicators proposed here. Current intelligent machines, including deep learning neural networks and agile robots, are not indicated to be conscious yet. Instead of assessing machine consciousness by a brief Turing-type of test, evidence for it may gradually accumulate when we study machines ethologically and across time, considering multiple behaviors that require flexibility, improvisation, spontaneous problem-solving and the situational conspectus typically associated with conscious experience. (shrink)

What kinds of fundamental limits are there in how capable artificial intelligence (AI) systems might become? Two questions in particular are of interest: (1) How much more capable could AI become relative to humans, and (2) how easily could superhuman capability be acquired? To answer these questions, we will consider the literature on human expertise and intelligence, discuss its relevance for AI, and consider how AI could improve on humans in two major aspects of thought and expertise, namely simulation and (...) pattern recognition. We find that although there are very real limits to prediction, it seems like AI could still substantially improve on human intelligence. (shrink)