Classicists and connectionists alike claim to be able to explain systematicity. The proposed classicist explanation, I argue, is little more than a promissory note, one that classicists have no idea how to redeem. Smolensky's (1995) proposed connectionist explanation fares little better: it is not vulnerable to recent classicist objections, but it nonetheless fails, particularly if one requires, as some classicists do, that explanations of systematicity take the form of a‘functional analysis’. Nonetheless, there are, I argue, reasons for cautious optimism about (...) the prospects of a connectionist explanation. (shrink)

I propose to consider the question, "Can machines think?" This should begin with definitions of the meaning of the terms "machine" and "think." The definitions might be framed so as to reflect so far as possible the normal use of the words, but this attitude is dangerous, If the meaning of the words "machine" and "think" are to be found by examining how they are commonly used it is difficult to escape the conclusion that the meaning and the answer to (...) the question, "Can machines think?" is to be sought in a statistical survey such as a Gallup poll. But this is absurd. Instead of attempting such a definition I shall replace the question by another, which is closely related to it and is expressed in relatively unambiguous words. The new form of the problem can be described in terms of a game which we call the 'imitation game." It is played with three people, a man (A), a woman (B), and an interrogator (C) who may be of either sex. The interrogator stays in a room apart front the other two. The object of the game for the interrogator is to determine which of the other two is the man and which is the woman. He knows them by labels X and Y, and at the end of the game he says either "X is A and Y is B" or "X is B and Y is A." The interrogator is allowed to put questions to A and B. We now ask the question, "What will happen when a machine takes the part of A in this game?" Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman? These questions replace our original, "Can machines think?". (shrink)

What psychological and philosophical significance should we attach to recent efforts at computer simulations of human cognitive capacities? In answering this question, I find it useful to distinguish what I will call "strong" AI from "weak" or "cautious" AI. According to weak AI, the principal value of the computer in the study of the mind is that it gives us a very powerful tool. For example, it enables us to formulate and test hypotheses in a more rigorous and precise fashion. (...) But according to strong AI, the computer is not merely a tool in the study of the mind; rather, the appropriately programmed computer really is a mind, in the sense that computers given the right programs can be literally said to. (shrink)

Although machine learning has been successful in recent years and is increasingly being deployed in the sciences, enterprises or administrations, it has rarely been discussed in philosophy beyond the philosophy of mathematics and machine learning. The present contribution addresses the resulting lack of conceptual tools for an epistemological discussion of machine learning by conceiving of deep learning networks as ‘judging machines’ and using the Kantian analysis of judgments for specifying the type of judgment they are capable of. At the center (...) of the argument is the fact that the functionality of deep learning networks is established by training and cannot be explained and justified by reference to a predefined rule-based procedure. Instead, the computational process of a deep learning network is barely explainable and needs further justification, as is shown in reference to the current research literature. Thus, it requires a new form of justification, that is to be specified with the help of Kant’s epistemology. (shrink)

In this paper, I argue that that polysemy is a two-sided phenomenon. It can be reduced neither to pragmatic modulation nor to ambiguity, for it is a mixture of both. The senses of a polysemous expression result from pragmatic modulation but they are stored in memory, as the senses of an ambiguous expression are. The difference with straightforward ambiguity is that the modulation relations between the senses are transparent to the language users: the senses are felt as related – they (...) form a family of senses. In other words, whereas two homonymous expressions are different expressions, with the same phonological realization but distinct meanings, a polyseme is a single expression, i.e. a semantic as well as a phonological unit. It has one meaning, which should not be confused with the separate senses which it contributes in context. Different ways of thinking of that unitary meaning will be discussed, and consequences drawn for the debate between more or less radical versions of Contextualism. (shrink)

In this paper I argue that the search for explainable models and interpretable decisions in AI must be reformulated in terms of the broader project of offering a pragmatic and naturalistic account of understanding in AI. Intuitively, the purpose of providing an explanation of a model or a decision is to make it understandable to its stakeholders. But without a previous grasp of what it means to say that an agent understands a model or a decision, the explanatory strategies will (...) lack a well-defined goal. Aside from providing a clearer objective for XAI, focusing on understanding also allows us to relax the factivity condition on explanation, which is impossible to fulfill in many machine learning models, and to focus instead on the pragmatic conditions that determine the best fit between a model and the methods and devices deployed to understand it. After an examination of the different types of understanding discussed in the philosophical and psychological literature, I conclude that interpretative or approximation models not only provide the best way to achieve the objectual understanding of a machine learning model, but are also a necessary condition to achieve post hoc interpretability. This conclusion is partly based on the shortcomings of the purely functionalist approach to post hoc interpretability that seems to be predominant in most recent literature. (shrink)

In this paper I argue that the search for explainable models and interpretable decisions in AI must be reformulated in terms of the broader project of offering a pragmatic and naturalistic account of understanding in AI. Intuitively, the purpose of providing an explanation of a model or a decision is to make it understandable to its stakeholders. But without a previous grasp of what it means to say that an agent understands a model or a decision, the explanatory strategies will (...) lack a well-defined goal. Aside from providing a clearer objective for XAI, focusing on understanding also allows us to relax the factivity condition on explanation, which is impossible to fulfill in many machine learning models, and to focus instead on the pragmatic conditions that determine the best fit between a model and the methods and devices deployed to understand it. After an examination of the different types of understanding discussed in the philosophical and psychological literature, I conclude that interpretative or approximation models not only provide the best way to achieve the objectual understanding of a machine learning model, but are also a necessary condition to achieve post hoc interpretability. This conclusion is partly based on the shortcomings of the purely functionalist approach to post hoc interpretability that seems to be predominant in most recent literature. (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)

In this paper, I utilise the growing literature on scientific modelling to investigate the nature of formal semantics from the perspective of the philosophy of science. Specifically, I incorporate the inferential framework proposed by Bueno and Colyvan : 345–374, 2011) in the philosophy of applied mathematics to offer an account of how formal semantics explains and models its data. This view produces a picture of formal semantic models as involving an embedded process of inference and representation applying indirectly to linguistic (...) phenomena. The final aim of the paper is directed at proposing a novel account of the syntax–semantics interface while shedding light on empty categories, semantically null forms, underspecified content and compositionality as a whole. (shrink)

This paper attempts to describe and address a specific puzzle related to compositionality in artificial networks such as Deep Neural Networks and machine learning in general. The puzzle identified here touches on a larger debate in Artificial Intelligence related to epistemic opacity but specifically focuses on computational applications of human level linguistic abilities or properties and a special difficulty with relation to these. Thus, the resulting issue is both general and unique. A partial solution is suggested.

Because of the “all-or-none” character of nervous activity, neural events and the relations among them can be treated by means of propositional logic. It is found that the behavior of every net can be described in these terms, with the addition of more complicated logical means for nets containing circles; and that for any logical expression satisfying certain conditions, one can find a net behaving in the fashion it describes. It is shown that many particular choices among possible neurophysiological assumptions (...) are equivalent, in the sense that for every net behaving under one assumption, there exists another net which behaves under the other and gives the same results, although perhaps not in the same time. Various applications of the calculus are discussed. (shrink)

In this article the central philosophical issues concerning human-level artificial intelligence (AI) are presented. AI largely changed direction in the 1980s and 1990s, concentrating on building domain-specific systems and on sub-goals such as self-organization, self-repair, and reliability. Computer scientists aimed to construct intelligence amplifiers for human beings, rather than imitation humans. Turing based his test on a computer-imitates-human game, describing three versions of this game in 1948, 1950, and 1952. The famous version appears in a 1950 article in Mind, ‘Computing (...) Machinery and Intelligence’ (Turing 1950). The interpretation of Turing's test is that it provides an operational definition of intelligence (or thinking) in machines, in terms of behavior. ‘Intelligent Machinery’ sets out the thesis that whether an entity is intelligent is determined in part by our responses to the entity's behavior. Wittgenstein frequently employed the idea of a human being acting like a reliable machine. A ‘living reading-machine’ is a human being or other creature that is given written signs, for example Chinese characters, arithmetical symbols, logical symbols, or musical notation, and who produces text spoken aloud, solutions to arithmetical problems, and proofs of logical theorems. Wittgenstein mentions that an entity that manipulates symbols genuinely reads only if he or she has a particular history, involving learning and training, and participates in a social environment that includes normative constraints and further uses of the symbols. (shrink)

The application of deep learning methods to problems in natural language processing has generated significant progress across a wide range of natural language processing tasks. For some of these applications, deep learning models now approach or surpass human performance. While the success of this approach has transformed the engineering methods of machine learning in artificial intelligence, the significance of these achievements for the modelling of human learning and representation remains unclear. Deep Learning and Linguistic Representation looks at the application of (...) a variety of deep learning systems to several cognitively interesting NLP tasks. It also considers the extent to which this work illuminates our understanding of the way in which humans acquire and represent linguistic knowledge. Key Features: combines an introduction to deep learning in AI and NLP with current research on Deep Neural Networks in computational linguistics. is self-contained and suitable for teaching in computer science, AI, and cognitive science courses; it does not assume extensive technical training in these areas. provides a compact guide to work on state of the art systems that are producing a revolution across a range of difficult natural language tasks. (shrink)

Semantics is the discipline that studies linguistic meaning generally, and the qualification ‘formal’ indicates something about the sorts of techniques used in investigating linguistic meaning. More specifically, formal semantics is the discipline that employs techniques from symbolic logic, mathematics, and mathematical logic to produce precisely characterized theories of meaning for natural languages or artificial languages. Formal semantics as we know it first arose in the twentieth century. It was made possible by certain developments in logic during that period. This article (...) chronicles those developments and how they led to the development of formal semantics. (shrink)

In the early days of natural language semantics, Donald Davidson issued a challenge to those, like Richard Montague, who would do semantics in a model-theoretic framework that gives a central role to a model-relative notion of truth. Davidson argued that no theory of this kind can claim to be an account of real truth conditions unless it first makes clear how the relativized notion relates to our ordinary non-relativized notion of truth. In the 1990s, Davidson’s challenge was developed by Etchemendy (...) into an argument against the model-theoretic account of logical consequence, one that also threatens the attempt to capture natural language entailment relations in modeltheoretic terms—one of the central desiderata of semantics. Nevertheless, the modeltheoretic framework has flourished within natural language semantics. But it has flourished without any consensus among semanticists as to how to answer Davidson’s challenge. The aim of this essay is to develop an answer. I argue that model-theoretic semantics is best understood as model-based science: a semantics for a natural language is a scientific model of truth conditions. This makes good sense of the way model-theoretic tools are used in natural language semantics. And it allows us to answer Davidson’s challenge by showing how a theory that employs a relativized notion of truth manages to tell us about ordinary truth conditions. Moreover, I argue that it helps us see how semantics can provide genuine explanations for natural language entailment and other truth-conditional phenomena. (shrink)

Extant contextualist theories have relied on the mechanism of pragmatically driven modulation to explain the way non-indexical expressions take on different interpretations in different contexts. In this paper I argue that a modulation-based contextualist semantics is untenable with respect to non-ambiguous expressions whose invariant meaning fails to determine a unique literal interpretation, such as ‘lawyer’ ‘musician’ ‘book’ and ‘game’. The invariant meaning of such an expression corresponds to a range of closely related and equally basic interpretations, none of which can (...) be distinguished as the literal interpretation. Moreover, what counts as a literal interpretation as opposed to a non-literal one is arguably vague. The nonuniqueness of the literal interpretation and the vagueness in the literal/non-literal divide doubly challenge a modulation-based semantics, for modulation is supposed to operate on a unique literal interpretation to generate a clearly non-literal interpretation. Lastly I contend that non-ambiguous expressions which lack determinate literal interpretation are amenable to a Radical Contextualist semantics, according to which the invariant meaning of such an expression directs its interpretation to congruent background information in context. Thereby, these expressions exhibit semantically driven context sensitivity without displaying indexicality. (shrink)

This paper investigates connectionism's potential to solve the frame problem. The frame problem arises in the context of modelling the human ability to see the relevant consequences of events in a situation. It has been claimed to be unsolvable for classical cognitive science, but easily manageable for connectionism. We will focus on a representational approach to the frame problem which advocates the use of intrinsic representations. We argue that although connectionism's distributed representations may look promising from this perspective, doubts can (...) be raised about the potential of distributed representations to allow large amounts of complexly structured information to be adequately encoded and processed. It is questionable whether connectionist models that are claimed to effectively represent structured information can be scaled up to a realistic extent. We conclude that the frame problem provides a difficulty to connectionism that is no less serious than the obstacle it constitutes for classical cognitive science. (shrink)

This paper explores the difference between Connectionist proposals for cognitive a r c h i t e c t u r e a n d t h e s o r t s o f m o d e l s t hat have traditionally been assum e d i n c o g n i t i v e s c i e n c e . W e c l a i m t h a t t h (...) e m a j o r d i s t i n c t i o n i s t h a t , w h i l e b o t h Connectionist and Classical architectures postulate representational mental states, the latter but not the former are committed to a symbol-level of representation, or to a ‘language of thought’: i.e., to representational states that have combinatorial syntactic and semantic structure. Several arguments for combinatorial structure in mental representations are then reviewed. These include arguments based on the ‘systematicity’ of mental representation: i.e., on the fact that cognitive capacities always exhibit certain symmetries, so that the ability to entertain a given thought implies the ability to entertain thoughts with semantically related contents. We claim that such arguments make a powerful case that mind/brain architecture is not Connectionist at the cognitive level. We then consider the possibility that Connectionism may provide an account of the neural (or ‘abstract neurological’) structures in which Classical cognitive architecture is implemented. We survey a n u m b e r o f t h e s t a n d a r d a r g u m e n t s t h a t h a v e b e e n o f f e r e d i n f a v o r o f Connectionism, and conclude that they are coherent only on this interpretation. (shrink)

In this paper, I shall provide a novel response to the argument from context-sensitivity against truth-conditional semantics. It is often argued that the contextual influences on truth-conditions outstrip the resources of standard truth-conditional accounts, and so truth-conditional semantics rests on a mistake. The argument assumes that truth-conditional semantics is legitimate if and only if natural language sentences have truth-conditions. I shall argue that this assumption is mistaken. Truth-conditional analyses should be viewed as idealised approximations of the complexities of natural language (...) meaning. From this perspective, disparity between the scientific model and its real-world target is to be expected. I elaborate on what such an approach to semantics would look like. (shrink)

This paper offers both a theoretical and an experimental perspective on the relationship between connectionist and Classical (symbol-processing) models. Firstly, a serious flaw in Fodor and Pylyshyn’s argument against connectionism is pointed out: if, in fact, a part of their argument is valid, then it establishes a conclusion quite different from that which they intend, a conclusion which is demonstrably false. The source of this flaw is traced to an underestimation of the differences between localist and distributed representation. It has (...) been claimed that distributed representations cannot support systematic operations, or that if they can, then they will be mere implementations of traditional ideas. This paper presents experimental evidence against this conclusion: distributed representations can be used to support direct structure-sensitive operations, in a man- ner quite unlike the Classical approach. Finally, it is argued that even if Fodor and Pylyshyn’s argument that connectionist models of compositionality must be mere implementations were correct, then this would still not be a serious argument against connectionism as a theory of mind. (shrink)

Deep learning is currently the most prominent and widely successful method in artificial intelligence. Despite having played an active role in earlier artificial intelligence and neural network research, philosophers have been largely silent on this technology so far. This is remarkable, given that deep learning neural networks have blown past predicted upper limits on artificial intelligence performance—recognizing complex objects in natural photographs and defeating world champions in strategy games as complex as Go and chess—yet there remains no universally accepted explanation (...) as to why they work so well. This article provides an introduction to these networks as well as an opinionated guidebook on the philosophical significance of their structure and achievements. It argues that deep learning neural networks differ importantly in their structure and mathematical properties from the shallower neural networks that were the subject of so much philosophical reflection in the 1980s and 1990s. The article then explores several different explanations for their success and ends by proposing three areas of inquiry that would benefit from future engagement by philosophers of mind and science. (shrink)

Minimal Semantics asks what a theory of literal linguistic meaning is for - if you were to be given a working theory of meaning for a language right now, what would you be able to do with it? Emma Borg sets out to defend a formal approach to semantic theorising from a relatively new type of opponent - advocates of what she call 'dual pragmatics'. According to dual pragmatists, rich pragmatic processes play two distinct roles in linguistic comprehension: as well (...) as operating in a post-semantic capacity to determine the implicatures of an utterance, they also operate prior to the determination of truth-conditional content for a sentence. That is to say, they have an integral role to play within what is usually thought of as the semantic realm. Borg believes dual pragmatic accounts constitute the strongest contemporary challenge to standard formal approaches to semantics since they challenge the formal theorist to show not merely that there is some role for formal processes on route to determination of semantic content, but that such processes are sufficient for determining content. Minimal Semantics provides a detailed examination of this school of thought, introducing readers who are unfamiliar with the topic to key ideas like relevance theory and contextualism, and looking in detail at where these accounts diverge from the formal approach. Borg's defence of formal semantics has two main parts: first, she argues that the formal approach is most naturally compatible with an important and well-grounded psychological theory, namely the Fodorian modular picture of the mind. Then she argues that the main arguments adduced by dual pragmatists against formal semantics - concerning apparent contextual intrusions into semantic content - can in fact be countered by a formal theory. The defence holds, however, only if we are sensitive to the proper conditions of success for a semantic theory. Specifically, we should reject a range of onerous constraints on semantic theorizing (e.g., that it answer epistemic or metaphysical questions, or that it explain our communicative skills) and instead adopt a quite minimal picture of semantics. (shrink)

Compositionality has been a central concept in linguistics and philosophy for decades, and it is increasingly prominent in many other areas of cognitive science. Its status, however, remains contentious. Here, I reassess the nature and scope of the principle of compositionality (Partee, 1995) from the perspective of psycholinguistics and cognitive neuroscience. First, I review classic arguments for compositionality and conclude that they fail to establish compositionality as a property of human language. Next, I state a new competence argument, acknowledging the (...) fact that any competent user of a language L can assign to most expressions in L at least one meaning which is a functiononlyof the meanings of the expression's parts and of its syntactic structure. I then discuss selected results from cognitive neuroscience, indicating that the human brain possesses the processing capacities presupposed by the competence argument. Finally, I outline a language processing architecture consistent with the neuroscience results, where semantic representations may be generated by a syntax‐driven streamandby an “asyntactic” processing stream, jointly or independently. Compositionality is viewed as a constraint on computation in the former stream only. (shrink)

Computation is central to the foundations of modern cognitive science, but its role is controversial. Questions about computation abound: What is it for a physical system to implement a computation? Is computation sufficient for thought? What is the role of computation in a theory of cognition? What is the relation between different sorts of computational theory, such as connectionism and symbolic computation? In this paper I develop a systematic framework that addresses all of these questions. Justifying the role of computation (...) requires analysis of implementation, the nexus between abstract computations and concrete physical systems. I give such an analysis, based on the idea that a system implements a computation if the causal structure of the system mirrors the formal structure of the computation. This account can be used to justify the central commitments of artificial intelligence and computational cognitive science: the thesis of computational sufficiency, which holds that the right kind of computational structure suffices for the possession of a mind, and the thesis of computational explanation, which holds that computation provides a general framework for the explanation of cognitive processes. The theses are consequences of the facts that (a) computation can specify general patterns of causal organization, and (b) mentality is an organizational invariant, rooted in such patterns. Along the way I answer various challenges to the computationalist position, such as those put forward by Searle. I close by advocating a kind of minimal computationalism, compatible with a very wide variety of empirical approaches to the mind. This allows computation to serve as a true foundation for cognitive science. (shrink)