Martin Roth

The current debate over systematicity concerns the formal conditions a scheme of mental representation must satisfy in order to explain the systematicity of thought.1 The systematicity of thought is assumed to be a pervasive property of minds, and can be characterized (roughly) as follows: anyone who can think T can think systematic variants of T, where the systematic variants of T are found by permuting T’s constituents. So, for example, it is an alleged fact that anyone who can think the thought that John loves Mary can think the thought that Mary loves John, where the latter thought is a systematic variant of the former.

While it may be a datum of common sense that perceptual experiences can justify beliefs, there is no clear consensus about how they can do so. According to what I call “inferentialism,” perceptual experiences can justify beliefs because perceptual experiences have propositional contents and thus can serve as reasons for belief. A critical commitment of inferentialism is that justification requires the obtaining of a nonarbitrary or nonaccidental semantic relation between justifier and justified, a requirement that I call semantic appropriateness (SA). By contrast, reliabilists reject SA and argue that perceptual experiences can justify beliefs because perceptual experiences are part of a reliable belief‐forming process. In this paper, I explore whether a commitment to SA inevitably leads to a commitment to inferentialism. This exploration is largely motivated by doubts over whether perceptual experiences have propositional contents. If those doubts prove to be well‐founded, then it seems either that perceptual experiences cannot justify beliefs or that some form of reliabilism is true. I argue that although we should take the doubts seriously, there is a way to make sense of SA that does not require inferentialism.

Central to the dispute between symbolic AI and connectionism is the issue of whether cognitive capacities can be or should be explained by the execution of a program. Recently, connectionist models have been developed that seem to exhibit such capacities without executing a program. In my dissertation, I examine one such model and argue that it does execute a program. The argument proceeds by showing that what is essential to running a program is preserving the functional structure of the program. It has generally been assumed that this can only be done by the temporally organized causal capacities of the instantiating system. I observe that counterfactual-preserving functional architecture can be instantiated in other ways, for example geometrically, which are realizable by connectionist networks. I then discuss the consequences of this argument for the role programs play in psychological explanation. The explanation of cognitive capacities via the specification of the functional structure of a program is an instance of functional analysis, and lies at the heart of psychological explanation in symbolic AI. The program specifies steps whose exercise constitutes the exercise of a larger disposition of the system being analyzed. Since the functional structure specified by the program constitutes the analysis of the cognitive capacity, it is the functional structure that does the explanatory work. Because this functional structure can be realized in systems with radically different causal structures, the particular causal structure of a cognitive system is incidental to the analysis of cognitive capacities as such, and any causal structure which realizes the functional structure of a program counts as executing that program. Finally, I explore how this perspective on the relationship between connectionist networks and programs might bear on the issue of learning in connectionist networks. I describe how a dual network architecture might execute a learning program, thus showing how connectionist networks could embody the learning principles of symbolic AI. In this architecture, one network encodes stored knowledge while the other network executes the learning program itself. I then discuss how the interaction of these networks gives rise to the changes in behavior that constitute learning

The Knowledge Argument of Frank Jackson has not persuaded physicalists, but their replies have not dispelled the intuition that someone raised in a black and white environment gains genuinely new knowledge when she sees colors for the first time. In what follows, we propose an explanation of this particular kind of knowledge gain that displays it as genuinely new, but orthogonal to both physicalism and phenomenology. We argue that Mary’s case is an instance of a common phenomenon in which something new is learned as the result of exploiting representational resources that were not previously exploited, and that this results in gaining genuinely new information

There is a long-standing debate in the philosophy of action and the philosophy of science over folk psychological explanations of human action: do the (perhaps implicit) generalizations that underwrite such explanations purport to state contingent, empirically established connections between beliefs, desires, and actions, or do such generalizations serve rather to define, at least in part, what it is to have a belief or desire, or perform an action? This question has proven important because of certain traditional assumptions made about the role of law-statements in scientific explanations. According to this tradition, law-statements take the form of generalizations, and the laws we find in well-established sciences are contingent and empirical; as such, if the kinds of generalizations at work in folk psychological explanations of human action act like definitions, or state conceptual connections, then such generalizations could not play the kind of explanatory role we find in mature sciences. This paper argues that the aforementioned way of framing the debate reflects a still powerful but impoverished conception of the role laws play in scientific explanations, a conception that, moreover, cannot be reconciled with a good deal of actual scientific practice. When we update the philosophy of science, we find the concerns that are raised for folk psychological explanations largely evaporate or are found not to be specific to such explanations

It has been commonplace in epistemology since its inception to idealize away from computational resource constraints, i.e., from the constraints of time and memory. One thought is that a kind of ideal rationality can be specified that ignores the constraints imposed by limited time and memory, and that actual cognitive performance can be seen as an interaction between the norms of ideal rationality and the practicalities of time and memory limitations. But a cornerstone of naturalistic epistemology is that normative assessment is constrained by capacities: you cannot require someone to do something they cannot or, as it is usually put, ought implies can. This much we take to be uncontroversial. We argue that differences in architectures, goals and resources imply substantial differences in capacity, and that some of these differences are ineliminable. It follows that some differences in goals and architectural and computational resources matter at the normative level: they constrain what principles of normative epistemology can be used to describe and prescribe their behavior. As a result, we can expect there to be important epistemic differences between the way brains, individuals, and science work.

Recently, connectionist models have been developed that seem to exhibit structuresensitive cognitive capacities without executing a program. This paper examines one such model and argues that it does execute a program. The argument proceeds by showing that what is essential to running a program is preserving the functional structure of the program. It has generally been assumed that this can only be done by systems possessing a certain temporalcausal organization. However, counterfactualpreserving functional architecture can be instantiated in other ways, for example geometrically, which are realizable by connectionist networks

What are the prospects for a cognitive science of meaning? As stated, we think this question is ill posed, for it invites the conflation of several importantly different semantic concepts. In this paper, we want to distinguish the sort of meaning that is an explanandum for cognitive science—something we are going to call meaning—from the sort of meaning that is an explanans in cognitive science—something we are not going to call meaning at all, but rather content. What we are going to call meaning is paradigmatically a property of linguistic expressions or acts: what one’s utterance or sentence means, and what one means by it. What we are going to call content is a property of, among other things, mental representations and indicator signals. We will argue that it is a mistake to identify meaning with content, and that, once this is appreciated, some serious problems emerge for grounding meaning in the sorts of content that cognitive science is likely to provide.

This paper considers two ways functions figure into scientific explanations: (i) via laws?events are causally explained by subsuming those events under functional laws; and (ii) via designs?capacities are explained by specifying the functional design of a system. We argue that a proper understanding of how functions figure into design explanations of capacities makes it clear why such functions are ill-suited to figure into functional-cum-causal law explanations of events, as those explanations are typically understood. We further argue that a proper understanding of how functions enter into design explanations of capacities enables us to show why two prominent objections to functionalism in the philosophy mind?the argument from metaphysically necessary effects (Bennett, 2007; Rupert, 2006) and causal exclusion (Kim, 1993, 1998; Malcolm, 1968)?are misguided when interpreted as posing a threat to functional explanation in science across the board. If those arguments pose a threat at all, they pose it to instances of (i); however, a great number of the functional explanations we find in psychology?and the sciences generally?are instances of (ii)