Catherine Stinson

This chapter explores some of the ways that mechanisms are invoked in neuroscience and looks at a selection of the philosophical problems that arise when trying to understand mechanistic explanations. It introduces a series of historical case studies that illustrate how neuroscientists have depended on mechanistic metaphors in their efforts to understand the mind and brain, and how their mechanistic explanations have developed over time. The chapter highlights what contemporary philosophers have identified as the fundamental features of mechanisms and mechanistic explanation. It considers some of the methodological issues that arise in neuroscience including how to integrate psychological with neural models, how to generalize findings in model organisms like the sea slug Aplysia to human learning and memory, and whether to favor top-down or bottom-up methods. Constructing multi-level mechanistic explanations involves intensive collaboration across different branches of science, and involves many challenges, both pragmatic and methodological.

This chapter explores the epistemic roles played by connectionist models of cognition, and offers a formal analysis of how connectionist models explain. It looks at how other types of computational models explain. Classical artificial intelligence (AI) programs explain using abductive reasoning, or inference to the best explanation; they begin with the phenomena to be explained, and devise rules that can produce the right outcome. The chapter also looks at several examples of connectionist models of cognition, observing what sorts of constraints are used in their design, and how their results are evaluated. It argues that the point of implementing networks roughly analogous to neural structures is to discover and explores the generic mechanisms at work in the brain, not to deduce the precise activities of specific structures. The chapter explores a formal analysis of the explanations offered, which interprets connectionist models and the cognitive theories they represent as sharing membership in a type of mechanism.

When Artificial Intelligence (AI) is applied in decision-making that affects people’s lives, it is now well established that the outcomes can be biased or discriminatory. The question of whether algorithms themselves can be among the sources of bias has been the subject of recent debate among Artificial Intelligence researchers, and scholars who study the social impact of technology. There has been a tendency to focus on examples, where the data set used to train the AI is biased, and denial on the part of some researchers that algorithms can also be biased. Here we illustrate the point that algorithms themselves can be the source of bias with the example of collaborative filtering algorithms for recommendation and search. These algorithms are known to suffer from cold-start, popularity, and homogenizing biases, among others. While these are typically described as statistical biases rather than biases of moral import; in this paper we show that these statistical biases can lead directly to discriminatory outcomes. The intuitive idea is that data points on the margins of distributions of human data tend to correspond to marginalized people. The statistical biases described here have the effect of further marginalizing the already marginal. Biased algorithms for applications such as media recommendations can have significant impact on individuals’ and communities’ access to information and culturally-relevant resources. This source of bias warrants serious attention given the ubiquity of algorithmic decision-making.

Diversity is often announced as a solution to ethical problems in artificial intelligence (AI), but what exactly is meant by diversity and how it can solve those problems is seldom spelled out. This lack of clarity is one hurdle to motivating diversity in AI. Another hurdle is that while the most common perceptions about what diversity is are too weak to do the work set out for them, stronger notions of diversity are often defended on normative grounds that fail to connect to the values that are important to decision-makers in AI. However, there is a long history of research in feminist philosophy of science and a recent body of work in social epistemology that taken together provide the foundation for a notion of diversity that is both strong enough to do the work demanded of it, and can be defended on epistemic grounds that connect with the values that are important to decision-makers in AI. We clarify and defend that notion here by introducing emergent expertise as a network phenomenon wherein groups of workers with expertise of different types can gain knowledge not available to any individual alone, as long as they have ways of communicating across types of expertise. We illustrate the connected epistemic and ethical benefits of designing technology with diverse groups of workers using the examples of an infamous racist soap dispenser, and the millimeter wave scanners used in US airport security.

Mechanism schemata are one of the least understood parts of MDC’s account of mechanistic explanation. Relatedly, there is a common misconception that there is no place for abstraction in MDC mechanisms. These two problems can be remedied by looking more carefully at what MDC say both in their 2000 paper and elsewhere about schemata and abstraction, and by following up on a comment of Machamer’s indicating that Piaget was the inspiration for schemata. Darden’s work on mechanism discovery reveals an important role for abstract mechanism schemata, not only as concise representations of known mechanisms, but also as hypotheses for how unknown mechanisms might work. Piaget’s schemata likewise both capture the developing child’s current state of knowledge, and serve as templates into which new experiences are fit. If we assume that mechanistic explanations should pick out real difference-makers, and that what makes a difference needn’t always be fine-grained details, we open up an explanatory role for abstract schemata to play. Schemata represent the coarser-grained features of mechanisms that are often among the most important difference-makers. In many contexts they might need to be supplemented with finer grained details in order to give an adequate explanation, but independent of those details, schemata remain explanatory in virtue of being part of the explanation. Detailed accounts of real difference-makers need not be seen as in competition with abstract models or generalizations. They can be combined in mechanistic explanations.

There is a vast literature within philosophy of mind that focuses on artificial intelligence, but hardly mentions methodological questions. There is also a growing body of work in philosophy of science about modeling methodology that hardly mentions examples from cognitive science. Here these discussions are connected. Insights developed in the philosophy of science literature about the importance of idealization provide a way of understanding the neural implausibility of connectionist networks. Insights from neurocognitive science illuminate how relevant similarities between models and targets are picked out, how modeling inferences are justified, and the metaphysical status of models.

Discussions of psychiatric nosology focus on a few popular examples of disorders, and on the validity of diagnostic criteria. Looking at Anorexia Nervosa, an example rarely mentioned in this literature, reveals a new problem: the DSM has a strict taxonomic structure, which assumes that disorders can only be located on one branch. This taxonomic assumption fails to fit the domain of psychopathology, resulting in obfuscation of cross-category connections. Poor outcomes for treatment of Anorexia may be due to it being pigeonholed as an Eating Disorder, when a disturbance of body perception may be a more central symptom than food restriction. This paper explores the possibility of restructuring the DSM taxonomy to allow for a pluralist classification of disorders. This change could improve treatment and research without requiring any changes to diagnostic criteria.

William James presaged, and Alan Allport voiced criticisms of cause theories of executive attention for involving a homunculus who directs attention. I review discussions of this problem, and argue that existing philosophical denials of the problem depend on equivocations between different senses of “Cartesian error”. Another sort of denial tries to get around the problem by offering empirical evidence that such an executive attention director exists in prefrontal cortex. I argue that the evidence does not warrant the conclusion that an executive director can be localized in prefrontal cortex unless dubious assumptions are made, and that computational models purporting to support these assumptions either beg the question, or fail to model executive attention in terms of cause theories.

Recent extensions of mechanistic explanation into psychology suggest that cognitive models are only explanatory insofar as they map neatly onto, and serve as scaffolding for more detailed neural models. Filling in those neural details is what these accounts take the integration of cognitive psychology and neuroscience to mean, and they take this process to be seamless. Critics of this view have given up on cognitive models possibly explaining mechanistically in the course of arguing for cognitive models having explanatory value independent of how well they align with neural mechanisms. We can have things both ways, however. The problem with seamless integration accounts is their seamlessness, not that they take cognitive models to be mechanistic. A non-componential view of mechanisms allows for cognitive and neural models that cross cut one another, and for cognitive models that don’t decompose into parts. I illustrate the inadequacy of seamless accounts of integration by contrasting how “filter” models of attention in psychology and of sodium channels in neuroscience developed; by questioning whether the mappings generated by neuroimaging subtraction studies achieve integration; and by reinterpreting the evidence for cognitive models of memory having been successfully integrated with neural models. I argue that the integrations we can realistically expect are more partial, patchy, and full of loose threads than the mosaic unity Craver describes