This paper presents the main topics, arguments, and positions in the philosophy of AI at present (excluding ethics). Apart from the basic concepts of intelligence and computation, the main topics of ar-tificial cognition are perception, action, meaning, rational choice, free will, consciousness, and normativity. Through a better understanding of these topics, the philosophy of AI contributes to our understand-ing of the nature, prospects, and value of AI. Furthermore, these topics can be understood more deeply through the discussion of AI; so (...) we suggest that “AI Philosophy” provides a new method for philosophy. (shrink)

How feasible are conceptual engineering projects of social concepts that aim for the engineered concept to be widely adopted in ordinary everyday life? Predominant frameworks on the psychology of concepts that shape work on stereotyping, bias, and machine learning have grim implications for the prospects of conceptual engineers: conceptual engineering efforts are ineffective in promoting certain social-conceptual changes. Specifically, since conceptual components that give rise to problematic social stereotypes are sensitive to statistical structures of the environment, purely conceptual change won’t (...) be possible without corresponding world change. This tradition, however, tends to ignore that concepts don’t only encode statistical, but also causal information. Paying attention to this feature of concepts, I argue, shows that conceptual engineering is not only possible. There is an imperative to conceptually-engineer. (shrink)

Necessity and sufficiency are the building blocks of all successful explanations. Yet despite their importance, these notions have been conceptually underdeveloped and inconsistently applied in explainable artificial intelligence, a fast-growing research area that is so far lacking in firm theoretical foundations. In this article, an expanded version of a paper originally presented at the 37th Conference on Uncertainty in Artificial Intelligence, we attempt to fill this gap. Building on work in logic, probability, and causality, we establish the central role of (...) necessity and sufficiency in XAI, unifying seemingly disparate methods in a single formal framework. We propose a novel formulation of these concepts, and demonstrate its advantages over leading alternatives. We present a sound and complete algorithm for computing explanatory factors with respect to a given context and set of agentive preferences, allowing users to identify necessary and sufficient conditions for desired outcomes at minimal cost. Experiments on real and simulated data confirm our method’s competitive performance against state of the art XAI tools on a diverse array of tasks. (shrink)

Recent studies indicate that indicative conditionals like "If people wear masks, the spread of Covid-19 will be diminished" require a probabilistic dependency between their antecedents and consequents to be acceptable (Skovgaard-Olsen et al., 2016). But it is easy to make the slip from this claim to the thesis that indicative conditionals are acceptable only if this probabilistic dependency results from a causal relation between antecedent and consequent. According to Pearl (2009), understanding a causal relation involves multiple, hierarchically organized conceptual dimensions: (...) prediction, intervention, and counterfactual dependence. In a series of experiments, we test the hypothesis that these conceptual dimensions are differentially encoded in indicative and counterfactual conditionals. If this hypothesis holds, then there are limits as to how much of a causal relation is captured by indicative conditionals alone. Our results show that the acceptance of indicative and counterfactual conditionals can become dissociated. Furthermore, it is found that the acceptance of both is needed for accepting a causal relation between two co-occurring events. The implications that these findings have for the hypothesis above, and for recent debates at the intersection of the psychology of reasoning and causal judgment, are critically discussed. Our findings are consistent with viewing indicative conditionals as answering predictive queries requiring evidential relevance (even in the absence of direct causal relations). Counterfactual conditionals in contrast target causal relevance, specifically. Finally, we discuss the implications our results have for the yet unsolved question of how reasoners succeed in constructing causal models from verbal descriptions. (shrink)

This paper aims to bridge philosophical and psychological research on causation, counterfactual thought, and the problem of backtracking. Counterfactual approaches to causation such as that by Lewis have ruled out backtracking, while on prominent models of causal inference interventionist counterfactuals do not backtrack. However, on various formal models, certain backtracking counterfactuals end up being true, and psychological evidence shows that people do sometimes backtrack when answering counterfactual questions in causal contexts. On the basis of psychological research, I argue that while (...) ordinarily both kinds of counterfactuals may be employed, non-backtracking counterfactuals are more easily used in causal inference because they are consistent with temporal order information embedded in the mental simulation heuristic, and they match reasoners’ experience of causation. While this approach is incompatible with the ambitions of counterfactual theories that seek to establish the non-backtracking interpretation as the only legitimate one, it can provide support for perspectival views on causation and open further inquiry on the functions of causal and counterfactual thought in the context of causal models. (shrink)

How can we best understand human cognitive architectural variability? We believe that the relationships between theories in neurobiology, cognitive science and evolutionary biology posited by evolutionary psychology’s Integrated Causal Model has unduly supported various essentialist conceptions of the human cognitive architecture, monomorphic minds, that mask HCA variability, and we propose a different set of relationships between theories in the same domains to support a different, non-essentialist, understanding of HCA variability. To set our case against essentialist theories of HCA variability, we (...) detail the general notion of an ICM and the specific ICM at the heart of evolutionary psychology. We briefly illustrate the type of essentialism fostered by evolutionary psychology’s ICM by showing how it grounds essentialist theories of cognitive gender. We shall not criticize these theories here since the literature is replete with compelling objections to them, but shall instead focus on motivating a replacement ICM to destabilize evolutionary psychology’s ICM wholesale. ICMs usually span larger than the models they support, hence larger than arguments against these models, and one reason the essentialist theories addressed here have the kind of staying power they do is that they are partly supported by the ICM in which they are grounded. In short, we offer “A New Hope” against the essentialist empire. True to the Hollywood trope, this new hope rests on an alliance between a young theory, cognitive network neuroscience, and two older, but still quite young, epistemic rebels: enactive cognitive science and developmental systems theory. Accordingly, we detail and discuss the proposed emerging ICM and test-drive it by sketching the multimorphic view of gender it grounds. (shrink)

I contrast the “data fitting” vs “data interpreting” approaches to data science along three dimensions: Expediency, Transparency, and Explainability. “Data fitting” is driven by the faith that the secret to rational decisions lies in the data itself. In contrast, the data-interpreting school views data, not as a sole source of knowledge but as an auxiliary means for interpreting reality, and “reality” stands for the processes that generate the data. I argue for restoring balance to data science through a task-dependent symbiosis (...) of fitting and interpreting, guided by the Logic of Causation. (shrink)

Can machine intelligence do clinical ethics? And if so, would applying it to actual medical cases be desirable? In a recent target article (Meier et al. 2022), we described the piloting of our advisory algorithm METHAD. Here, we reply to commentaries published in response to our project. The commentaries fall into two broad categories: concrete criticism that concerns the development of METHAD; and the more general question as to whether one should employ decision-support systems of this kind—the debate we set (...) out to ignite with our target article. (shrink)

One of the best known arguments against the connectionist approach to artificial intelligence and cognitive science is that neural networks are black boxes, i.e., there is no understandable account of their operation. This difficulty has impeded efforts to explain how categories arise from raw sensory data. Moreover, it has complicated investigation about the role of symbols and language in cognition. This state of things has been radically changed by recent experimental findings in artificial deep learning research. Two kinds of artificial (...) deep learning networks, namely the convolutional neural network and the generative adversarial network have been found to possess the capability to build internal states that are interpreted by humans as complex visual categories, without any specific hints or any grammatical processing. This emergent ability suggests that those categories do not depend on human knowledge or the syntactic structure of language, while they do rely on their visual context. This supports a mild form of empiricism, while it does not assume that computational functionalism is true. Some consequences are extracted regarding the debate about amodal and grounded representations in the human brain. Furthermore, new avenues for research on cognitive science are open. (shrink)

Accounts of mental disorders focusing either on the brain as neurophysiological substrate or on systematic connections between symptoms are insufficient to account for the multifactorial nature of mental illnesses. Recently, multiplexes have been suggested to provide a holistic view of psychopathology that integrates data from different factors, at different scales, or across time. Intuitively, these multi-layered network structures present quite appealing models of mental disorders that can be constructed by powerful computational machinery based on increasing amounts of real-world data. In (...) this paper, I systematically examine what challenges psychopathology models face and to what extent different species of psychopathology models can address them. My analysis highlights that while multiplexes, as they are usually conceived, appear promising, they suffer from the same problems as other approaches. To remedy this, I suggest, we must go a step further and combine different kinds of multiplexes into 4D models. Once we embrace 4D multiplexes and identify appropriate ways to constrain them, we might unlock the true potential of multiplexes for making headway in psychopathology research. (shrink)

The question of how disability affects wellbeing has occupied a number of philosophers in recent years. However, this literature has proceeded without a careful examination of the fairly vast empirical research on the topic. In this paper, I review the scholarly literature and discuss some philosophically-relevant aspects of it. On average, those with disabilities have a significantly lower level of wellbeing than those without disabilities. Furthermore, there is strong evidence that this reduction in wellbeing is not due entirely to ableist (...) factors. Hence, I argue that it shows that the Mere Difference View of disability is most likely false. However, the literature indeed shows that disabled people typically live good lives. Thus, despite its rejection of the Mere Difference View, it still upholds a disability-affirmative view. These findings allow for a measured critique of some of Peter Singer’s writings on disability. (shrink)

The causal closure of physics is usually discussed in a context free way. Here I discuss it in the context of engineering systems and biology, where strong emergence takes place due to a combination of upwards emergence and downwards causation. Firstly, I show that causal closure is strictly limited in terms of spatial interactions because these are cases that are of necessity strongly interacting with the environment. Effective Spatial Closure holds ceteris parabus, and can be violated by Black Swan Events. (...) Secondly, I show that causal closure in the hierarchy of emergence is a strictly interlevel affair, and in the cases of engineering and biology encompasses all levels from the social level to the particle physics level. However Effective Causal Closure can usefully be defined for a restricted set of levels, and one can experimentally determine Effective Theories that hold at each level. This does not however imply those effective theories are causally complete by themselves. In particular, the particle physics level is not causally complete by itself in the contexts of solid state physics, digital computers, or biology. Furthermore Inextricably Intertwined Levels occur in all these contexts. (shrink)

We develop a mathematical and interpretative foundation for the enterprise of decision-theoretic (DT) statistical causality, which is a straightforward way of representing and addressing causal questions. DT reframes causal inference as “assisted decision-making” and aims to understand when, and how, I can make use of external data, typically observational, to help me solve a decision problem by taking advantage of assumed relationships between the data and my problem. The relationships embodied in any representation of a causal problem require deeper justification, (...) which is necessarily context-dependent. Here we clarify the considerations needed to support applications of the DT methodology. Exchangeability considerations are used to structure the required relationships, and a distinction drawn between intention to treat and intervention to treat forms the basis for the enabling condition of “ignorability.” We also show how the DT perspective unifies and sheds light on other popular formalisations of statistical causality, including potential responses and directed acyclic graphs. (shrink)

In ‘Computing Machinery and Intelligence’, Turing, sceptical of the question ‘Can machines think?’, quickly replaces it with an experimentally verifiable test: the imitation game. I suggest that for such a move to be successful the test needs to be relevant, expansive, solvable by exemplars, unpredictable, and lead to actionable research. The Imitation Game is only partially successful in this regard and its reliance on language, whilst insightful for partially solving the problem, has put AI progress on the wrong foot, prescribing (...) a top-down approach for building thinking machines. I argue that to fix shortcomings with modern AI systems a nonverbal operationalisation is required. This is provided by the recent Animal-AI Testbed, which translates animal cognition tests for AI and provides a bottom-up research pathway for building thinking machines that create predictive models of their environment from sensory input. (shrink)

In this paper, two central questions will be addressed: ought we to implement medical AI technology in the medical domain? If yes, how ought we to implement this technology? I will critically engage with three options that exist with respect to these central questions: the Neo-Luddite option, the Assistive option, and the Substitutive option. I will first address key objections on behalf of the Neo-Luddite option: the Objection from Bias, the Objection from Artificial Autonomy, the Objection from Status Quo, and (...) the Objection from Inscrutability. I will thereafter present the Demographic Trends Argument and the Human Enhancement Argument in support of alternatives to the Neo-Luddite option. In the second half of the paper, I will argue against the Substitutive option and in favour of the Assistive option, given the existence of two chief formal deficits in medical AI technology: the causality deficit and the care deficit. (shrink)

In this paper I argue that the Integrated Information Theory of Consciousness has an underlying emergentist metaphysics, specifically of a kind that has received minimal attention and we may call functionalist emergentism. I will try to show that in this scientific theory conscious experience is a functional-role property possessed by the whole system, not by their parts, which is dependent on, but also (purportedly) causally powerful over and above, the properties of the parts. However, I will argue that depicting conscious (...) experience as a functional-role emergent property threatens the whole coherence of the theory because, by definition, functional-realizers do all the causal work associated with the instantiation of any functional-role property on any occasion. Hence, to preserve the causal power of consciousness, which, as we will see, is one of the fundamental building blocks of the theory, linked to the assertion of the very existence of consciousness, phenomenal properties should be re-thought as realizers of the relevant informationally integrated causal-roles. (shrink)

This paper argues that machine learning and epidemiology are on collision course over causation. The discipline of epidemiology lays great emphasis on causation, while ML research does not. Some epidemiologists have proposed imposing what amounts to a causal constraint on ML in epidemiology, requiring it either to engage in causal inference or restrict itself to mere projection. We whittle down the issues to the question of whether causal knowledge is necessary for underwriting predictions about the outcomes of public health interventions. (...) While there is great plausibility to the idea that it is, conviction that something is impossible does not by itself motivate a constraint to forbid trying. We disambiguate the possible motivations for such a constraint into definitional, metaphysical, epistemological, and pragmatic considerations and argue that “Proceed with caution” is the outcome of each. We then argue that there are positive reasons to proceed, albeit cautiously. Causal inference enforces existing classification schema prior to the testing of associational claims, but associations and classification schema are more plausibly discovered in a back-and-forth process of gaining reflective equilibrium. ML instantiates this kind of process, we argue, and thus offers the welcome prospect of uncovering meaningful new concepts in epidemiology and public health—provided it is not causally constrained. (shrink)

Back in 1987 the physicist/theoretical biologist Walter Elsasser reviewed a range of philosophical issues at the foundation of organismal biology above the molecular level. Two of these are particularly relevant to quantifications of form: the concept of ordered heterogeneity and the principle of nonstructural memory, the truism that typically the forms of organisms substantially resemble the forms of their ancestors. This essay attempts to weave Elsasser’s principles together with morphometrics for one prominent data type, the representation of animal forms by (...) positions of landmark points. I introduce a new spectrum of pattern descriptors of the shape variation of landmark configurations, ranging from the most homogeneous through growth gradients and focal features to a model of totally disordered heterogeneity incompatible with the rhetoric of functional explanation. An investigation may end up reporting its findings by one of these descriptors or by several. These descriptors all derive from one shared diagrammatic device: a log–log plot of sample variance against one standard formalism of today’s geometric morphometrics, the bending energies of the principal warps that represent all the scales of variability around the sample average shape. The new descriptors, which I demonstrate over a variety of contemporary morphometric examples, may help build the bridge we urgently need between the arithmetic of today’s burgeoning image-based data resources and the rhetoric of biological explanations aligned with the principles of Elsasser along with an even earlier philosopher of biology, the Viennese visionary Hans Przibram. (shrink)

Habilitation thesis in psychology. The book consists of a collection of reasoning studies. The experimental investigations will take us from people’s reasoning about probabilities, entailments, pragmatic factors, argumentation, and causality to morality. An overarching theme of the book is norm pluralism and individual differences in rationality research.

Cognitive science is the interdisciplinary investigation of mind and intelligence, embracing psychology, neuroscience, anthropology, artificial intelligence, and philosophy. There are many important philosophical questions related to this investigation, but this short chapter will focus on the following three. What is the nature of the explanations and theories developed in cognitive science? What are the relations among the five disciplines that comprise cognitive science? What are the implications of cognitive science research for general issues in the philosophy of science? I will (...) argue that cognitive theories and explanations depend on representations of mechanisms and that the relations among the five disciplines, especially psychology and neuroscience, depend on relations between kinds of mechanisms. These conclusions have implications for central problems in general philosophy of science such as the nature of theories, explanations, and reduction between theories at different levels. (shrink)

Lucas Matthews and I substantially revised my SEP entry on Heritability. This version includes discussion of the missing heritability problem and other issues that arise from the use of Genome Wide Association Studies by Behavioral Geneticists.