Vincent C. Müller

This paper investigates the role of Inference to the Best Explanation (IBE) in Explainable Artificial Intelligence (XAI), arguing that IBE can both act as a framework for revealing the considerations and preferences behind XAI explanations and for guiding and evaluating them in a more rigorous manner. The paper focusses specifically on post-hoc explainability and model agnostic techniques, illustrating the argument through examples of salience maps and feature importance techniques. Here, IBE in XAI happens in two steps. The first aims at maximising the range and accuracy of possible inferences made; the second aims to provide a certain balance of explanatory virtues given a certain why-question and stakeholder that requires an explanation. Using the language of explanatory virtues enables a more nuanced discussion about what we may like or dislike about certain explanations in the current XAI landscape and provides actionable ways of improving explanations.

We describe a method, based on Jennifer Nado’s proposal for classification procedures as targets of conceptual engineering, that implements such procedures by prompting a large language model. We apply this method, using data from the Wikidata knowledge graph, to evaluate stipulative definitions related to two paradigmatic conceptual engineering projects: the International Astronomical Union’s redefinition of PLANET and Haslanger’s ameliorative analysis of WOMAN. Our results show that classification procedures built using our approach can exhibit good classification performance and, through the generation of rationales for their classifications, can contribute to the identification of issues in either the definitions or the data against which they are being evaluated. We consider objections to this method, and discuss implications of this work for three aspects of theory and practice of conceptual engineering: the definition of its targets, empirical methods for their investigation, and their practical roles. The data and code used for our experiments, together with the experimental results, are available in a Github repository (https://github.com/bradleypallen/zero-shot-classifiers-for-conceptual-engineering).

Taking collective artificial musical improvisation as a case study, we address software AI agents’ limitations due to their lack of embodiment and explore how machine learning resources may overcome them, without turning to robotics. Provided generative models are trained on embodied data, it is conceivable that the outputs of software agents may indirectly inherit some of the advantages of embodied cognitive processes. Drawing on empirical studies involving the software Somax2, we formulate an Embodiment via Machine Learning hypothesis and articulate the notion of embodiment that it involves.

This article delves into the evolving role of philosophy in governance debates, as affected by the emergence of Artificial Moral Agents (AMAs). It examines how AMAs, designed to make moral decisions, are shaping moral discourse and standards in various spheres of action. The analysis distinguishes between the traditional role of philosophers in critique and the potential shift towards policy drafting prompted by the development of AMAs. Ethical implications of these trends are explored, focusing on concerns about negotiability in societal standards and the alignment of moral standards with human conduct. The article highlights the need for careful consideration in policy and ethical discussions in this development.

Some AI systems employ simulated representations of real-world scenarios, and these simulations are sometimes called ‘imagination’ in the engineering literature. Can imagination be attributed to AI systems? This paper will consider this question from a conceptual engineering perspective. Some key properties of imagination are identified, which characterise imagination’s relation to other mental states and its epistemic function. Simulations run by AI systems resemble human imagination with respect to these properties only partially. Thus a purely descriptive approach, which aims to address the question of AI imagination by identifying the conditions that delineate our present concept of imagination, has limited prospects. Nevertheless, there are normative reasons to countenance AI imagination when simulation-based AI systems are seen as possible models of human intelligence. Some suggestions for adopting an AI-friendly concept of imagination will be offered.

Discussions surrounding AI-generated images mainly rest on questions regarding aesthetics; however, underneath this debate are more pressing questions regarding the labour. With the rise of AI, invisible labour extends to large-scale operations where billions of images and text samples are needed. I argue that the labour required for AI-generated images not only goes unnoticed but becomes actively hidden. I demonstrate that AI-generated images do not simply appear, but instead require a significant amount of human labour in the form of data creation, gathering, and classification. Besides an extensive theoretical exploration of invisible labour, this adds an empirical dimension by conducting semi-structured interviews with artists, computer scientists, and people inhibiting a ‘double role’. I show that the dichotomy between visibility and invisibility is not applicable in AI; instead, the socio-economic context and labour relations classify if something is invisible labour or not.

In this chapter, I argue against the view that existing AI systems can be deemed agents comparably to human beings or other organisms. I especially focus on the criteria of interactivity, autonomy, and adaptivity, provided by the seminal work of Luciano Floridi and José Sanders to determine whether an artificial system can be considered an agent. I argue that the tentacles of octopuses also fit those criteria. However, I argue that octopuses’ tentacles cannot be attributed agency because their behavior can be meaningfully interpreted only in reference to the octopus’ organism as an entire system. I argue that attributing agency to AI systems faces similar difficulties, and propose an alternative characterization of these systems as responsive and contested scaffolds.

In the aftermath of the success of attention-based transformer networks, the debate over the potential and role of consciousness in artificial systems has intensified. Prominently, the Global Neuronal Workspace Theory emerges as a front-runner in the endeavor to model consciousness in computational terms. A recent advancement in the direction of mapping the theory onto state-of-the-art machine learning tools is the model of a Global Latent Workspace. It introduces a central latent representation around which multiple modules are constructed. Leveraging dedicated encoder-decoder structures, content from the central representation or any individual module, integrated via the latent space, can be translated to any other module and back with minimal loss. This paper presents a thought experiment involving a minimal setup with one deep sensory and one deep motor module, which illustrates the emergence of “globally” accessible sensorimotor representations in the central latent space connecting both modules. In the human brain, neuronally enacted knowledge of laws relating changes in sensory information to changes in motor output or corresponding efferent copy information have been proposed to constitute the biological correlates of phenomenal conscious experience. The underlying Sensorimotor Contingency Theory encompasses a rich mathematical framework. Yet, the implementation of intelligent systems based on this framework has thus far been confined to proof-of-concept and basic prototype applications. Here, the natural appearance of global latent sensorimotor representations links two major neuroscientific theories of consciousness in a powerful machine learning setup. A remaining question is whether this artificial system is conscious.

Artificial intelligence (AI) involves considerable risks and uncertainties. Consequently, there is a growing need for normative guidance. In addition to an ethical understanding of AI-related issues, a legal framework is needed to ensure that important ethical aspects are adequately considered in the development and application of AI technologies. Therefore, regulatory frameworks have emerged, such as the European Union’s 2024 AI Act. However, it is an open question to what extent AI regulation aligns with AI ethics guidelines. This paper aims to fill this research gap by comparing relevant ethics guidelines with legal frameworks. We focus on ethical aspects that are understood in these texts as ‘principles,’ ‘values,’ ‘recommendations’ or ‘requirements’ and provide a systematization of these aspects. A specific finding of the comparison is that in the current legal proposals, the value ‘well-being’ and its connected principles are not yet operationalized. However, these aspects are shared elements of the main ethics guidelines. This knowledge can be useful for the further development of the legal framework by illustrating important ethical aspects of AI that are not (yet) transferred to the legal realm. It is an open question how the legal system should deal with this issue.

This paper explores the profound cognitive and affective impacts of Large Language Models (LLMs) and Generative AI (GenAI) on human thought, imagination, and especially our emotional life by analysing our interaction with different styles of GenAI agents especially when presented as virtual personalities. I describe how LLL-based agents can be presented using one of three broad metaphors of interaction, namely as Assistants, Extenders and Companions. Each metaphor of interaction implies distinctive cognitive patterns for their human interactant. I explore especially the shape of some current interactions with GenAI companions and analyse their roles as cognitive and affective mind shapers. I analyse how Companion AI agents based on LLM technology can come to be treated as ‘Ersatz Others’ taking up deep roles in our cognitive and affective lives. I explore some of the perhaps unexpected implications of the deep reliance on such systems especially from the point of view of the theory of the extended mind and the mind-technology problem and show how this new sort of cognitive and affective environment for the human mind produces some new ethical challenges and may even reshape our future understandings of what minds are.

The Digital Revolution has given rise to a new era of perceptual experiences, marked by the convergence of analogue and digital dimensions into what is collectively known as Mixed Reality (MR). This paper explores the complexities of perception within digital environments, addressing fundamental questions regarding body and object experiences in Virtual Reality and Augmented Reality. The study aims to redefine MR by adopting a phenomenological perspective applicable to the intertwining of both analogue and digital dimensions.The reality-virtuality (RV) continuum is revisited, challenging the dichotomy between completely analogue and fully digital dimensions. Drawing inspiration from Maurice Merleau-Ponty’s concept of flesh, the paper proposes a phenomenological theory of perception rooted in the interconnectedness of living and non-living entities. This perspective fosters a posthuman reading of Merleau-Ponty’s phenomenology, highlighting the co-participation of the extended body in analogue and digital environments. The study contributes to the understanding of transdimensionality within the MR spectrum, providing a new definition of the continuum as occurring between analogue and digital aspects (AD continuum). In conclusion, the AD continuum is presented as an amalgamation of the human living body and digital technology, offering new insights for the understanding of immersive MR environments.

Many Artificial Intelligence (AI) tools have been developed for suicide prevention. These technologies are wide ranging, from Language Models (LM) in chatbots for therapy or triaging, computer vision algorithms designed to detect specific human motions to the use of limited emotion theories to detect emotional states. Both their design and implementation raise ethical concerns, especially considering that - unlike clinical research on human subjects - they are not necessarily subjected to any regulations or ethics governance structures, nor does the design or implementation require any formal expertise in mental health. Considering the nature of such technologies, the high-risk area they are applied in, and the explicit targeting of vulnerable populations, this is concerning. In this work, we aim to (a) give an overview of such tools, (b) provide some insight into key ethical concerns, (c) demonstrate that these technologies do not necessarily comply with existing guidelines regarding human subject research, although some of the tools would most likely be subjected to it, had they been done in a clinical setting (d) argue that there is an unmet need for specific ethics guidelines for research, scientists, and practitioners working at the intersection of AI and suicide prevention, and (e) provide such guidelines.

Chatbots powered by large language models (LLMs) are increasingly capable of engaging in what seems like natural conversations with humans. This raises the question of whether we should interact with these chatbots in a morally considerate manner. In this chapter, we examine how to answer this question from within the normative framework of virtue ethics. In the literature, two kinds of virtue ethics arguments, the moral cultivation and the moral character argument, have been advanced to argue that we should afford moral treatment to social robots. However, we propose a moral character argument against the view that we should afford moral treatment to LLM-based chatbots drawing on the notion of practical wisdom. Practical wisdom in this context consists in the skill to discern genuine and artificial sociality. Drawing on ideas from phenomenological philosophy, we argue that this involves the ability to distance oneself from direct social perception and the ability to widen one’s awareness over one’s apparently social interactions. We conclude by suggesting that this skill is a kind of technomoral wisdom required to live well with advanced, social AI systems.

This paper explores how our concept of creativity has to be rethought in light of the increasing role of emerging generative Artificial Intelligence (GenAI) technologies in creative practice, thereby enriching the concept to include collective creativity produced by humans and AI. Collective creativity arises from the collective practices of people in relation to their nature, environment, cultures, societies, techniques, technologies, and AI. This conception of collective creativity challenges the traditional idea of the individual genius. By expanding our concept of creativity, we are able to acknowledge the processes and products of humans and AI as creative. Exploring case studies of co-created art, music, and text, I argue that AI can be creative and demonstrate how AI can be creative and enhance our creativity.

In an era of pervasive misinformation, the help of artificial intelligence like Large Language Models (LLMs) in identifying and correcting falsehoods is critical. This study investigates the alignment between humans’ and GPT-4’s judgments on lying and falsity, with a focus on deceptive implicatures—statements that are literally true but imply something false. Using a cross-cultural human dataset (3660 participants across ten countries) as a benchmark, we evaluated GPT-4’s performance in six languages. Results indicate that while GPT-4’s judgments correlate significantly with human responses, notable discrepancies also exist. GPT-4 tends to attribute higher lie scores and lower falsity scores than humans and exhibits more severe moral judgments. These findings highlight the potential risks of deploying LLMs like GPT-4 for misinformation detection, suggesting possible over- or under classification (depending on the task). The study highlights the importance of cautious implementation and the need for further refinement to enhance human-LLM alignment in identifying deception.

Robert Sparrow argues that while we should recognize vice in mistreatments and cruelty towards robots, like kicking, we shouldn’t recognize virtue. I discuss some problems with this argument. While we’re supposed to investigate whether it is wrong to kick robots, Sparrow’s premise isn’t neutral about such actions, describing them with morally loaded terms like ‘cruel’. I show that such premises with morally loaded terms cannot be taken as given in Human-Robot Interaction (HRI). By focusing on graphic mistreatment scenarios and on what robots resemble or represent, Sparrow and others fail to account for broader, everyday interactions. Where robots come with unique affordance mixtures, don’t necessarily resemble or represent other beings and are directly perceived (rather than fictionalized as Sparrow claims). Thus, robots can be treated in ways different than humans and animals. I emphasize how situation, context, dynamics, interaction and direct affordance perception matter in HRI, where robots play an active role, rather than being passively anthropomorphized or fictionalized. Thus, there is a need for more cognitive scientific insight into HRI ethics. Especially into the indirect robot moral patiency debate, which has been stuck with graphic mistreatment scenarios and stuck with limited descriptive concepts like anthropomorphism and taking robots as representation of other beings.

Modern AI systems based on deep learning are neither traditional tools nor full-blown agents. Rather, they are characterised by idiosyncratic agential profiles, i.e., combinations of agency-relevant properties. Modern AI systems lack superficial features which enable people to recognise agents but possess sophisticated information processing capabilities which can undermine human goals. I argue that systems fitting this description, when they are adversarial with respect to human users, pose particular risks to those users. To explicate my argument, I provide conditions under which agential profiles are explanatorily relevant to harms caused. I then contend that the role of recommender systems in producing harmful outcomes like digital addiction satisfies these conditions.

We can formulate an argument against AI creativity from agency. By some accounts, creativity requires agency, and agency is, many think, not possible for AI. This is due to the typical conception of agency as a capacity for intentional action. Intentional action is thought to require mental states, a severe challenge for machine intelligence. On the face of things, the agency argument seems to provide a straightforward route to argue for the impossibility of AI creativity. However, this path, I argue, is not so clear. In this paper, I outline the agency argument against AI creativity, before calling into question the apparent simplicity of this argument. I argue, ultimately, that the reasoning behind the inclusion of agency in accounts of creativity does not necessitate the use of intentional action, but can instead be satisfied by a minimal teleological account of agency.

This paper examines how metaphors shape our thinking about and conceptualizing of artificial intelligence (AI), noting that their inherent imprecision leads to discrepancies in our understanding and objectives for AI. By exploring the concept of ‘bad metaphors’ that equate artificial intelligence with human intelligence, paper argues that these metaphors often carry additional, unintended meanings that distort our understanding and expectations of AI. The terms “artificial” and “intelligence” themselves are ambiguous and ideologically loaded, contributing to the complexity. The paper critiques the anthropocentric and mechanistic metaphors, like “AI as human” and “brain-computer,” which perpetuate unrealistic expectations. By deconstructing these metaphors within broader cultural and historical contexts, the paper calls for a more nuanced and precise understanding of AI, moving beyond simplistic and potentially misleading analogies.

Despite significant advancements in XAI, scholars note a persistent lack of solid conceptual foundations and integration with broader scientific discourse on explanation. In response, emerging research draws on explanatory strategies from various sciences and the philosophy of science literature to fill these gaps. This paper outlines a mechanistic strategy for explaining the functional organization of deep learning systems, situating recent developments in explainable AI within a broader philosophical context. According to the mechanistic approach, the explanation of opaque AI systems involves identifying mechanisms that drive decision making. For deep neural networks, this means discerning functionally relevant components, such as neurons, layers, circuits, or activation patterns, and understanding their roles through decomposition, localization, and recomposition. Proof-of-principle case studies from image recognition and language modeling align these theoretical approaches with mechanistic interpretability research from OpenAI and Anthropic. The findings suggest that pursuing mechanistic explanations can uncover elements that traditional explainability techniques may overlook, ultimately contributing to more thoroughly explainable AI.

This paper examines the problem of disparity between AI systems and human agents in key human characteristics. We point out that this disparity, or gap, despite being recognized, has not been systematically studied. We also argue that without a clearer understanding of this gap, we cannot accurately predict how AI systems will interact with human agents or how they will account for human values and priorities. This paper defines this gap and proposes a methodology for studying it. We also illustrate this conceptual gap with two examples, namely the differences between a synthetic and human consciousness and synthetic agency versus human moral agency. Finally, we emphasize the importance of studying this gap for developing safe AI technology.

Medical expertise is an essential concept when discussing the potential value of Artificial Intelligence (AI) systems in advancing the medical field, as one of AI’s main promises is to positively complement the expertise of medical professionals. Yet, relatively few articles on medical AI explicitly mention what medical expertise entails, how it is practiced in medical contexts, and how AI systems relate to medical professionals’ expertise. We believe it is essential to focus on answering these questions to determine how AI should be integrated into medical practice and how we should formulate the expected benefits of AI systems in clinical care. With this article, we conceptualize medical expertise and identify four forms of medical expertise: epistemic expertise, performative expertise, adaptive expertise, and virtuous expertise. Additionally, we highlight how this conceptualization of expertise can support the ethical evaluation of AI’s augmenting role in medical practices.

A well-known problem in machine learning is “fairness impossibility”: that is, the mathematically demonstrable result that several intuitively compelling standards of fairness cannot be simultaneously realized, except in a narrow and rare set of cases. Many researchers have proposed responses to fairness impossibility, but there is still no consensus on how to handle the problem. In this paper I briefly review the problem and catalog previous efforts to respond, noting their strengths and weaknesses. I then propose an approach that situates concerns about fairness within a more general theory of ethical decision-making (an “integrative ethical framework”) and recommends responses based on ethically relevant features of the decision context. Several distinctions in this framework are especially important for guiding responses to fairness impossibility in particular situations, including rights vs. goods and substantive vs. procedural ethical considerations. Implications for the fairness impossibility debate include identification of cases where specific responses such as softening fairness requirements, subjecting fairness trade-offs to public opinion, or favoring one fairness metric over others, are preferable over alternatives.

There is, in some quarters, concern about high–level machine intelligence and superintelligent AI coming up in a few decades, bringing with it significant risks for humanity. In other quarters, these issues are ignored or considered science fiction. We wanted to clarify what the distribution of opinions actually is, what probability the best experts currently assign to high–level machine intelligence coming up within a particular time–frame, which risks they see with that development, and how fast they see these developing. We thus designed a brief questionnaire and distributed it to four groups of experts in 2012/2013. The median estimate of respondents was for a one in two chance that high-level machine intelligence will be developed around 2040–2050, rising to a nine in ten chance by 2075. Experts expect that systems will move on to superintelligence in less than 30 years thereafter. They estimate the chance is about one in three that this development turns out to be ‘bad’ or ‘extremely bad’ for humanity.

There seems to be dilemma whether we should direct our efforts in AI ethics towards problems that are clearly visible on the horizon today (short-term), or towards problems for which we see significant probability of them occurring at some point (long-term), provided they are significant enough. Some authors have argued that we should put a heavy focus on the one or the other. I will argue that this is a false dilemma: Any rational agent will consider both short- and long-term consequences (as well as other factors). My analysis is that the supposed dilemma rests on the assumption that we are at a very unusual situation that forces favouring of one option – what I call an “ethical singularity”. The only serious argument for this view is that the longer term involves a) an ethically “significantly different” situation, and b) that this demands a version of “ethical fanaticism”. Both premises turn out to have little in their favour. We should thus return to the “normal balance” of expected utility in AI ethics and solve problems in all relevant time horizons concurrently.

It is known that big data analytics and AI pose a threat to privacy, and that some of this is due to some kind of “black box problem” in AI. I explain how this becomes a problem in the context of justification for judgments and actions. Furthermore, I suggest distinguishing three kinds of opacity: 1) the subjects do not know what the system does (“shallow opacity”), 2) the analysts do not know what the system does (“standard black box opacity”), or 3) the analysts cannot possibly know what the system might do (“deep opacity”). If the agents, data subjects as well as analytics experts, operate under opacity, then these agents cannot provide justifications for judgments that are necessary to protect privacy, e.g., they cannot give “informed consent”, or guarantee “anonymity.” It follows from these points that agents in big data analytics and AI ofen cannot make the judgments needed to protect privacy. So I conclude that big data analytics makes the privacy problems worse and the remedies less effective. As a positive note, I provide a brief outlook on technical ways to handle this situation.

Digital ethics, also known as computer ethics or information ethics, is now a lively field that draws a lot of attention, but how did it come about and what were the developments that lead to its existence? What are the traditions, the concerns, the technological and social developments that pushed digital ethics? How did ethical issues change with digitalisation of human life? How did the traditional discipline of philosophy respond? The article provides an overview, proposing historical epochs: ‘pre-modernity’ prior to digital computation over data, via the ‘modernity’ of digital data processing to our present ‘post-modernity’ when not only the data is digital, but our lives themselves are largely digital. In each section, the situation in technology and society is sketched, and then the developments in digital ethics are explained. Finally, a brief outlook is provided.