Organizations such as the EU High-Level Expert Group on AI and the IEEE have recently formulated ethical principles and (moral) values that should be adhered to in the design and deployment of artificial intelligence (AI). These include respect for autonomy, non-maleficence, fairness, transparency, explainability, and accountability. But how can we ensure and verify that an AI system actually respects these values? To help answer this question, I propose an account for determining when an AI system can be said to embody (...) certain values. This account understands embodied values as the result of design activities intended to embed those values in such systems. AI systems are here understood as a special kind of sociotechnical system that, like traditional sociotechnical systems, are composed of technical artifacts, human agents, and institutions but—in addition—contain artificial agents and certain technical norms that regulate interactions between artificial agents and other elements of the system. The specific challenges and opportunities of embedding values in AI systems are discussed, and some lessons for better embedding values in AI systems are drawn. (shrink)

Why should moral philosophers, moral psychologists, and machine ethicists care about computational complexity? Debates on whether artificial intelligence (AI) can or should be used to solve problems in ethical domains have mainly been driven by what AI can or cannot do in terms of human capacities. In this paper, we tackle the problem from the other end by exploring what kind of moral machines are possible based on what computational systems can or cannot do. To do so, we analyze normative (...) ethics through the lens of computational complexity. First, we introduce computational complexity for the uninitiated reader and discuss how the complexity of ethical problems can be framed within Marr’s three levels of analysis. We then study a range of ethical problems based on consequentialism, deontology, and virtue ethics, with the aim of elucidating the complexity associated with the problems themselves (e.g., due to combinatorics, uncertainty, strategic dynamics), the computational methods employed (e.g., probability, logic, learning), and the available resources (e.g., time, knowledge, learning). The results indicate that most problems the normative frameworks pose lead to tractability issues in every category analyzed. Our investigation also provides several insights about the computational nature of normative ethics, including the differences between rule- and outcome-based moral strategies, and the implementation-variance with regard to moral resources. We then discuss the consequences complexity results have for the prospect of moral machines in virtue of the trade-off between optimality and efficiency. Finally, we elucidate how computational complexity can be used to inform both philosophical and cognitive-psychological research on human morality by advancing the moral tractability thesis. (shrink)

Machine ethics is the field that strives to develop ‘artificial moral agents’ (AMAs), artificial systems that can autonomously make moral decisions. Some authors have questioned the feasibility of machine ethics, by questioning whether artificial systems can possess moral competence, or the capacity to reach morally right decisions in various situations. This paper explores this question by drawing on the work of several moral philosophers (McDowell, Wiggins, Hampshire, and Nussbaum) who have characterised moral competence in a manner inspired by Aristotle. Although (...) disparate in many ways, these philosophers all emphasise what may be called ‘moral sensitivity’ as a precondition for moral competence. Moral sensitivity is the uncodified, practical skill to recognise, in a range of situations, which features of the situations are morally relevant, and how they are relevant. This paper argues that the main types of AMAs currently proposed are incapable of full moral sensitivity. First, top-down AMAs that proceed from fixed rule-sets are too rigid to respond appropriately to the wide range of qualitatively unique factors that moral sensitivity gives access to. Second, bottom-up AMAs that learn moral behaviour from examples are at risk of generalising from these examples in undesirable ways, as they lack embedding in what Wittgenstein calls a ‘form of life’, which allows humans to appropriately learn from moral examples. The paper concludes that AMAs are unlikely to possess full moral competence, but closes by suggesting that they may still be feasible in restricted domains of public morality, where moral sensitivity plays a smaller role. (shrink)

This paper concerns top-down approaches in machine ethics. It is divided into three main parts. First, I briefly describe top-down design approaches, and in doing so I make clear what those approaches are committed to and what they involve when it comes to training an AI to behave ethically. In the second part, I formulate two underappreciated motivations for endorsing them, one relating to predictability of machine behavior and the other relating to scrutability of machine decision-making. Finally, I present three (...) major worries about such approaches, and I attempt to show that advocates of top-down approaches have some plausible avenues of response. I focus most of my attention on what I call the ‘technical manual objection’ to top-down approaches, inspired by the work of Annas (2004). In short, the idea is that top-down approaches treat ethical decision-making as being merely a matter of following some ethical instructions in the same way that one might follow some set of instructions contained in a technical manual (e.g., computer manual), and this invites sensible skepticism about the ethical wisdom of machines that have been trained on those approaches. I respond by claiming that the objection is successful only if it is understood as targeting machines that have certain kinds of goals, and it should not compel us to totally abandon top-down approaches. Such approaches could still be reasonably employed to design ethical AI that operate in contexts that include fairly noncontroversial answers to ethical questions. In fact, we should prefer top-down approaches when it comes to those types of context, or so I argue, due to the advantages I claim for them. (shrink)

Recent advancements in artificial intelligence have fueled widespread academic discourse on the ethics of AI within and across a diverse set of disciplines. One notable subfield of AI ethics is machine ethics, which seeks to implement ethical considerations into AI systems. However, since different research efforts within machine ethics have discipline-specific concepts, practices, and goals, the resulting body of work is pestered with conflict and confusion as opposed to fruitful synergies. The aim of this paper is to explore ways to (...) alleviate these issues, both on a practical and theoretical level of analysis. First, we describe two approaches to machine ethics: the philosophical approach and the engineering approach and show how tensions between the two arise due to discipline specific practices and aims. Using the concept of disciplinary capture, we then discuss potential promises and pitfalls to cross-disciplinary collaboration. Drawing on recent work in philosophy of science, we finally describe how metacognitive scaffolds can be used to avoid epistemological obstacles and foster innovative collaboration in AI ethics in general and machine ethics in particular. (shrink)

Virtue ethics has many times been suggested as a promising recipe for the construction of artificial moral agents due to its emphasis on moral character and learning. However, given the complex nature of the theory, hardly any work has de facto attempted to implement the core tenets of virtue ethics in moral machines. The main goal of this paper is to demonstrate how virtue ethics can be taken all the way from theory to machine implementation. To achieve this goal, we (...) critically explore the possibilities and challenges for virtue ethics from a computational perspective. Drawing on previous conceptual and technical work, we outline a version of artificial virtue based on moral functionalism, connectionist bottom–up learning, and eudaimonic reward. We then describe how core features of the outlined theory can be interpreted in terms of functionality, which in turn informs the design of components necessary for virtuous cognition. Finally, we present a comprehensive framework for the technical development of artificial virtuous agents and discuss how they can be implemented in moral environments. (shrink)

Although virtue ethics has repeatedly been proposed as a suitable framework for the development of artificial moral agents, it has been proven difficult to approach from a computational perspective. In this work, we present the first technical implementation of artificial virtuous agents in moral simulations. First, we review previous conceptual and technical work in artificial virtue ethics and describe a functionalistic path to AVAs based on dispositional virtues, bottom-up learning, and top-down eudaimonic reward. We then provide the details of a (...) technical implementation in a moral simulation based on a tragedy of the commons scenario. The experimental results show how the AVAs learn to tackle cooperation problems while exhibiting core features of their theoretical counterpart, including moral character, dispositional virtues, learning from experience, and the pursuit of eudaimonia. Ultimately, we argue that virtue ethics provides a compelling path toward morally excellent machines and that our work provides an important starting point for such endeavors. (shrink)

Virtue ethics has many times been suggested as a promising recipe for the construction of artificial moral agents due to its emphasis on moral character and learning. However, given the complex nature of the theory, hardly any work has de facto attempted to implement the core tenets of virtue ethics in moral machines. The main goal of this paper is to demonstrate how virtue ethics can be taken all the way from theory to machine implementation. To achieve this goal, we (...) critically explore the possibilities and challenges for virtue ethics from a computational perspective. Drawing on previous conceptual and technical work, we outline a version of artificial virtue based on moral functionalism, connectionist bottom–up learning, and eudaimonic reward. We then describe how core features of the outlined theory can be interpreted in terms of functionality, which in turn informs the design of components necessary for virtuous cognition. Finally, we present a comprehensive framework for the technical development of artificial virtuous agents and discuss how they can be implemented in moral environments. (shrink)

We examine Van Wynsberghe and Robbins (JAMA 25:719-735, 2019) critique of the need for Artificial Moral Agents (AMAs) and its rebuttal by Formosa and Ryan (JAMA 10.1007/s00146-020-01089-6, 2020) set against a neo-Aristotelian ethical background. Neither Van Wynsberghe and Robbins (JAMA 25:719-735, 2019) essay nor Formosa and Ryan’s (JAMA 10.1007/s00146-020-01089-6, 2020) is explicitly framed within the teachings of a specific ethical school. The former appeals to the lack of “both empirical and intuitive support” (Van Wynsberghe and Robbins 2019, p. 721) for (...) AMAs, and the latter opts for “argumentative breadth over depth”, meaning to provide “the essential groundwork for making an all things considered judgment regarding the moral case for building AMAs” (Formosa and Ryan 2019, pp. 1–2). Although this strategy may benefit their acceptability, it may also detract from their ethical rootedness, coherence, and persuasiveness, characteristics often associated with consolidated ethical traditions. Neo-Aristotelian ethics, backed by a distinctive philosophical anthropology and worldview, is summoned to fill this gap as a standard to test these two opposing claims. It provides a substantive account of moral agency through the theory of voluntary action; it explains how voluntary action is tied to intelligent and autonomous human life; and it distinguishes machine operations from voluntary actions through the categories of poiesis and praxis respectively. This standpoint reveals that while Van Wynsberghe and Robbins may be right in rejecting the need for AMAs, there are deeper, more fundamental reasons. In addition, despite disagreeing with Formosa and Ryan’s defense of AMAs, their call for a more nuanced and context-dependent approach, similar to neo-Aristotelian practical wisdom, becomes expedient. (shrink)

Some authors have recently suggested that it is time to consider rights for robots. These suggestions are based on the claim that the question of robot rights should not depend on a standard set of conditions for ‘moral status’; but instead, the question is to be framed in a new way, by rejecting the is/ought distinction, making a relational turn, or assuming a methodological behaviourism. We try to clarify these suggestions and to show their highly problematic consequences. While we find (...) the suggestions ultimately unmotivated, the discussion shows that our epistemic condition with respect to the moral status of others does raise problems, and that the human tendency to empathise with things that do not have moral status should be taken seriously—we suggest that it produces a “derived moral status”. Finally, it turns out that there is typically no individual in real AI that could even be said to be the bearer of moral status. Overall, there is no reason to think that robot rights are an issue now. (shrink)

Machine intelligence already helps medical staff with a number of tasks. Ethical decision-making, however, has not been handed over to computers. In this proof-of-concept study, we show how an algorithm based on Beauchamp and Childress’ prima-facie principles could be employed to advise on a range of moral dilemma situations that occur in medical institutions. We explain why we chose fuzzy cognitive maps to set up the advisory system and how we utilized machine learning to train it. We report on the (...) difficult task of operationalizing the principles of beneficence, non-maleficence and patient autonomy, and describe how we selected suitable input parameters that we extracted from a training dataset of clinical cases. The first performance results are promising, but an algorithmic approach to ethics also comes with several weaknesses and limitations. Should one really entrust the sensitive domain of clinical ethics to machine intelligence? (shrink)

As AI Systems become increasingly autonomous, they are expected to engage in decision-making processes that have moral implications. In this research we integrate theoretical and empirical lines of thought to address the matters of moral reasoning and moral uncertainty in AI Systems. We reconceptualize the metanormative framework for decision-making under moral uncertainty and we operationalize it through a latent class choice model. The core idea being that moral heterogeneity in society can be codified in terms of a small number of (...) classes with distinct moral preferences and that this codification can be used to express moral uncertainty of an AI. Choice analysis allows for the identification of classes and their moral preferences based on observed choice data. Our reformulation of the metanormative framework is theory-rooted and practical in the sense that it avoids runtime issues in real time applications. To illustrate our approach we conceptualize a society in which AI Systems are in charge of making policy choices. While one of the systems uses a baseline morally certain model, the other uses a morally uncertain model. We highlight cases in which the AI Systems disagree about the policy to be chosen, thus illustrating the need to capture moral uncertainty in AI systems. (shrink)

Programming artificial intelligence to make fairness assessments of texts through top-down rules, bottom-up training, or hybrid approaches, has presented the challenge of defining cross-cultural fairness. In this paper a simple method is presented which uses vectors to discover if a verb is unfair or fair. It uses already existing relational social ontologies inherent in Word Embeddings and thus requires no training. The plausibility of the approach rests on two premises. That individuals consider fair acts those that they would be willing (...) to accept if done to themselves. Secondly, that such a construal is ontologically reflected in Word Embeddings, by virtue of their ability to reflect the dimensions of such a perception. These dimensions being: responsibility vs. irresponsibility, gain vs. loss, reward vs. sanction, joy vs. pain, all as a single vector. The paper finds it possible to quantify and qualify a verb as fair or unfair by calculating the cosine similarity of the said verb’s embedding vector against FairVec—which represents the above dimensions. We apply this to Glove and Word2Vec embeddings. Testing on a list of verbs produces an F1 score of 95.7, which is improved to 97.0. Lastly, a demonstration of the method’s applicability to sentence measurement is carried out. (shrink)

Programming artificial intelligence to make fairness assessments of texts through top-down rules, bottom-up training, or hybrid approaches, has presented the challenge of defining cross-cultural fairness. In this paper a simple method is presented which uses vectors to discover if a verb is unfair or fair. It uses already existing relational social ontologies inherent in Word Embeddings and thus requires no training. The plausibility of the approach rests on two premises. That individuals consider fair acts those that they would be willing (...) to accept if done to themselves. Secondly, that such a construal is ontologically reflected in Word Embeddings, by virtue of their ability to reflect the dimensions of such a perception. These dimensions being: responsibility vs. irresponsibility, gain vs. loss, reward vs. sanction, joy vs. pain, all as a single vector. The paper finds it possible to quantify and qualify a verb as fair or unfair by calculating the cosine similarity of the said verb’s embedding vector against FairVec—which represents the above dimensions. We apply this to Glove and Word2Vec embeddings. Testing on a list of verbs produces an F1 score of 95.7, which is improved to 97.0. Lastly, a demonstration of the method’s applicability to sentence measurement is carried out. (shrink)

Artificial intelligence is shaping strategy, activities, interactions, and relationships in business and specifically in marketing. The drawback of the substantial opportunities AI systems and applications provide in marketing are ethical controversies. Building on the literature on AI ethics, the authors systematically scrutinize the ethical challenges of deploying AI in marketing from a multi-stakeholder perspective. By revealing interdependencies and tensions between ethical principles, the authors shed light on the applicability of a purely principled, deontological approach to AI ethics in marketing. To (...) reconcile some of these tensions and account for the AI-for-social-good perspective, the authors make suggestions of how AI in marketing can be leveraged to promote societal and environmental well-being. (shrink)

Artificial intelligence can be a game changer to address the global challenge of humanity-threatening climate change by fostering sustainable development. Since chemical research and development lay the foundation for innovative products and solutions, this study presents a novel chemical research and development process backed with artificial intelligence and guiding ethical principles to account for both process- and outcome-related sustainability. Particularly in ethically salient contexts, ethical principles have to accompany research and development powered by artificial intelligence to promote social and environmental (...) good and sustainability while preventing any harm for all stakeholders affected. (shrink)

I describe a suite of reinforcement learning environments in which artificial agents learn to value and respond to moral content and contexts. I illustrate the core principles of the framework by characterizing one such environment, or “gridworld,” in which an agent learns to trade-off between monetary profit and fair dealing, as applied in a standard behavioral economic paradigm. I then highlight the core technical and philosophical advantages of the learning approach for modeling moral cognition, and for addressing the so-called value (...) alignment problem in AI. (shrink)

In this philosophical exploration, we investigate the concept of robotic moral environment interaction. The common view understands moral interaction to occur between agents endowed with ethical and interactive capacities. However, recent developments in moral philosophy argue that moral interaction also occurs in relation to the environment. Here conditions and situations of the environment contribute to human moral cognition and the formation of our moral experiences. Based on this philosophical position, we imagine robots interacting as moral environments—a novel conceptualization of human–robot (...) moral interaction with an inherent capacity for moral augmentation. To explicate this idea, we first define moral environments as moral systems providing moral affordances. We then intuit and explicate two constitutive conditions of moral environments: the environment’s moral ambiance and its moral atmosphere and compare them with real-life cases of moral environments—the moral landmarks. Based on these explications, we construct several explanatory cases to illustrate robots interacting as moral environments. Lastly, we set to evaluate the ethical challenges and wider social ramifications of using moral environment robots in the public space. (shrink)

It is widely agreed that possession of consciousness contributes to an entity’s moral status, even if it is not necessary for moral status (Levy and Savulescu 2009). An entity is considered to have...

Should artificial intelligences ever be included as co-authors of democratic decisions? According to the conventional view in democratic theory, the answer depends on the relationship between the political unit and the entity that is either affected or subjected to its decisions. The relational conditions for inclusion as stipulated by the all-affected and all-subjected principles determine the spatial extension of democratic inclusion. Thus, AI qualifies for democratic inclusion if and only if AI is either affected or subjected to decisions by the (...) political unit in relevant ways. This paper argues that the conventional view is too simple; that it neglects democratic reasons to recognize only agents and/or moral patients as participants in decision-making. The claim defended is that AAP and ASP implicitly affirm requirements for agency and patiency. In ASP, the entity included must be an agent understood either in terms of legal status, capacity to comply with the law or ability to recognize legitimate authority. In AAP, the entity included must be a patient, understood either in terms of capacity for sentience or consciousness. Thus, the idea here is to explore the potential democratic inclusion of artificial intelligences by an updated account of the relevant conditions of agency and patiency that are implicit in democratic theory. Although conceivable that AI is or will be either affected or subjected in relevant ways to decisions made by political units, it is far less clear that AI will ever be agents or patients in the sense required for democratic inclusion. (shrink)

In general, existential threats are those that may potentially result in the extinction of the entire human species, if not significantly endanger its living population. Among the said threats include, but not limited to, pandemics and the impacts of a technological singularity. As regards pandemics, significant work has already been done on how to mitigate, if not prevent, the aftereffects of this type of disaster. For one, certain problem areas on how to properly manage pandemic responses have already been identified, (...) like the following: (a) not being able to learn from previous experiences, (b) the inability to act on warning signals, and (c) the failure to reach a global consensus on a problem (i.e., in a timely manner). In terms of a singularity, however, it may be said that further research is still needed, specifically on how to aptly respond to its projected negative outcomes. In this paper, by treating the three problem areas noted above as preliminary assessment measures of a country’s capacity to coordinate a national response to large-scale disasters, we examine the readiness of the Philippines in preparing for an intelligence explosion. By citing certain instances of how the said country, specifically its national government, faced the coronavirus disease 2019 pandemic, it puts forward the idea that the likely Philippine disaster response towards a singularity needs to be worked on, appealing for a more comprehensive assessment of such for a more informed response plan. (shrink)