William Hasselberger

The creation of machines with intelligence comparable to human beings—so-called "human-level” and “general” intelligence—is often regarded as the Holy Grail of Artificial Intelligence (AI) research. However, many prominent discussions of AI lean heavily on the notion of human-level intelligence to frame AI research, but then rely on conceptions of human cognitive capacities, including “common sense,” that are sketchy, one-sided, philosophically loaded, and highly contestable. Our goal in this essay is to bring into view some underappreciated features of the practical intelligence involved in ordinary human agency. These features of practical intelligence are implicit in the structure of our first-person experience of embodied and situated agency, deliberation, and human interaction. We argue that spelling out these features and their implications reveals a fundamental distinction between two forms of intelligence in action, or what we call “efficient task-completion” versus “intelligent engagement in activity.” This distinction helps us to see what is missing from some widely accepted ways of thinking about human-level intelligence in AI, and how human common sense is actually tied, conceptually, to the ideal of practical wisdom, or good (normative) judgment about how to act and live well. Finally, our analysis, if sound, also has implications for the important ethical question of what it means to have AI systems that are aligned with human values, or the so-called “value alignment” problem for artificial intelligence.

This paper focuses on a central question for Human-AI interaction: Can you be friends with an AI agent? If not, why not? Some have argued that friendship with AI agents is impossible because software artifacts do not, and cannot, care about you. Proponents of human–machine friendships have responded that such relationships may indeed be one-sided, but still count as relationships of genuine love and affection—perhaps constituting a whole new category of friendship. Our paper takes a different path. We argue that you cannot be friends with an AI agent because you cannot sensibly be a friend to an AI agent. Being a friend to an AI would require caring about the good of the AI agent for its own sake, and it does not make sense to care about an AI agent in that way, since these agents lack a good of their own. After spelling out this argument, and responding to several objections, we highlight some initial implications of our argument, the most important of which is that the very idea of a tool – or, technological fix – to address social isolation and loneliness is misguided.

It is widely believed that AGI has the potential to be a wonderful tool that humans can use to meet our needs, solve our problems, and improve our lives. Against this view, we argue that any entity with truly general, human-level intelligence would have the capacity to lead its own life, with its own purposes and integrated hierarchy of goals. And thus any true AGI could not be merely a tool, even if it turned out to be extremely helpful for human beings. If we are correct, there is a dilemma at the heart of the ambition to build AGI as a valuable tool. On the one hand, any mere tool that we might build would lack capacities essential to the kind of general intelligence exhibited by human beings; it would not be genuine AGI. On the other hand, were we to create genuine AGI, then what we would have created would not be a mere tool, but something more. In this paper, we make the case for this dilemma. In so doing, we illuminate the connections between a set of core ideas: intelligence, agency, tools, and life.

Artificial Intelligence (AI) is predicted to play an increasingly central role in warfare, with weaponized robots taking over more battlefield operations, and military algorithms mediating in, or substituting for, human decision-making in areas such as intelligence collection and analysis, targeting, and strategic decision-making. The primary focus of this article is the potential impact of the widespread use of AI systems on soldiers and military leaders themselves – namely, on their moral character, skills, emotions, and agency – and on how they are perceived by the society which they defend. Increasing the use of AI in the military may undermine the conditions necessary for the cultivation and practice of martial virtues and commitment to the common good. The essay focuses on the three virtues of courage, deep loyalty, and prudence or practical wisdom. It then asks what kinds of “technocratic” vices might be encouraged in warfighters by the use of AI systems. It closes by analyzing the wider social and political implications of an expansive use of AI in the military. The risk is that, as AI takes over more military functions, the military ceases to be seen as a civic and republican institution and instead is seen as a mere extension of technocratic power.

The creation of machines with intelligence comparable to human beings—so-called "human-level” and “general” intelligence—is often regarded as the Holy Grail of Artificial Intelligence (AI) research. However, many prominent discussions of AI lean heavily on the notion of human-level intelligence to frame AI research, but then rely on conceptions of human cognitive capacities, including “common sense,” that are sketchy, one-sided, philosophically loaded, and highly contestable. Our goal in this essay is to bring into view some underappreciated features of the practical intelligence involved in ordinary human agency. These features of practical intelligence are implicit in the structure of our first-person experience of embodied and situated agency, deliberation, and human interaction. We argue that spelling out these features and their implications reveals a fundamental distinction between two forms of intelligence in action, or what we call “efficient task-completion” versus “intelligent engagement in activity.” This distinction helps us to see what is missing from some widely accepted ways of thinking about human-level intelligence in AI, and how human common sense is actually tied, conceptually, to the ideal of practical wisdom, or good (normative) judgment about how to act and live well. Finally, our analysis, if sound, also has implications for the important ethical question of what it means to have AI systems that are aligned with human values, or the so-called “value alignment” problem for artificial intelligence.

‘Skill models’ of ethical virtues offer a promising way of explaining the distinctive kind of ethical knowledge or understanding had by a virtuous person: virtues are akin to practical skills (in carpentry, sailing, musicianship, etc.) in that both are experience-based capacities of agency that yield non-codifiable knowledge of how-to-act-well in particular circumstances. This paper poses a puzzle for skill models of virtue concerning the non-deliberative character of much skillful and virtuous activity, and critiques two opposing ways of responding to the puzzle, reflecting two different skill models—Julia Annas’s intellectualist account and Hubert Dreyfus’s anti-intellectualism. The paper then offers an alternative skill model of virtue that draws on Wittgenstein’s remarks on pre-reflective perceptual discernment, and on a distinction between propositional (discursive) knowledge and a broader form conceptual understanding operative in the phenomenology of skillful agency. This view aims to respect what is true in Annas’s and Dreyfus’s views while avoiding the problems they encounter with non-deliberative action. It also reveals continuities between practical understanding and evaluative appreciation in ethical life and in other skillful activities, as well as important limits to discursive articulacy about these domains.

Propositionalism in the philosophy of action is the popular view that intentional actions are bodily movements caused and rationalized by certain ‘internal’ propositional attitude states that constitute the agent's perspective. I attack propositionalism's background claim that the genuinely mental/cognitive dimension of human action resides solely in some range of ‘internal’ agency-conferring representational states that causally trigger, and thus are always conceptually disentangle-able from, bodily activity itself. My opposing claim, following Ryle, Wittgenstein, and others, is that mentality and intentionality can be constitutively implicated in bodily actions themselves, as exercises of a distinctive form of embodied practical understanding. I attempt to show this by attending to the fine-grained contours of various skillful actions.

Popular Frankfurt-style theories of autonomy hold that (i) autonomy is motivation in action by psychological attitudes that have ‘authority’ to constitute the agent's perspective, and (ii) attitudes have this authority in virtue of their formal role in the individual's psychological system, rather than their substantive content. I pose a challenge to such ‘psychologistic’ views, taking Frankfurt's and Bratman's theories as my targets. I argue that motivation by attitudes that play the roles picked out by psychologistic theories is compatible with radically unintelligible behavior. Because of this, psychologistic views are committed to classifying certain agents as ‘autonomous’ whom we intuitively find to be dysfunctional. I then argue that a necessary condition for autonomy is that an agent's behavior is intelligible in a particular way: autonomy necessarily involves acting from a subjective practical perspective that is, in principle, minimally comprehensible to others – not simply in a causal sense, but in a substantive, socio-cultural sense

In this essay I argue that the mainstream ‘Standard Story’ of action – according to which actions are bodily motions with the right internal mental states as their causal triggers (e.g., ‘belief-desire-pairs’, ‘intentions’) – gives rise to a deeply problematic conception of inter-subjective action-understanding. For the Standard Story, since motivating reasons are internal mental states and bodily motions are not intrinsically intentional, an observer must ascribe internal states to others to make rational sense of their outwardly observable bodily motions. I argue this is both phenomenologically distorted and requires, on pain of infinite regress, a deeper, non-inferential, practical-perceptual form of understanding: ‘knowledge-how’, in a broadly Rylean sense. Recognizing the irreducible role of practical-perceptual knowledge-how in inter-subjective understanding, I argue, undermines core assumptions of the Standard Story concerning what an agent can directly perceive in interacting with others, and how our everyday practices of explaining actions with reasons function – and this opens the space for a radically opposed alternative view of inter-subjective action understanding.