Deep neural networks are said to be opaque, impeding the development of safe and trustworthy artificial intelligence, but where this opacity stems from is less clear. What are the sufficient properties for neural network opacity? Here, I discuss five common properties of deep neural networks and two different kinds of opacity. Which of these properties are sufficient for what type of opacity? I show how each kind of opacity stems from only one of these five properties, and then discuss to (...) what extent the two kinds of opacity can be mitigated by explainability methods. (shrink)

In the past decade, the fields of machine learning and artificial intelligence (AI) have seen unprecedented developments that raise human-machine interactions (HMI) to the next level.Smart machines, i.e., machines endowed with artificially intelligent systems, have lost their character as mere instruments. This, at least, seems to be the case if one considers how humans experience their interactions with them. Smart machines are construed to serve complex functions involving increasing degrees of freedom, and they generate solutions not fully anticipated by humans. (...) Consequently, their performances show a touch of action and even autonomy. HMI is therefore often described as a sort of “cooperation” rather than as a mere application of a tool. Some authors even go as far as subsuming cooperation with smart machines under the label ofpartnership, akin to cooperation between human agents sharing a common goal. In this paper, we explore how far the notion of shared agency and partnership can take us in our understanding of human interaction with smart machines. Discussing different topoi related to partnerships in general, we suggest that different kinds of “partnership” depending on the form of interaction between agents need to be kept apart. Building upon these discussions, we propose a tentative taxonomy of different kinds of HMI distinguishing coordination, collaboration, cooperation, and social partnership. (shrink)

Much discussion of the ethics of algorithms has focused on harms to autonomy—especially harms stemming from manipulation. Nonetheless, although manipulation can often be harmful, we suggest that in certain contexts it may not impair autonomy. To fully assess the impact of algorithm design on autonomy, we argue for a need to move beyond a focus on manipulation towards a multidimensional account of autonomy itself. Drawing on the autonomy literature and recent data ethics, we propose a novel account which takes autonomy (...) to supervene on three distinct but related elements: agency, authenticity and individual control over decision-making. By distinguishing autonomy from control, the account can explain the variable effects of manipulation on user autonomy within algorithm-driven systems. In particular, it can explain why improving user control may improve autonomy in some contexts, while in other contexts—such as in some kinds of newsfeeds—certain algorithm designs that instead _reduce_ user control may nevertheless improve autonomy. As a result, the account can accommodate the sometimes convoluted interplay between control, autonomy, manipulation, and commercial versus prosocial design goals. (shrink)

Research under real-world conditions is crucial to the development and deployment of robust AI systems. Exposing large language models to complex use settings yields knowledge about their performance and impact, which cannot be obtained under controlled laboratory conditions or through anticipatory methods. This epistemic need for real-world research is exacerbated by large-language models’ opaque internal operations and potential for emergent behavior. However, despite its epistemic value and widespread application, the ethics of real-world AI research has received little scholarly attention. To (...) address this gap, this paper provides an analysis of real-world research with LLMs and generative AI, assessing both its epistemic value and ethical concerns such as the potential for interpersonal and societal research harms, the increased privatization of AI learning, and the unjust distribution of benefits and risks. This paper discusses these concerns alongside four moral principles influencing research ethics standards: non-maleficence, beneficence, respect for autonomy, and distributive justice. I argue that real-world AI research faces challenges in meeting these principles and that these challenges are exacerbated by absent or imperfect current ethical governance. Finally, I chart two distinct but compatible ways forward: through ethical compliance and regulation and through moral education and cultivation. (shrink)

Many digital technologies, designed and controlled by intensive data-driven corporate platforms, have become ubiquitous for many of our daily activities. This has raised political and ethical concerns over how they might be threatening our personal autonomy. However, not much philosophical attention has been paid to the specific role that their hyper-designed (sensorimotor) interfaces play in this regard. In this paper, we aim to offer a novel framework that can ground personal autonomy on sensorimotor interaction and, from there, directly address how (...) technological design affects personal autonomy. To do this, we will draw from enactive sensorimotor approaches to cognition, focusing on the central notion of habits, understood as sensorimotor schemes that, in networked relations, give rise to sensorimotor agency. Starting from sensorimotor agency as a basis for more complex forms of personal autonomy, our approach gives us grounds to analyse our relationship with technology (in general) and to distinguish between autonomy-enhancing and autonomy-diminishing technologies. We argue that, by favouring/obstructing the enactment of certain (networks of) habits over others, technologies can directly act upon our personal autonomy, locally and globally. With this in mind, we then discuss how current digital technologies are often being designed to be autonomy-diminishing (as is the case of “dark patterns” in design), and sketch some ideas on how to build more autonomy-enhancing digital technologies. (shrink)

Aylsworth and Castro’s recent paper, Should I Use ChatGPT to Write My Papers?, argues that students in the humanities have a moral obligation to refrain from using AI tools such as ChatGPT for writing assignments. Their claim is that writing is an autonomy-fostering activity, essential for intellectual growth and critical reflection, and that every agent has the moral duty to respect their own autonomy. While the authors raise significant ethical concerns, the paper lacks the identification of which specific features of (...) writing promote autonomy. Writing is a multifaceted process which involves distinct activities, not all of which equally contribute to fostering autonomy. Furthermore, writing can be a shared endeavour, for example when co-authoring a paper with a colleague. This prompts the question of which features or activities exactly are the ones to support autonomy. The usage in AI in education is a double-edged sword: while some scholars emphasize the limitations of generated texts, others have reported positive outcomes from using AI tools in educational settings. This invites a deeper examination of how technology can be designed or used to support, rather than mine, students’ autonomy. (shrink)