Trust is a central dimension in the relation between human beings and technologies. In many discourses about technology, the relation between human beings and technologies is conceptualized as an external relation: a relation between pre-given entities that can have an impact on each other but that do not mutually constitute each other. From this perspective, relations of trust can vary between _reliance_, as is present for instance in technological extensionism, and _suspicion_, as in various precautionary approaches in ethics that focus (...) on technological risks. Against these two interpretations of trust, this article develops a third one. Based on a more internal account of the relations between human beings and technologies, it becomes possible to see that every technological development puts at stake what it means to be a human being. Using technologies, then, implies trusting ourselves _to_ technologies. We argue that this does not imply an uncritical subjection to technology. Rather, recognizing that technologies help to constitute human subjectivity implies that human beings can get actively involved in processes of technological mediation. Trust then has the character of _confidence_: deliberately trusting oneself _to_ technology. (shrink)

I present and defend an account of three-place trustworthiness according to which B is trustworthy with respect to A in domain of interaction D, if and only if she is competent with respect to that domain, and she would take the fact that A is counting on her, were A to do so in this domain, to be a compelling reason for acting as counted on. This is not the whole story of trustworthiness, however, for we want those we can (...) count on to identify themselves so that we can place our trust wisely. (shrink)

Trust is difficult to define. Instead of doing so, I propose that the best way to understand the concept is through a genealogical account. I show how a root notion of trust arises out of some basic features of what it is for humans to live socially, in which we rely on others to act cooperatively. I explore how this concept acquires resonances of hope and threat, and how we analogically apply this in related but different contexts. The genealogical account (...) explains both why the notion has such value for us and why it is difficult to define. (shrink)

In this essay, we argue that trust is a dynamic emotional relationship which entails responsibility. Trust is not a social substance, a medium, or a mysterious entity but rather a set of social practices, defined by our choices, to trust or not to trust. We discuss the differences and the relationship between trust and trustworthiness, and we distinguish several different kinds or “levels” of trust, simple trust, basic trust, “blind” trust, and authentic trust. We then argue that trust as an (...) emotional practice, can be “willful,” voluntary and amatter of personal responsibility. (shrink)

Reasons are given to justify the claim that computer simulations and computational science constitute a distinctively new set of scientific methods and that these methods introduce new issues in the philosophy of science. These issues are both epistemological and methodological in kind.

Artificial agents (AAs), particularly but not only those in Cyberspace, extend the class of entities that can be involved in moral situations. For they can be conceived of as moral patients (as entities that can be acted upon for good or evil) and also as moral agents (as entities that can perform actions, again for good or evil). In this paper, we clarify the concept of agent and go on to separate the concerns of morality and responsibility of agents (most (...) interestingly for us, of AAs). We conclude that there is substantial and important scope, particularly in Computer Ethics, for the concept of moral agent not necessarily exhibiting free will, mental states or responsibility. This complements the more traditional approach, common at least since Montaigne and Descartes, which considers whether or not (artificial) agents have mental states, feelings, emotions and so on. By focussing directly on mind-less morality we are able to avoid that question and also many of the concerns of Artificial Intelligence. A vital component in our approach is the Method of Abstraction for analysing the level of abstraction (LoA) at which an agent is considered to act. The LoA is determined by the way in which one chooses to describe, analyse and discuss a system and its context. The Method of Abstraction is explained in terms of an interface or set of features or observables at a given LoA. Agenthood, and in particular moral agenthood, depends on a LoA. Our guidelines for agenthood are: interactivity (response to stimulus by change of state), autonomy (ability to change state without stimulus) and adaptability (ability to change the transition rules by which state is changed) at a given LoA. Morality may be thought of as a threshold defined on the observables in the interface determining the LoA under consideration. An agent is morally good if its actions all respect that threshold; and it is morally evil if some action violates it. That view is particularly informative when the agent constitutes a software or digital system, and the observables are numerical. Finally we review the consequences for Computer Ethics of our approach. In conclusion, this approach facilitates the discussion of the morality of agents not only in Cyberspace but also in the biosphere, where animals can be considered moral agents without their having to display free will, emotions or mental states, and in social contexts, where systems like organizations can play the role of moral agents. The primary cost of this facility is the extension of the class of agents and moral agents to embrace AAs. (shrink)

A series of recent developments points towards auditing as a promising mechanism to bridge the gap between principles and practice in AI ethics. Building on ongoing discussions concerning ethics-based auditing, we offer three contributions. First, we argue that ethics-based auditing can improve the quality of decision making, increase user satisfaction, unlock growth potential, enable law-making, and relieve human suffering. Second, we highlight current best practices to support the design and implementation of ethics-based auditing: To be feasible and effective, ethics-based auditing (...) should take the form of a continuous and constructive process, approach ethical alignment from a system perspective, and be aligned with public policies and incentives for ethically desirable behaviour. Third, we identify and discuss the constraints associated with ethics-based auditing. Only by understanding and accounting for these constraints can ethics-based auditing facilitate ethical alignment of AI, while enabling society to reap the full economic and social benefits of automation. (shrink)

A series of recent developments points towards auditing as a promising mechanism to bridge the gap between principles and practice in AI ethics. Building on ongoing discussions concerning ethics-based auditing, we offer three contributions. First, we argue that ethics-based auditing can improve the quality of decision making, increase user satisfaction, unlock growth potential, enable law-making, and relieve human suffering. Second, we highlight current best practices to support the design and implementation of ethics-based auditing: To be feasible and effective, ethics-based auditing (...) should take the form of a continuous and constructive process, approach ethical alignment from a system perspective, and be aligned with public policies and incentives for ethically desirable behaviour. Third, we identify and discuss the constraints associated with ethics-based auditing. Only by understanding and accounting for these constraints can ethics-based auditing facilitate ethical alignment of AI, while enabling society to reap the full economic and social benefits of automation. (shrink)

This article considers the issue of opacity as a problem for socially consequential mechanisms of classification and ranking, such as spam filters, credit card fraud detection, search engines, news trends, market segmentation and advertising, insurance or loan qualification, and credit scoring. These mechanisms of classification all frequently rely on computational algorithms, and in many cases on machine learning algorithms to do this work. In this article, I draw a distinction between three forms of opacity: opacity as intentional corporate or state (...) secrecy, opacity as technical illiteracy, and an opacity that arises from the characteristics of machine learning algorithms and the scale required to apply them usefully. The analysis in this article gets inside the algorithms themselves. I cite existing literatures in computer science, known industry practices, and do some testing and manipulation of code as a form of lightweight code audit. I argue that recognizing the distinct forms of opacity that may be coming into play in a given application is a key to determining which of a variety of technical and non-technical solutions could help to prevent harm. (shrink)

We are sceptical of concerns over the opacity of algorithmic decision tools. While transparency and explainability are certainly important desiderata in algorithmic governance, we worry that automated decision-making is being held to an unrealistically high standard, possibly owing to an unrealistically high estimate of the degree of transparency attainable from human decision-makers. In this paper, we review evidence demonstrating that much human decision-making is fraught with transparency problems, show in what respects AI fares little worse or better and argue that (...) at least some regulatory proposals for explainable AI could end up setting the bar higher than is necessary or indeed helpful. The demands of practical reason require the justification of action to be pitched at the level of practical reason. Decision tools that support or supplant practical reasoning should not be expected to aim higher than this. We cast this desideratum in terms of Daniel Dennett’s theory of the “intentional stance” and argue that since the justification of action for human purposes takes the form of intentional stance explanation, the justification of algorithmic decisions should take the same form. In practice, this means that the sorts of explanations for algorithmic decisions that are analogous to intentional stance explanations should be preferred over ones that aim at the architectural innards of a decision tool. (shrink)

Many of the computing systems programmed using Machine Learning are opaque: it is difficult to know why they do what they do or how they work. Explainable Artificial Intelligence aims to develop analytic techniques that render opaque computing systems transparent, but lacks a normative framework with which to evaluate these techniques’ explanatory successes. The aim of the present discussion is to develop such a framework, paying particular attention to different stakeholders’ distinct explanatory requirements. Building on an analysis of “opacity” from (...) philosophy of science, this framework is modeled after accounts of explanation in cognitive science. The framework distinguishes between the explanation-seeking questions that are likely to be asked by different stakeholders, and specifies the general ways in which these questions should be answered so as to allow these stakeholders to perform their roles in the Machine Learning ecosystem. By applying the normative framework to recently developed techniques such as input heatmapping, feature-detector visualization, and diagnostic classification, it is possible to determine whether and to what extent techniques from Explainable Artificial Intelligence can be used to render opaque computing systems transparent and, thus, whether they can be used to solve the Black Box Problem. (shrink)

I have a very particular reason to be grateful to Stewart Sutherland, our late President, which is connected to some of the themes of this lecture, so want to begin by recalling a long conversation I had with him on these topics.

Western societies are experiencing a crisis of trust: we no longer enjoy high levels of confidence in social institutions and are increasingly skeptical of those holding positions of authority. The crisis of trust, however, seems paradoxical: at the same time we report greater feelings of mistrust or an erosion of trust in institutions and technologies we increasingly entrust our wellbeing and security to these very same technologies and institutions. Analyzing trust not only will help resolve the paradox but suggests that (...) the crisis entails normative expectations linked to collective identity, and as such, trust can be understood as a public virtue. (shrink)

One of the most noticeable trends in recent years has been the increasing reliance of public decision-making processes on algorithms, i.e. computer-programmed step-by-step instructions for taking a given set of inputs and producing an output. The question raised by this article is whether the rise of such algorithmic governance creates problems for the moral or political legitimacy of our public decision-making processes. Ignoring common concerns with data protection and privacy, it is argued that algorithmic governance does pose a significant threat (...) to the legitimacy of such processes. Modelling my argument on Estlund’s threat of epistocracy, I call this the ‘threat of algocracy’. The article clarifies the nature of this threat and addresses two possible solutions. It is argued that neither solution is likely to be successful, at least not without risking many other things we value about social decision-making. The result is a somewhat pessimistic conclusion in which we confront the possibility that we are creating decision-making processes that constrain and limit opportunities for human participation. (shrink)

In this paper we raise the question whether technological artifacts can properly speaking be trusted or said to be trustworthy. First, we set out some prevalent accounts of trust and trustworthiness and explain how they compare with the engineer’s notion of reliability. We distinguish between pure rational-choice accounts of trust, which do not differ in principle from mere judgments of reliability, and what we call “motivation-attributing” accounts of trust, which attribute specific motivations to trustworthy entities. Then we consider some examples (...) of technological entities that are, at first glance, best suited to serve as the objects of trust: intelligent systems that interact with users, and complex socio-technical systems. We conclude that the motivation-attributing concept of trustworthiness cannot be straightforwardly applied to these entities. Any applicable notion of trustworthy technology would have to depart significantly from the full-blown notion of trustworthiness associated with interpersonal trust. (shrink)

In this paper I argue that the search for explainable models and interpretable decisions in AI must be reformulated in terms of the broader project of offering a pragmatic and naturalistic account of understanding in AI. Intuitively, the purpose of providing an explanation of a model or a decision is to make it understandable to its stakeholders. But without a previous grasp of what it means to say that an agent understands a model or a decision, the explanatory strategies will (...) lack a well-defined goal. Aside from providing a clearer objective for XAI, focusing on understanding also allows us to relax the factivity condition on explanation, which is impossible to fulfill in many machine learning models, and to focus instead on the pragmatic conditions that determine the best fit between a model and the methods and devices deployed to understand it. After an examination of the different types of understanding discussed in the philosophical and psychological literature, I conclude that interpretative or approximation models not only provide the best way to achieve the objectual understanding of a machine learning model, but are also a necessary condition to achieve post hoc interpretability. This conclusion is partly based on the shortcomings of the purely functionalist approach to post hoc interpretability that seems to be predominant in most recent literature. (shrink)

There is widespread agreement that there should be a principle requiring that artificial intelligence be ‘explicable’. Microsoft, Google, the World Economic Forum, the draft AI ethics guidelines for the EU commission, etc. all include a principle for AI that falls under the umbrella of ‘explicability’. Roughly, the principle states that “for AI to promote and not constrain human autonomy, our ‘decision about who should decide’ must be informed by knowledge of how AI would act instead of us” :689–707, 2018). There (...) is a strong intuition that if an algorithm decides, for example, whether to give someone a loan, then that algorithm should be explicable. I argue here, however, that such a principle is misdirected. The property of requiring explicability should attach to a particular action or decision rather than the entity making that decision. It is the context and the potential harm resulting from decisions that drive the moral need for explicability—not the process by which decisions are reached. Related to this is the fact that AI is used for many low-risk purposes for which it would be unnecessary to require that it be explicable. A principle requiring explicability would prevent us from reaping the benefits of AI used in these situations. Finally, the explanations given by explicable AI are only fruitful if we already know which considerations are acceptable for the decision at hand. If we already have these considerations, then there is no need to use contemporary AI algorithms because standard automation would be available. In other words, a principle of explicability for AI makes the use of AI redundant. (shrink)

One staple of living in our information society is having access to the web. Web-connected devices interpret our queries and retrieve information from the web in response. Today’s web devices even purport to answer our queries directly without requiring us to comb through search results in order to find the information we want. How do we know whether a web device is trustworthy? One way to know is to learn why the device is trustworthy by inspecting its inner workings, 156–170 (...) 1995; Humphreys 2004, Episteme, 6, 221–229 2009). But ordinary users of web devices cannot inspect their inner workings because of their scale, complexity, and the corporate secrecy which enshrouds both the procedures by which the devices operate and the companies that make them. Further piling on this predicament, authors have criticized web technology on the grounds that the invisibility of the web devices’ inner workings prevents users from critically assessing the procedures that produce a given output, in some cases, barring users from fulfilling their epistemic responsibilities, 343–355 2010; Miller and Record Episteme, 10, 117–134 2013). I consider four broad kinds of reasons which we can acquire without inspecting the inner workings of black-boxed technology: individual understanding, expert testimony, testing through experience, and social vetting; and show how each is a viable method of appraising black-boxed technology. By deploying these methods, we can remain responsible inquirers while nonetheless benefitting from today’s epistemic resources on the web. (shrink)

A socio-cognitive approach to trust can help us envisage a notion of networked trust for multi-agent systems (MAS) based on different interacting agents. In this framework, the issue is to evaluate whether or not a socio-cognitive analysis of trust can apply to the interactions between human and autonomous agents. Two main arguments support two alternative hypothesis; one suggests that only reliance applies to artificial agents, because predictability of agents’ digital interaction is viewed as an absolute value and human relation is (...) judged to be a necessary requirement for trust. The other suggests that trust may apply to autonomous agents because predictability of agents’ interaction is viewed only as a relative value since the digital normativity that grows out of the communication process between interacting agents in MAS has always deal with some unpredictable outcomes (_reduction of uncertainty_). Furthermore, human touch is not judged to be a necessary requirement for trust. In this perspective, a diverse notion of trust is elaborated, as trust is no longer conceived only as a relation between interacting agents but, rather, as a relation between cognitive states of control and lack of control (_double bind_). (shrink)

It is argued that the question “Can we trust technology?” is unanswerable because it is open-ended. Only questions about specific issues that can have specific answers should be entertained. It is further argued that the reason the question cannot be answered is that there is no such thing as Technology _simpliciter_. Fundamentally, the question comes down to trusting people and even then, the question has to be specific about trusting a person to do this or that.

In our information societies, tasks and decisions are increasingly outsourced to automated systems, machines, and artificial agents that mediate human relationships, by taking decisions and acting on the basis of algorithms. This raises a critical issue: how are algorithmic procedures and applications to be appraised and governed? This question needs to be investigated, if one wishes to avoid the traps of ICTs ending up in isolating humans behind their screens and digital delegates, or harnessing them in a passive role, by (...) curtailing their freedom and autonomy. (shrink)