Rune Nyrup

Concerns about threats to human autonomy feature prominently in the field of AI ethics. One aspect of this concern relates to the use of AI systems for problematically manipulative influence. In response to this, the European Union’s draft AI Act (AIA) includes a prohibition on AI systems deploying subliminal techniques that alter people’s behavior in ways that are reasonably likely to cause harm (Article 5(1)(a)). Critics have argued that the term ‘subliminal techniques’ is too narrow to capture the target cases of AI-based manipulation. We propose a definition of ‘subliminal techniques’ that (a) is grounded on a plausible interpretation of the legal text; (b) addresses all or most of the underlying ethical concerns motivating the prohibition; (c) is defensible from a scientific and philosophical perspective; and (d) does not over-reach in ways that impose excessive administrative and regulatory burdens. The definition is meant to provide guidance for design teams seeking to pursue responsible and ethically aligned AI innovation.

Many authors have proposed constraining the behaviour of intelligent systems with ‘machine ethics’ to ensure positive social outcomes from the development of such systems. This paper critically analyses the prospects for machine ethics, identifying several inherent limitations. While machine ethics may increase the probability of ethical behaviour in some situations, it cannot guarantee it due to the nature of ethics, the computational limitations of computational agents and the complexity of the world. In addition, machine ethics, even if it were to be ‘solved’ at a technical level, would be insufficient to ensure positive social outcomes from intelligent systems.

This paper defends an account of explanatory reasoning generally, and inference to the best explanation in particular, according to which it first and foremost justifies pursuing hypotheses rather than accepting them as true. This side-steps the problem of why better explanations should be more likely to be true. I argue that this account faces no analogous problems. I propose an account of justification for pursuit and show how this provides a simple and straightforward connection between explanatoriness and justification for pursuit.

This article highlights a use of analogies in science that so far has received relatively little systematic discussion: providing reasons for pursuing a model or theory. Using the development of the liquid drop model as a test case, I critically assess two extant pursuit worthiness accounts: that analogies justify pursuit by supporting plausibility arguments and that analogies can serve as a guide to potential theoretical unification. Neither of these fit the liquid drop model case. Instead, I develop an alternative account, based on the idea that analogies facilitate the transfer of a well-understood modelling strategy to a new domain. 1Introduction2Case Study: The Development of the Liquid Drop Model3Plausibility Accounts 3.1Bartha on plausibility and analogical inference3.2Plausibility and the drop analogy4Analogies as a Guide to Unification5Generative Accounts 5.1Analogy-based modelling strategies5.2Did analogies play a merely generative role?6A New Pursuit Worthiness Account of Analogies 6.1Transferring understanding-with through analogies6.2Understanding-with and the liquid drop model7Conclusion

Explainable artificial intelligence (XAI) is an emerging, multidisciplinary field of research that seeks to develop methods and tools for making AI systems more explainable or interpretable. XAI researchers increasingly recognise explainability as a context-, audience- and purpose-sensitive phenomenon, rather than a single well-defined property that can be directly measured and optimised. However, since there is currently no overarching definition of explainability, this poses a risk of miscommunication between the many different researchers within this multidisciplinary space. This is the problem we seek to address in this paper. We outline a framework, called Explanatory Pragmatism, which we argue has two attractive features. First, it allows us to conceptualise explainability in explicitly context-, audience- and purpose-relative terms, while retaining a unified underlying definition of explainability. Second, it makes visible any normative disagreements that may underpin conflicting claims about explainability regarding the purposes for which explanations are sought. Third, it allows us to distinguish several dimensions of AI explainability. We illustrate this framework by applying it to a case study involving a machine learning model for predicting whether patients suffering disorders of consciousness were likely to recover consciousness.

I highlight a lacuna in recent debates concerning analogies in science. Most philosophers focus on analogical inferences as a way to justify accepting hypotheses. I argue that analogies play another important role, namely to justify pursuing hypotheses. In particular, I argue that both Paul Bartha's formal account and John Norton's material view of analogical inference, as they stand, fail to address this issue. I present my own account of justification for pursuit and show how analogies on this account can justify pursuit independently of whether they can also justify acceptance.

Transparency has been proposed as a way of handling value-ladenness in machine learning (ML). This article highlights limits to this strategy. I distinguish three kinds of transparency: epistemic transparency, retrospective value transparency, and prospective value transparency. This corresponds to different approaches to transparency in ML, including so-called explainable artificial intelligence and governance based on disclosing information about the design process. I discuss three sources of value-ladenness in ML—problem formulation, inductive risk, and specification gaming—and argue that retrospective value transparency is only well-suited for dealing with the first, while the third raises serious challenges even for prospective value transparency.

We distinguish three aspects of medical diagnosis: generating new diagnostic hypotheses, selecting hypotheses for further pursuit, and evaluating their probability in light of the available evidence. Drawing on Peirce’s account of abduction, we argue that hypothesis generation is amenable to normative analysis: physicians need to make good decisions about when and how to generate new diagnostic hypothesis as well as when to stop. The intertwining relationship between the generation and selection of diagnostic hypotheses is illustrated through the analysis of a detailed clinical case study. This interaction is not adequately captured by the existing probabilistic, decision-theoretic models of the threshold approach to clinical decision-making. Instead, we propose to conceptualize medical diagnosis in terms of strategic reasoning.

It is often observed that explanatory reasoning, especially the so-called inference to best explanation, plays a crucial role in theory choice in scientific practice. This paper defends an alternative account of the justificatory role of explanatory reasoning in general, and IBE in particular, inspired by C.S. Peirce's account of abduction. Most contemporary proponents of IBE take it to provide justification for accepting hypotheses as true. In contrast to this, the Peircean view defended here takes explanatory reasoning to first and foremost provide justification for pursuing hypotheses, i.e. to subject them to empirical testing. The standard view of IBE faces a problem of accounting for the connection between explanatoriness and truth-likeliness : Why should the fact that a hypothesis would be a good explanation if it were true give us any reason to regard it as actually true? The Peircean view avoids this problem since justification for pursuing a hypothesis H does not require showing H any more probable. Furthermore, the Peircean view faces no analogous challenge: I develop an account of pursuit where the connection between explanatoriness and justification for pursuit is simple and intuitive. Briefly: assuming that one important goal of science is to find good explanations, showing that H would provide a better explanation than previously thought also shows that it would be more valuable to learn whether H is actually true, thus increasing the justification for pursuing H. Finally, the Peircean view provides an alternative account of the role played by explanatory reasoning in many of the historical case studies cited by proponents of IBE. This undermines historical-empirical arguments for the reliability of explanatory reasoning as a guide to true hypotheses.

This thesis draws a distinction between reasoning about which scientific hypothesis to accept, reasoning concerned with generating new hypotheses and reasoning about which hypothesis to pursue. I argue that and should be evaluated according to the same normative standard, namely whether the hypotheses generated/selected are pursuit worthy. A consequentialist account of pursuit worthiness is defended, based on C. S. Peirce’s notion of ‘abduction’ and the ‘economy of research’, and developed as a family of formal, decision-theoretic models. This account is then deployed to discuss four more specific topics concerning scientific reasoning. First, I defend an account according to which explanatory reasoning mainly provides reasons for pursuing hypotheses, and criticise empirical arguments for the view that it also provides reasons for acceptance. Second, I discuss a number of pursuit worthiness accounts of analogical reasoning in science, arguing that, in some cases, analogies allow scientists to transfer an already well-understood modelling framework to a new domain. Third, I discuss the use of analogies within archaeological theorising, arguing that the distinction between using analogies for acceptance, generation and pursuit is implicit in methodological discussions in archaeology. A philosophical analysis of these uses is presented. Fourth, diagnostic reasoning in medicine is analysed from the perspective of Peircean abduction, where the conception of abduction as strategic reasoning is shown to be particularly important.

Simion and Kelp defend a non-anthropocentric account of trustworthy AI, based on the idea that the obligations of AI systems should be sourced in purely functional norms. In this commentary, I highlight some pressing counterexamples to their account, involving AI systems that reliably fulfil their functions but are untrustworthy because those functions are antagonistic to the interests of the trustor. Instead, I outline an alternative account, based on the idea that AI systems should not be considered primarily as tools but as technological participants in social practices. Specifically, I propose to source the obligations of an AI system in the norms that should govern the role it plays within the social practices it participates in, taking into account any changes to the social practices that its participation may bring about. This proposal is anthropocentric insofar as it ascribes obligations to AI systems that are similar to those of human participants in social practices, but only modestly so, as it does not require trustworthy AI to have contentious anthropomorphic capacities (e.g. for consciousness or moral responsibility).

Analogies often provide reasons for pursuing hypotheses or models. This is illustrated with a case study on the liquid drop model of the atomic nucleus. I criticise accounts in which analogies provide reasons for pursuit through epistemic support, proposing instead that analogies increase the value of learning the truth. I consider two accounts of this type: first, that analogies indicate potentials for theoretical unification; second, that analogies facilitate the transfer of already well-understood modelling frameworks to new domains. While the first is plausible for some cases, only the second can account for the liquid drop case study.

Most optimistic accounts of analogies in archaeology focus on cases where analogies lead to accurate or well-supported interpretations of the past. This chapter offers a complementary argument: analogies can also provide a valuable form of understanding of cultural and social phenomena when they fail, in the sense of either being shown inaccurate or the evidence being insufficient to determine their accuracy. This type of situation is illustrated through a case study involving the mortarium, a characteristic type of Roman pottery, and its relation to the so-called Romanization debate in Romano-British archaeology. I develop an account of comparative understanding, based on the idea that humans have a natural desire to understand ourselves comparatively, i.e., in terms of how we resemble and differ from societies at other times and places. Pursuing analogies can provide this type of understanding regardless of whether they turn out to be accurate. Furthermore, analogies can provide a similar form of understanding even when the evidence turns out to be insufficient to determine their accuracy.