Geoff Keeling

Many people feel compelled to interpret, describe, and respond to Large Language Models (LLMs) as if they possess inner mental lives sim- ilar to our own. Responses to this phenomenon have varied. Inflation- ists hold that at least some folk psychological ascriptions to LLMs are warranted. Deflationists argue that all such attributions of mentality to LLMs are misplaced, often cautioning against the risk that anthropomor- phic projection may lead to misplaced trust or potentially even confusion about the moral status of LLMs. We advance this debate by assessing two common deflationary arguments against LLM mentality. What we term the robustness strategy aims to undercut one justification for believing that LLMs are minded entities by showing that putatively cognitive and humanlike behaviours are not robust, failing to generalise appropriately. What we term the etiological strategy undercuts attributions of mental- ity by challenging naive causal explanations of LLM behaviours, offering alternative causal accounts that weaken the case for mental state attribu- tions. While both strategies offer powerful challenges to full-blown infla- tionism, we find that neither strategy provides a knock-down case against ascriptions of mentality to LLMs simpliciter. With this in mind, we ex- plore a modest form of inflationism that permits ascriptions of mentality to LLMs under certain conditions. Specifically, we argue that folk prac- tice provides a defeasible basis for attributing mental states and capacities to LLMs provided those mental states and capacities can be understood in metaphysically undemanding terms (e.g. knowledge, beliefs and de- sires), while greater caution is required when attributing metaphysically demanding mental phenomena such as phenomenal consciousness.

Credences are mental states corresponding to degrees of confidence in propositions. Attribution of credences to Large Language Models (LLMs) is commonplace in the empirical literature on LLM evaluation. Yet the theoretical basis for LLM credence attribution is unclear. We defend three claims. First, our semantic claim is that LLM credence attributions are (at least in general) correctly interpreted literally, as expressing truth-apt beliefs on the part of scientists that purport to describe facts about LLM credences. Second, our metaphysical claim is that the existence of LLM credences is at least plausible, although current evidence is inconclusive. Third, our epistemic claim is that LLM credence attributions made in the empirical literature on LLM evaluation are subject to non-trivial sceptical concerns. It is a distinct possibility that even if LLMs have credences, LLM credence attributions are generally false because the experimental techniques used to assess LLM credences are not truth-tracking. Our analysis has implications for the assessment of LLMs for properties like factuality and trustworthiness, and the broader scientific debate over the use of human cognition as a blueprint for understanding LLM behaviour.

The normative challenge of AI alignment centres upon what goals or values ought to be encoded in AI systems to govern their behaviour. A number of answers have been proposed, including the notion that AI must be aligned with human intentions or that it should aim to be helpful, honest and harmless. Nonetheless, both accounts suffer from critical weaknesses. On the one hand, they are incomplete: neither specification provides adequate guidance to AI systems, deployed across various domains with multiple parties. On the other hand, the justification for these approaches is questionable and, we argue, of the wrong kind. More specifically, neither approach takes seriously the need to justify the operation of AI systems to those affected by their actions – or what this means for pluralistic societies where people have different underlying beliefs about value. To address these limitations, we propose an alternative account of AI alignment that focuses on fair processes. We argue that principles that are the product of these processes are the appropriate target for alignment. This approach can meet the necessary standard of public justification, generate a fuller set of principles for AI that are sensitive to variation in context, and has explanatory power insofar as it makes sense of our intuitions about AI systems and points to a number of hitherto underappreciated ways in which an AI system may cease to be aligned.

“Moral imagination” is the capacity to register that one’s perspective on a decision-making situation is limited, and to imagine alternative perspectives that reveal new considerations or approaches. We have developed a Moral Imagination approach that aims to drive a culture of responsible innovation, ethical awareness, deliberation, decision-making, and commitment in organizations developing new technologies. We here present a case study that illustrates one key aspect of our approach – the technomoral scenario – as we have applied it in our work with product and engineering teams. Technomoral scenarios are fictional narratives that raise ethical issues surrounding the interaction between emerging technologies and society. Through facilitated role-playing and discussion, participants are prompted to examine their own intentions, articulate justifications for actions, and consider the impact of decisions on various stakeholders. This process helps developers to re-envision their choices and responsibilities, ultimately contributing to a culture of responsible innovation.

We propose a ‘Moral Imagination’ methodology to facilitate a culture of responsible innovation for engineering and product teams in technology companies. Our approach has been operationalized over the past two years at Google, where we have conducted over 50 workshops with teams from across the organization. We argue that our approach is a crucial complement to existing formal and informal initiatives for fostering a culture of ethical awareness, deliberation, and decision-making in technology design such as company principles, ethics and privacy review procedures, and compliance controls. We characterize some distinctive benefits of our methodology for the technology sector in particular.

The use of machine learning systems for decision-support in healthcare may exacerbate health inequalities. However, recent work suggests that algorithms trained on sufficiently diverse datasets could in principle combat health inequalities. One concern about these algorithms is that their performance for patients in traditionally disadvantaged groups exceeds their performance for patients in traditionally advantaged groups. This renders the algorithmic decisions unfair relative to the standard fairness metrics in machine learning. In this paper, we defend the permissible use of affirmative algorithms; that is, algorithms trained on diverse datasets that perform better for traditionally disadvantaged groups. Whilst such algorithmic decisions may be unfair, the fairness of algorithmic decisions is not the appropriate locus of moral evaluation. What matters is the fairness of final decisions, such as diagnoses, resulting from collaboration between clinicians and algorithms. We argue that affirmative algorithms can permissibly be deployed provided the resultant final decisions are fair.

The proper function of the heart is pumping the blood. According to what we call the type etiological view, this is because previous tokens of the type HEART were selected for pumping the blood. Nanay :412–431, 2010) argues that the type etiological view is viciously circular. He claims that the only plausible accounts of trait type individuation use proper functions, such that whenever the type etiological view is supplemented with a plausible account of trait type individuation, the result is a view that uses proper functions to explain proper functions. We refine this objection, and argue that Nanay at most establishes a potentially benign definitional circularity. However, we show that the type etiological view’s reliance on types nevertheless generates a vicious regress. Hence the type etiological view is false. We reject dispositional and modal alternatives to the type etiological view for the reason that they either cannot accommodate malfunction, or do so at the cost of proliferation; and then formulate a novel token etiological view that overcomes both problems because it makes no reference to trait types.

The application of machine-learning technologies to medical practice promises to enhance the capabilities of healthcare professionals in the assessment, diagnosis, and treatment, of medical conditions. However, there is growing concern that algorithmic bias may perpetuate or exacerbate existing health inequalities. Hence, it matters that we make precise the different respects in which algorithmic bias can arise in medicine, and also make clear the normative relevance of these different kinds of algorithmic bias for broader questions about justice and fairness in healthcare. In this paper, we provide the building blocks for an account of algorithmic bias and its normative relevance in medicine.

This paper presents a dilemma for the additive model of reasons. Either the model accommodates disjunctive cases in which one ought to perform some act $$\phi $$ just in case at least one of two factors obtains, or it accommodates conjunctive cases in which one ought to $$\phi $$ just in case both of two factors obtains. The dilemma also arises in a revised additive model that accommodates imprecisely weighted reasons. There exist disjunctive and conjunctive cases. Hence the additive model is extensionally inadequate. The upshot of the dilemma is that one of the most influential accounts of how reasons accrue to determine what we ought to do is flawed.

Lake et al. propose three criteria which, they argue, will bring artificial intelligence (AI) systems closer to human cognitive abilities. In this paper, we explore the application of these criteria to a particular domain of human cognition: our capacity for moral reasoning. In doing so, we explore a set of considerations relevant to the development of AI moral decision-making. Our main focus is on the relation between dual-process accounts of moral reasoning and model-free/model-based forms of machine learning. We also discuss how work in embodied and situated cognition could provide a valu- able perspective on future research.

Suppose a driverless car encounters a scenario where harm to at least one person is unavoidable and a choice about how to distribute harms between different persons is required. How should the driverless car be programmed to behave in this situation? I call this the moral design problem. Santoni de Sio defends a legal-philosophical approach to this problem, which aims to bring us to a consensus on the moral design problem despite our disagreements about which moral principles provide the correct account of justified harm. He then articulates an answer to the moral design problem based on the legal doctrine of necessity. In this paper, I argue that Santoni de Sio’s answer to the moral design problem does not achieve the aim of the legal-philosophical approach. This is because his answer relies on moral principles which, at least, utilitarians have reason to reject. I then articulate an alternative reading of the doctrine of necessity, and construct a partial answer to the moral design problem based on this. I argue that utilitarians, contractualists and deontologists can agree on this partial answer, even if they disagree about which moral principles offer the correct account of justified harm.

In his excellent essay, ‘Nudges in a post-truth world’, Neil Levy argues that ‘nudges to reason’, or nudges which aim to make us more receptive to evidence, are morally permissible. A strong argument against the moral permissibility of nudging is that nudges fail to respect the autonomy of the individuals affected by them. Levy argues that nudges to reason do respect individual autonomy, such that the standard autonomy objection fails against nudges to reason. In this paper, I argue that Levy fails to show that nudges to reason respect individual autonomy.