Ethan Landes

A rift has opened between philosophical and non-philosophical perspectives on what kind of conceptual stance we can take towards artificial intelligence: are AI-powered machines the sort of things that can be trusted? Outside of philosophy, an ever-increasing body of research is looking at features of AI that promote trust and trustworthiness to understand how and why people “trust” AI. Within philosophy, this entire endeavor is widely seen as resting on a conceptual confusion about trust; trust requires the object of trust having agential capacities that modern AI lacks, and so whatever this body of research is studying, it is not, properly speaking, “trust”. Who is right? To shed light on this question, we explore the extent to which people thought sentences involving trusting different sorts of things make sense. We find that people do think it makes sense to talk about trusting AI, although less than it makes sense to trust other humans – even if they do not think AI has the sort of agential capacities philosophers think are required to be the object of trust. We conclude that either trust does not require agency or that there are multiple senses of trust, and neither conclusion is consistent with “trustworthy AI” being a conceptual confusion. While this undermines one criticism of AI and AI-related stakeholders, it nonetheless does not entail that AI is worthy of our trust.

As AI large language models (LLMs) become increasingly embedded in everyday technologies, should we be concerned about their capacity to influence human beliefs - particularly in the moral domain? Being persuaded because one is convinced by the LLM-generated reasons can support the moral and intellectual growth of users, while being persuaded because one defers to the LLM can prevent, or even reverse, growth and understanding. In three studies, we investigate whether and how people revise their moral judgments after receiving advice from LLMs. In Study 1, we find that despite rating human advisors as more trustworthy, participants were equally persuaded by LLMs in everyday moral dilemmas. In Study 2, we used a methodologically realistic paradigm in which participants interacted with a genuine LLM, finding that the LLM’s past performance and judged trustworthiness did not have an effect on its persuasiveness in everyday moral dilemmas. In Study 3, participants interacted with an LLM that defended its moral recommendation with good reasons, no reasons, or bad (i.e., patently absurd) reasons. While high-quality reasons do not increase persuasion relative to no reasons, bad reasons may actively undermine it. Our findings suggest that users defer to the LLM on a response-by-response basis, not based on past performance or the presence of high-quality reasons alone. That people defer to AI moral advice, even if not blindly, raises concerns about the effects of AI moral advisors - a heuristic of “this advice seems good enough” is not the way we should approach moral advice.

Conceptual engineering is a philosophical method that aims to design and spread conceptual and linguistic devices to cause meaningful changes in the world. So far, however, conceptual engineers have struggled to successfully spread the conceptual and linguistic entities they have designed to their target communities. This paper argues that conceptual engineering is far more likely to succeed if it incorporates empirical data and empirical methods. Because the causal factors influencing the successful propagation of linguistic or conceptual devices are as complicated and interwoven as they are, proper empirical research will greatly boost the likelihood that propagation is successful. In arguing for the superiority of empirical conceptual engineering over armchair-based conceptual engineering, this paper proposes a framework for understanding the causal forces at play in propagation. This is a three-part framework between the label of a lexical item, the psychological states associated with the lexical item, and the worldly things associated with the lexical item. By understanding the way causal forces affecting propagation play out at these three levels, conceptual engineers can better conceptualize, study, and harness the different causal forces affecting the success of their conceptual engineering projects.

Empathizing with another is perhaps the most human emotion of all and a long shot from the cold calculations of vectorized tokens that drive the inner workings of contemporary LLMs. This chapter looks beyond the questions of what to make of AI-generated “empathy” and instead asks whether AI can be used to develop our capacities for genuine human empathy. Empathy is not a static trait, instead capable of growth and development, and this paper explores whether AI can and, more importantly, should be used to increase one’s empathy. Nothing in principle stands in the way of AI improving our empathy, but the possibility raises unanswered questions about whether such an approach would be effective or would backfire in unexpected ways, such as encouraging the commodification of empathy as a technological tool that can make companies money, rather than a fundamental part of the human experience.

Conceptual engineering is the practice of revising concepts to improve how people talk and think. Its ability to improve talk and thought ultimately hinges on the successful dissemination of desired conceptual changes. Unfortunately, the field has been slow to develop methods to directly test what barriers stand in the way of propagation and what methods will most effectively propagate desired conceptual change. In order to test such questions, this paper introduces the masked time-lagged method. The masked time-lagged method tests people’s concepts at a later time than the intervention without participant’s knowledge, allowing us to measure conceptual revision in action. Using a masked time-lagged design on a content internalist framework, we attempted to revise planet and dinosaur in online participants to match experts’ concepts. We successfully revised planet but not dinosaur, demonstrating some of the difficulties conceptual engineers face. Nonetheless, this paper provides conceptual engineers, regardless of framework, with the tools to tackle questions related to implementation empirically and head-on.

Everyone who has come up with a new idea or concept faces the same problem: they need to call it something. This choice is simultaneously arbitrary and important. The new idea could have any name, but some names for it are clearly better than others. This paper builds upon existing research on the perceived relationship between label and word meaning to better understand the way that label choice affects language design, particularly conceptual engineering. Across three experiments, the paper finds that when learning novel concepts, the labels serving as vehicles for the novel concepts affect how people perceive the novel concepts’ content. In particular, the experiments find that labels influence people’s initial perception of novel concepts’ valence. Not all forms of information are readily extracted from labels, however, and subtle forms of information, particularly phonesthemes, do not appear to influence how people learn new concepts. Accordingly, conceptual engineers and other forms of language designers need to do careful research into how a given label will affect how their novel concept is interpreted by others.

This paper aims to clear up the epistemology of learning morality from artificial moral advisors (AMAs). We start with a brief consideration of what counts as moral enhancement and consider the risk of deskilling raised by machines that offer moral advice. We then shift focus to the epistemology of moral advice and show when and under what conditions moral advice can lead to enhancement. We argue that people’s motivational dispositions are enhanced by inspiring people to act morally, instead of merely telling them how to act. Drawing upon these insights, we claim that if AMAs are to genuinely enhance people morally, they should be designed as inspiration and not authority machines. In the final section, we evaluate existing AMA models to shed light on which holds the most promise for helping to make users better moral agents.

This paper aims to clear up the epistemology of learning morality from Artificial Moral Advisors (AMAs). We start with a brief consideration of what counts as moral enhancement and consider the risk of deskilling raised by machines that offer moral advice. We then shift focus to the epistemology of moral advice and show when and under what conditions moral advice can lead to enhancement. We argue that people’s motivational dispositions are enhanced by inspiring people to act morally, instead of merely telling them how to act. Drawing upon these insights, we claim that if AMAs are to genuinely enhance people morally, they should be designed as inspiration and not authority machines. In the final section, we evaluate existing AMA models to shed light on which holds the most promise for helping to make users better moral agents.

It is impossible to study philosophical methodology without being struck by the state of absolute chaos of the field’s methodological practices, methodological norms, and metaphilosophical beliefs. Not only are the methods of formal epistemology nothing like the methods of aesthetics, but even within specific debates and subfields, there are often significant disagreements about standards of proof, to say nothing about disagreements about the ultimate nature of the debate. The question facing metaphilosophers is whether this chaos is a feature or a bug. Is the chaos a part of philosophy’s value as an incubator for new ideas or a sign that many philosophers have lost their way and are holding back philosophical progress?

According to a prevailing view, conceptual engineering introduces a revolutionary philosophical methodology, challenging traditional conceptual analysis. However, in our paper, we argue that closer scrutiny reveals not only the falsity but also the inherent ambiguity of this narrative. We explore four interpretations of the "Anti-Novelty Claim", the claim that conceptual engineering is not a new way of doing philosophy. Discussing the Anti-Novelty Claim from the perspective of a text’s producer, the text’s consumers, and the exegetical potential of the text, we examine each perspective’s metaphilosophical implications and demonstrate that taking each perspective requires different methods. Adopting these different methods, we argue that the different interpretations of the Anti-Novelty Claim range from nearly trivially true to unlikely but untested. Importantly, we emphasize that each interpretation offers unique philosophical insights, yet addressing them requires diverse types of evidence, preventing a singular, straightforward answer to whether conceptual engineering is new.

Conceptual engineering is a philosophical method that aims to design and spread conceptual and linguistic devices to cause meaningful changes in the world. So far, however, conceptual engineers have struggled to successfully spread the conceptual and linguistic entities they have designed to their target communities. This paper argues that conceptual engineering is far more likely to succeed if it incorporates empirical data and empirical methods. Because the causal factors influencing successful propagation of linguistic or conceptual devices are as complicated and interwoven as they are, proper empirical research will greatly boost the likelihood that propagation is successful. In arguing for the superiority of empirical conceptual engineering over armchair-based conceptual engineering, this paper proposes a framework for understanding the causal forces at play in propagation. This is a three-part framework between the label of a lexical item, the psychological states associated with the lexical item, and the worldly things associated with the lexical item. By understanding the way causal forces affecting propagation play out at these three levels, conceptual engineers can better conceptualize, study, and harness the different causal forces affecting the success of their conceptual engineering projects.

Conceptual engineering is the practice of revising concepts to improve how people talk and think. Its ability to improve talk and thought ultimately hinges on the successful dissemination of desired conceptual changes. Unfortunately, the field has been slow to develop methods to directly test what barriers stand in the way of propagation and what methods will most effectively propagate desired conceptual change. In order to test such questions, this paper introduces the masked time-lagged method. The masked time-lagged method tests people’s concepts at a later time than the intervention without participant’s knowledge, allowing us to measure conceptual revision in action. Using a masked time-lagged design on a content internalist framework, we attempted to revise planet and dinosaur in online participants to match experts’ concepts. We successfully revised planet but not dinosaur, demonstrating some of the difficulties conceptual engineers face. Nonetheless, this paper provides conceptual engineers, regardless of framework, with the tools to tackle questions related to implementation empirically and head-on.

The last two decades have seen the proliferation of the empirical study of philosophy. This dissertation defends the practice and argues that to understand the way contingent features of the practice of philosophy affect the epistemic standing of philosophers, we need to draw upon a wider and more varied set of empirical data than is sometimes supposed. To explore this, the dissertation focuses on two places where the practices of the discipline of philosophy have an effect on the epistemology of philosophy. First, the dissertation discusses the interaction between notable works of philosophy and their readers. In particular, it critiques the method of defending the epistemic standing of philosophers through careful examination of notable works of philosophy to discern the methods in the text. Ultimately this method is epistemically unmotivated. It is instead far more important to study how people have interacted and reacted to works of philosophy. Second, the dissertation defends the use of lexicography in philosophy. Using "intuition" as a case study, the dissertation argues metasemantically and lexicographically that philosophers often use common words with meanings unique to philosophy. Through both discussions it is argued that experimental philosophers and epistemologists of philosophy need to drastically expand the sorts of data they collect and consider in their theorizing. ​

The beginning of the COVID-19 pandemic offers a unique case study for understanding conceptual and linguistic propagation. In early 2020, scientists, politicians, journalists, and other public figures had to, with great urgency, propagate several public health-related concepts and terms to every person they could. This paper examines the propagation of coronavirus and social distancing and develops a framework for understanding how the language used to express a notion can help or hinder propagation. I argue that anyone designing a representational device for propagation needs to appreciate the three-way causal relationship between language, people’s mental representations, and the extramental world. Using this framework, I explore what makes “social distancing” a bad name for social distancing and why it is unproblematic that “coronavirus” is a scientifically loose way of speaking about the virus. Through this and further study of the many historical examples of linguistic and conceptual propagation, conceptual engineers can better understand the complex challenges facing people who want to propagate representational devices.

In recent years, two competing methodological frameworks have developed in the study of the epistemology of philosophy. The traditional camp, led by experimental philosophy and its allies, has made inferences about the epistemology of philosophy based on the reactions, or intuitions, people have to works of philosophy. In contrast, multiple authors have followed the lead of Deutsch and Cappelen by setting aside experimental data in favor of inferences based on careful examination of the text of notable works of philosophy. In other words, the debate is split between authors focusing on philosophy’sconsumptionand those focusing on philosophy’sproduction. This paper examines the motivation for focusing on original texts and other evidence of philosophy’s production and finds it lacking. Drawing upon Hills’ distinction between propagation and transmission, I argue that the social epistemology of philosophy does not justify the recent focus on original texts of philosophy. Because the philosophical knowledge of consumers of philosophy is likelyinspiredby producers of philosophy, as opposed to epistemically grounded in the producers’ epistemic states, experimental philosophy had it right all along—if we want to know the epistemic standing of philosophy, we need to look to philosophy’s consumers.

The assumption that philosophers rely on intuitions to justify their philosophical positions has recently come under substantial criticism. In order to protect philosophy from experimental findings that suggest that intuitions are epistemically problematic, a number of metaphilosophers have argued that intuitions play no substantial epistemic role in philosophy. This paper focuses on attempts to deny intuitions’ epistemic role through exegetical analysis of original thought experiments. Using Deutsch’s particularly well-developed exegesis of Gettier’s 10 coin case as an exemplar of this method, I examine the challenges the strategy faces. I argue that intuition denial fails to provide a satisfactory account of how verdicts of thought experiments are justified. Instead, it commits intuition deniers to the conclusion that the arguments of Gettier, Kripke, Thomson, and others are bad arguments. As a result, rather than defusing challenges to the case method raised by experimental philosophers, intuition denial ultimately leads to the same troubling conclusion – that philosophers hold their positions on bad grounds.