Daniel C. Friedman

Prevailing approaches to interpreting large language models (LLMs) risk addressing the field’s central questions at the wrong level of analysis. As LLMs develop, researchers have turned to “under-thehood” methods to investigate whether LLMs possess states analogous to beliefs, desires, or intentions. These methods typically map internal representations, feature directions, or neural circuits onto folk-psychological categories. We argue that these methods are too fragile to reap the results we need. Instead, interpretability ought to be reoriented toward and grounded in what we call “ecologically sensitive behavioral analyses” which integrates mechanistic tools with severe testing of behavioral signatures across ecological contexts, aligning interpretability efforts with emerging evaluation sciences for AI. We outline a multi-disciplinary trajectory for developing interpretability methods that are both philosophically coherent and practically relevant for the deployment of increasingly capable LLMs.

What does it take to properly recognize someone as having made a scientific discovery? According to the 'Cognitivist', discovery attribution properly depends on the exercise of distinctive cognitive capacities such as competence, meta-reflective awareness, or domain-general understanding. Since AI systems lack such capacities, they cannot, on this view, be discoverers. If the Cognitivist is right, AI-driven science will be a markedly impoverished enterprise. Here, we argue otherwise. We develop an alternative, non-cognitivist conception of scientific discovery according to which discovery turns on successfully negotiating epistemic gaps. This reconception, we argue, better captures both familiar human cases and novel AI contributions, thereby re-framing the grounds for attributions of discovery in contemporary science. AI systems, we argue, can be appropriately attributed scientific discoveries. Along the way we develop a general moral for philosophical reflection in the age of AI-infused science.

When experts make judgments that inform public policy, what kinds of reasons should they consider in order to provide informed and responsible recommendations? Extant discussions of this question typically focus on the role of non-epistemic values in the evaluation of scientific hypotheses. However, the kinds of epistemic reasons that should undergird such assessments have received comparatively little attention. This paper argues that evidence is not the only kind of epistemic reason important for guiding expert judgment. Experts must also be sensitive to two kinds of social epistemic reasons, which have so far been overlooked in discussions on expertise. First, experts must be sensitive to higher-order evidence about other experts' views, including about disagreements between experts. Second, they must also be sensitive to inquisitive reasons concerning, for example, the pursuitworthiness of various research programs in the field, or the distribution of labor across those research programs. The paper illustrates the fruitfulness of this account by showing the importance of these overlooked reasons for understanding examples of expert judgment from the history of science and contemporary scientific practice.

We inquire together all the time, yet the norms of such inquiring are poorly understood. Parallels from norms of individual inquiry fall short in accurately characterizing our inquiring together. The need then for an account of inquiring together which provides normative guidance is pressing. This paper unpacks and defends a version of a crucial norm of such inquiry, inspired by Harman (1986), which codifies the kind of evidence necessary for a shared inquirer to permissibly settle her shared question. It is informed by the demands of directed openness between inquirers, characteristic of much shared activity generally, and shared inquiry specifically. In defending this norm, a general account of shared inquiry is developed, one which sheds light on the complex and surprising relationship between human cooperation and epistemic rationality.

What do we owe those with whom we inquire? Presumably, quite a bit. Anything beyond what is necessary to secure knowledge? Yes. In this paper, I argue for a class of ‘zetetic rights.’ These are rights distinctive to participants in group inquiry. Zetetic rights help protect important central interests of inquirers. These include a right to aid, a right against interference, and a right to exert influence over the course of inquiry. Building on arguments by Fricker (2015), I defend these rights, and explore cases of their violation: zetetic wrongings. I argue that zetetic wrongings constitute a distinctive, ubiquitous, and significant form of injustice in our epistemic and zetetic lives. To improve our inquiries and avoid epistemic injustice, we need to contend with zetetic wronging. Making this case helps show that a complete picture of epistemic life must account for the deep influence of our cooperative ties to one another.

This paper defends the Intention Account of Inquiry. On this account, inquiry is best understood by appeal to a ‘question-directed intention’ (QDI), an intention to answer a question broadly construed. This account’s core commitments help meet recent challenges plaguing extant approaches to characterizing inquiry. First, QDIs are the type of mental state central to inquiry, not attitudes like curiosity or wonder. Second, holding a QDI towards a question and acting in service of it constitutes the start of inquiry. Third, controversial norms which mandate a rational inquirer's ignorance towards the answer to her question can be reformulated and defended by appeal to rational constraints on intention. Fourth, instrumental pressures inquirers face are the standard pressures of plan-rationality. In defending these theses, I show that the Intention Account provides compelling explanations to standing challenges, in ways competitors cannot. It does so by advancing understanding of how our epistemic and practical agency are intertwined.

Scientific disagreements are an important catalyst for scientific progress. But what happens when scientists disagree amidst times of crisis, when we need quick yet reliable policy guidance? In this paper we provide a normative account for how scientists facing disagreement in the context of ‘fast science’ should respond, and how policy makers should evaluate such disagreement. Starting from an argumentative, pragma-dialectic account of scientific controversies, we argue for the importance of ‘higher-order evidence’ (HOE) and we specify desiderata for scientifically relevant HOE. We use our account to analyze the controversy about the aerosol transmission of COVID-19.