Randomized controlled trials (RCTs) are widely employed to inform policy decisions in the social and behavioral sciences. Yet both philosophers and scientists have questioned the reliability of effect-size evidence derived from RCTs for policy extrapolation, arguing that RCT findings should be complemented by evidence about the mechanism through which a policy works. We agree that mechanisms are important but argue that existing accounts face the challenge of mechanism individuation. In this article, we explain what the challenge consists in, focusing on common practices of policy scaling-up, transferring policy to new domains, and aggregating policy-relevant evidence.

The contemporary debate over scientific progress is fragmented into four competing, independent approaches, making systematic comparison difficult. To resolve this, we propose a formal meta-theory of progress, within which the different approaches can be represented and compared. Generalizing the formal core of truthlikeness theory, we develop a neutral meta-theory of “target-likeness.” We demonstrate how to represent the competing approaches as alternative measures of target-likeness. We conclude by discussing the implications and applications of our theory both inside and outside the current debate on scientific progress.

Scientific models can be performative: Their uptake may alter the phenomena they represent. Existing debates on performativity have not provided a criterion for model selection when models perceived to be in admissibility parity differ in their foreseeable performative effects. Using game-theoretic models of proenvironmental behavior as a test case, I introduce the One-Step Principle: a conservative constraint on performative model selection. The principle permits replacing a pessimistic model with a more optimistic alternative only under conditions of salience, favorable conditions for self-fulfilling performativity, and proximity to the model it replaces.

This article examines how biological models are combined in biomedical research. It analyzes the notion of epistemological space, understood not as an abstract space of epistemic possibilities but as a framework for coordinating heterogeneous models through zones of partial compatibility structured by shared features, biological constraints, and research objectives. It also questions the ways in which models are combined, highlighting the methodological choices and epistemic commitments that guide pluralistic modeling practices. Finally, it proposes a representation of these modalities as a heuristic tool for comparing and reflecting on strategies of model combination.

Camouflage is a ubiquitous phenomenon in nature, yet most researchers in animal communication have overlooked it because it does not typically qualify as a signal. I argue against this prevailing view and show that at least one type of camouflage, masquerade, can be regarded as a signal. I then introduce two notions—information-transmitting signals and information-masking signals—and characterize them in such a way that typical signals fall into the former category and camouflage into the latter, drawing on an information-theoretic approach to animal signaling. Lastly, I compare information-masking signals with misinformation and deception.

Cognitive scientists have applied advances in our formal understanding of causation in philosophy and computer science, particularly within the Causal Bayes Net formalism, to understand human causal learning. In parallel, reinforcement learning (RL) researchers have developed the idea of an intrinsic reward signal called “empowerment.” An agent is rewarded for maximizing the mutual information between its actions and their outcomes, regardless of the external reward value of those outcomes. Empowerment may be an important bridge between classical Bayesian causal learning and RL and may help to characterize causal learning in humans and enable it in machines.

In October 1950, Alan Turing proposed that a machine’s success in an imitation game provides a criterion for thinking. Turing’s test became one of the philosophical foundations of AI, but three-quarters of a century later, it is widely claimed that generative AI models pass the test, rendering it obsolete. Our aim here is to establish definitively Turing’s conceptual design for his game. We argue that influential interpretations of Turing’s design are mistaken, with the consequence that his test has not been passed. We also present the results of carrying out, for the first time, Turing’s experiment as he proposed it.

Predictive benchmarking, evaluating machine learning models based on predictive performance and competitive ranking, is central to machine learning research and scientific inquiry. However, benchmark scores at best measure performance relative to a specific dataset and learning problem. Drawing substantial scientific inferences requires additional assumptions. Adapting ideas from psychological validity theory, we propose validity conditions that make these assumptions explicit. In two case studies—ImageNet and the Fragile Families Challenge—we show how benchmark results can support inferences about research progress and limits of predictability, situating predictive benchmarking as a distinct epistemic practice in machine learning.

This article proposes restricting selected effect (SE) proper functions to effectively fixed traits—traits at or close to fixation in the population. SE functions, as historical fitness contributions in general, can only ground gradual normative standards at the character-state level, whereas proper functions are further expected to ground a stronger normativity at the character level. Proper function thus requires the trait’s fitness contributions to be sufficiently strong and sustained, and fixation provides a principled standard for such contributions. The restricted account better addresses the explanatory challenge to proper functions while preserving their theoretical goals of grounding robust teleology and normativity.

Text-analytic methods increasingly promise to transform surviving textual traces into measurements of past psychological attributes. This paper argues that such claims can mistake technical success for representational warrant. A scoring regime may perform well within a constructed evidential space without thereby validating the construct-to-trace relation required for measurement. The central problem is endogenous validation: The same operations that make texts scoreable also delimit the evidence by which success is assessed. Strong measurement claims therefore require credible conditions under which the underlying representational mapping could fail.

Although scientific narratives have often been studied in relation to causality and explanation, not much attention has been given to their noncausal content. This less explored aspect of narratives is addressed here. I argue that colligatory inferences underlie narrative connections and become significant when they help constitute new scientific phenomena from scattered information. Through a sedimentological case study, I identify kinship relations as noncausal connections that provide continuity to changing phenomena without relying on causal explanations. These relations depend on more basic affinities between apparently unrelated phenomena. As I will defend, such connections contribute to a nonexplanatory form of understanding.

By detailing a long-disputed measurement application from astrobiology, I analyze the relations between theoretical assumptions, research questions, measurement procedures, manufactured effects, and genuine phenomena in measurement robustness (MR). First, I present a problem about measurement standards in MR by examining divergent results, partially reliable procedures, and contradictory measurement standards in the “arsenic life” case study. Second, I show that the resolution to this problem requires examining the central role of divergence upstream in measurement methodology and extending MR beyond results and procedures. I illustrate how discipline-specific assumptions moderate measurement standards and research questions, as well as how those assumptions influence the manipulation of measurement conditions, which can result in artifact production, genuine effect destruction, or the creation of instructive effects.

Ever since John L. Austin, “performativity” has meant many things to many scholars—something that did not change when phrases like “performativity of science” became popular, obscuring various debates in philosophy of science. In this article, I revisit Austin’s original use of the term and propose a nomenclature of three kinds of performativity in science. This revises earlier attempts to identify “authentic” or “genuine” performativity in science and instead advances an integrative disambiguation—one that preserves what scholars in science and technology studies and philosophy of science have found helpful about the concept while addressing demonstrable conceptual conflations.

The goal of this paper is to argue for a systematic set of connections between the philosophy literature on the empirical significance of symmetries and the physics literature on generating effective field theories using symmetry-breaking data via the coset construction. A salient case for exhibiting these connections is the symmetry principles exhibited by Newton’s Corollaries V and VI. In the following, we (i) argue for an interpretation of symmetry breaking in terms of subsystem–environment reasoning, and (ii) display a novel derivation of Newtonian gravitational dynamics from Newton’s Corollaries V and VI using this interpretation.

This article applies the Copernican principle and Gott’s delta-t method to the infamous Humean problem of induction. Copernican reasoning uses only the observed duration of a phenomenon to predict its future duration, so this reasoning is noncircular. It also does not involve positing necessary connections, as do many other attempts to answer Hume’s challenge. Copernican reasoning is controversial, as are many of its applications, but the application to Hume’s problem is one that even most critics of the Copernican principle should accept. If Copernican reasoning is ever applicable, it is applicable here. Copernicus might have an answer for Hume.

AI provides a rich source of examples for theories of agency because it allows relevant features to be relatively freely combined. This paper considers agency in sequence models: models that have been trained to predict the next item in a sequence. Training models on sequences of observations and actions can yield systems that use reinforcement learning in context, as well as ones that merely imitate agents. Investigating systems from this group suggests a requirement that agents must learn about the environment in part from their own interaction with it. It also suggests a distinction between two approaches to understanding agency.

Failures of retraction are common in science. Why do they occur? And what determines whether a retraction is successful? We use data from citation records and Altmetrics to test proposed answers to these questions. LaCroix et al. (2021) employ network models to argue the social spread of information helps explain failures of retraction. One prediction is that widely known results, surprisingly, should be easier to retract, since their retraction is more relevant. Our results support this conclusion. We find highly cited papers show more significant reductions in citation after retraction and garner more attention to their retractions as they occur.

Novel tools have allowed researchers to intervene into circuits at the mesoscale. The results of these interventions are often explained by appeal to functions. How are functions ascribed to circuit parts experimentally? I identify two kinds of function ascription practices in circuit interventions. Analysis of these practices shows us that what makes function ascriptions challenging is a lack of interventive control and insufficient constraints on the class of candidate functions to discriminate in practice. One kind of function ascription practice—subtractive analysis—fares better at addressing these challenges.