Alexander Tolbert

This essay introduces the special issue Contemporary Issues in Black Philosophy: Pluralism in Methodological Approaches and advances a metaphilosophical argument about method in Black philosophy. We distinguish the question of what makes philosophy Black from the question of what counts as philosophy, and argue that conflating these questions produces a misleading methodological monism. Attention to the difference between substance and method shows that methodological choice must be guided by the problems under investigation. We defend a disciplined pluralism according to which methods are justified by problem-fit rather than by allegiance to a single tradition. The essays in this collection are a demonstration, not a discussion, of methodological pluralism and its benefits to philosophy.

The biases exhibited by text-to-image (TTI) models are often treated as independent, though in reality, they may be deeply interrelated. Addressing bias along one dimension—such as ethnicity or age—can inadvertently affect another, like gender, either mitigating or exacerbating existing disparities. Understanding these interdependencies is crucial for designing fairer generative models, yet measuring such effects quantitatively remains a challenge. To address this, we introduce BiasConnect, a novel tool for analyzing and quantifying bias interactions in TTI models. BiasConnect uses counterfactual interventions along different bias axes to reveal the underlying structure of these interactions and estimates the effect of mitigating one bias axis on another. These estimates show strong correlation (+0.65) with observed postmitigation outcomes. Building on BiasConnect, we propose InterMit, an intersectional bias mitigation algorithm guided by a user-defined target distribution and priority weights. InterMit achieves lower bias (0.33 vs. 0.52) with fewer mitigation steps (2.38 vs. 3.15 average steps), and yields superior image quality compared to traditional techniques. Although our implementation is training-free, InterMit is modular and can be integrated with many existing debiasing approaches for TTI models, making it a flexible and extensible solution.

Variable selection poses a significant challenge in causal modeling, particularly within the social sciences, where constructs often rely on interrelated factors such as age, socioeconomic status, gender, and race. Indeed, it has been argued that such attributes must be modeled as macro-level abstractions of lower-level manipulable features, in order to preserve the modularity assumption essential to causal inference (Mossé ́e et al., 2025). This paper accordingly extends the theoretical framework of Causal Feature Learning (CFL). Empirically, we apply the CFL algorithm to diverse social science datasets, evaluating how CFL-derived macrostates compare with traditional microstates in downstream modeling tasks.

Disparities in lending to minority applicants persist even as algorithmic lending practices proliferate. Further, disparities in interest rates charged can remain large even when loan applicants from different groups are equally creditworthy. We study the role of risk-management constraints, specifically Value-at-Risk (VaR) constraints, in the persistence of segregation in loan approval decisions. We develop a formal model in which a mainstream (low-interest) bank is more sensitive to variance risk than a subprime (high-interest) bank. If the mainstream bank has an inflated prior belief about the variance of the minority group, it may deny that group credit indefinitely, thus never learning the true risk of lending to that group, while the subprime lender serves this population at higher rates. We formalize this as a “subprime trap” equilibrium. Finally, we show that a small, finite subsidy (or partial guarantee) can help minority groups escape the trap by covering enough of the mainstream bank’s downside so that it can afford to lend and learn the minority group’s true risk. Once it has sufficiently many data points, it meets its VaR requirement with no further assistance, minority groups are approved for loans by the mainstream bank, and competition drives down the interest rates of subprime lenders.

Ambition and risk-taking have been heralded as important ways for marginalized communities to get out of cycles of poverty. As a result, educational messaging often encourages individuals to strengthen their personal resolve and develop characteristics such as discipline and grit to succeed in ambitious ends. However, recent work in philosophy and sociology highlights that this messaging often does more harm than good for students in these situations. We study similar questions using a different epistemic approach and in simple theoretical models – we provide a quantitative model of decision-making between stable and risky choices in the improving multi-armed bandits framework. We use this model to first study how individuals’ “strategies” are affected by their level of grittiness and how this affects their accrued rewards. Then, we study the impact of various interventions, such as increasing grit or providing a financial safety net. Our investigation of rational decision making involves two different formal models of rationality, the competitive ratio between the accrued reward and the optimal reward and Bayesian quantification of uncertainty.

In this paper, we relate the philosophical literature on pessimism traps to information cascades, a formal model derived from the economics and mathematics literature. A pessimism trap is a social pattern in which individuals in a community, in situations of uncertainty, copy the sub-optimal actions of others, despite their individual beliefs. This maps nicely onto the concept of an information cascade, which involves a sequence of agents making a decision between two alternatives, with a private signal of the superior alternative and a public history of others’ actions. Key results from the economics literature show that information cascades occur with probability one in many contexts, and depending on the strength of the signal, populations can fall into the incorrect cascade very easily and quickly. Once formed, in the absence of external perturbation, a cascade cannot be broken – therefore, we derive an intervention that can be used to nudge a population from an incorrect to a correct cascade and, importantly, maintain the cascade once the subsidy is discontinued. We extend this to the case of multiple communities, each of which might have a different optimal action, and a government providing subsidies that cannot discriminate between communities and does not know which action is optimal for each. We study this both theoretically and empirically.

Many machine learning applications predict individual probabilities, such as the likelihood that a person develops a particular illness. Since these probabilities are unknown, a key question is how to address situations in which different models trained on the same dataset produce varying predictions for certain individuals. This issue is exemplified by the model multiplicity (MM) phenomenon, where a set of comparable models yield inconsistent predictions. Roth, Tolbert, and Weinstein recently introduced a reconciliation procedure, the Reconcile algorithm, to address this problem. Given two disagreeing models, the algorithm leverages their disagreement to falsify and improve at least one of the models. In this paper, we empirically analyze the Reconcile algorithm using five widely-used fairness datasets: COMPAS, Communities and Crime, Adult, Statlog (German Credit Data), and the ACS Dataset. We examine how Reconcile fits within the model multiplicity literature and compare it to existing MM solutions, demonstrating its effectiveness. We also discuss potential improvements to the Reconcile algorithm theoretically and practically. Finally, we extend the Reconcile algorithm to the setting of causal inference, given that different competing estimators can again disagree on specific causal average treatment effect (CATE) values. We present the first extension of the Reconcile algorithm in causal inference, analyze its theoretical properties, and conduct empirical tests. Our results confirm the practical effectiveness of Reconcile and its applicability across various domains.

Pejorative or harmful stereotypical language appears in large language model (LLM) outputs, and despite various mitigation approaches — such as reinforcement learning from human feedback (RLHF) or manual fixes after red teaming — these issues persist. This paper argues that generics, which are generalizations lacking explicit quantification (e.g., statements like “mosquitoes carry malaria” or “birds lay eggs”), contribute significantly to these harmful stereotypes. Whereas humans naturally contextualize and quantify such generalizations (e.g., discerning that only female birds lay eggs), LLMs may misinterpret these vague expressions. Given the frequency of generics in natural language, if LLMs cannot correctly interpret or quantify them, there exists a potential for persistent, harmful stereotypical outputs. This work employs a philosophy of language approach to show that deficiencies in processing generics can yield problematic assumptions and to demonstrate that current LLMs lack robust mechanisms to handle these ambiguities.

In this chapter, I argue that the algorithmic fairness literature systematically under-theorizes race by treating bias as a modular technical problem. Drawing on Charles Mills's The Racial Contract and Black philosophical traditions, I demonstrate why incremental technical reforms often fail and may even reinforce patterns of racial stratification. I introduce the concept of the Racial Algorithm: The process through which causal mechanisms structured by the Racial Contract generate data distributions, algorithms learn from those distributions, and public or private actors make decisions that, in turn, reaffirm and lock in the structures of the Racial Contract. With this analysis in mind, I then discuss the tension between reform and abolition, concluding with a non-ideal framework of procedural and substantive criteria to determine when algorithmic governance is legitimate and when it must be abolished.

Individual probabilities refer to the probabilities of outcomes that are realized only once: the probability that it will rain tomorrow, the probability that Alice will die within the next 12 months, the probability that Bob will be arrested for a violent crime in the next 18 months, etc. Individual probabilities are fundamentally unknowable. Nevertheless, we show that two parties who agree on the data—or on how to sample from a data distribution—cannot agree to disagree on how to model individual probabilities. This is because any two models of individual probabilities that substantially disagree can together be used to empirically falsify and improveat least one of the two models. This can be efficiently iterated in a process of “reconciliation” that results in models that both parties agree are superior to the models they started with, and which themselves (almost) agree on the forecasts of individual probabilities (almost) everywhere. We conclude that although individual probabilities are unknowable, they are contestable via a computationally and data efficient process that must lead to agreement. Thus we cannot find ourselves in a situation in which we have two equally accurate and unimprovable models that disagree substantially in their predictions—providing an answer to what is sometimes called the predictive or model multiplicity problem.

I consider the problem of learning from data corrupted by underrepresentation bias, where positive examples are filtered out at different, unknown rates for a fixed number of sensitive groups. I show that with a small amount of unbiased data, I can efficiently estimate the group-wise drop-out rates, even in settings where intersectional group membership makes learning each intersectional rate computationally infeasible. Using these estimates, I construct a reweighting scheme that allows me to approximate the loss of any hypothesis on the true distribution and present an algorithm encapsulating this process. Finally, I define a bespoke notion of PAC learnability for the underrepresentation and intersectional bias setting and show that my algorithm allows efficient learning for model classes of finite VC dimension.

This paper proposes a novel view in the the philosophy of race & causation literature known as “causal agnosticism” about race. Causal agnosticism about race implies that it is reasonable to refrain from making judgments about whether race is a cause. The paper’s thesis asserts that certain conditions must be met to infer that something is a cause, according to the fundamental assumptions of causal inference. However, in the case of race, these conditions are often violated. By advocating for causal agnosticism, the paper suggests a more modest approach to understanding the role of race in causal relationships.

Recent debates in political philosophy of science underscore the need for normative frameworks to determine which values should guide scientific inquiry (Douglas 2009). Philosophers increasingly draw on principles from political theory, such as democratic majoritarianism (Schroeder 2021), deliberative proceduralism (Lusk 2021), and Rawlsian public reason (Rawls 1999; Badano 2025), to inform scientific investigations. I argue these approaches fail in societies where basic institutions developed through historical racial domination—including chattel slavery, Jim Crow segregation, COINTELPRO’s targeting of Black movements, racially exclusionary housing policies, and mass incarceration without reparative correction. Drawing on Mills’s (2017) critique of ideal theory, I contend that these frameworks abstract away from the racialized structures that organize contemporary scientific and technological practices, and that the normative constraints they propose are insufficient to secure legitimacy under such conditions. Using Artificial Intelligence as a case study, I introduce the concept of the Racial Algorithm to show how machine learning systems encode and amplify racial inequities through algorithmic decision-making. I conclude by sketching a framework for scientific legitimacy grounded in non-ideal theory.

This paper argues that in settings where you have intense social stratification in particular, those along racial lines, you will in turn have positivity violations which impede causal inference. Racial stratification often causes large differences in the distribution of various outcomes across different racial groups such that the covariate distribution for those groups do not overlap significantly. This lack of overlap constitutes a positivity violation. Because positivity is a necessary condition for causal inference, any failure to repair social stratification that constitues a positivity violation undermines causal inference. As a result, I claim that if we have the epistemic goal of engaging in causal inquiry, we must remove barriers that impede the acquisition of causal knowledge. This means that repairing social stratification is required to have causal knowledge of the social world going forward.

We draw a distinction between the traditional reference class problem, which describes an obstruction to estimating a single individual probability—which we rename the individual reference class problem—and what we call the reference class problem at scale, which can result when using tools from statistics and machine learning to systematically make predictions about many individual probabilities simultaneously. We argue that scale actually helps to mitigate the reference class problem, and purely statistical tools can be used to efficiently minimize the reference class problem at scale, even though they cannot be used to solve the individual reference class problem.

This paper argues that causal inference is a necessary condition for achieving fairness in algorithmic decision-making. Dominant machine learning models are typically limited to associative methods. However, we often need to modify the very probability distributions that produce social injustice, not merely identify predictive patterns from them, an undertaking standard machine learning neglects. Fairness often requires identifying who or what is responsible for a particular outcome of interest, uncovering the source of disparities, and determining which policies are effective interventions to address these issues. Normative questions such as these require causal modeling. I then confront a key objection: certain social variables, notably race, appear to violate modularity in causal models. To address this, I propose several strategies, including bracketing subsystems into coarser causal abstractions/macro-variables, refining the study design, and relaxing local invariances to handle these non-local influences. In sum, I claim that causal modeling is indispensable for responsible algorithmic decision-making.

This paper challenges the view that race is a reliable scientific variable or kind for the purpose of inductive inference within the social sciences. I characterize stability in terms of Extended Conditional Independence (ECI) and show that the heterogeneity and instability of racial categories across different background circumstances undermines their ability to support robust inductive inference and explanatory power. I claim this, in turn, undermines racial categories' status as real scientific variables or kinds. Race, has local stability within restricted sets of target systems and thus, its reality is limited to those domains. I argue for a restricted form of racial realism, a view I call Restricted Racial Realism (RRR).