What is AI slop? The term has recently become a popular way to deride certain kinds of AI-generated content. But what does it actually pick out? In this paper, I argue that calling something "AI slop" isn't just calling it a bad product of AI. Working through a series of increasingly sophisticated accounts — AI slop as any AI output, as low-quality AI output, as AI output with low social value, as AI output in domains that call for human authorship — I argue that each fails. The failures are instructive: the accounts don't work because calling something "AI slop" is not just a judgment about the quality of the output; it is a political claim about how society should organize the relationship between human and artificial intelligence in creative, professional, and civic life. Indiscriminately or unthinkingly calling things "AI slop" smuggles these contested political judgments into public conversations, which could really matter as we collectively negotiate the roles AI should play in society. As AI becomes more embedded in high-stakes domains — medical diagnosis, scientific research, education — the foreclosure of careful collective deliberation about AI's appropriate role in this way may carry serious costs, including the loss of transformative benefits in precisely the domains where getting AI's role right matters most. So we should be careful about what AI slop is and how we use the term.

Much recent work argues that workers are owed a voice in firm governance, through things like workplace democracy, codetermination, works councils, and worker cooperatives. I argue that the case for mandating this is weaker than its defenders suggest. The defenses fall into two camps. Instrumental defenses hold that voice produces goods like better wages, higher productivity, and more dignified workers. But neither the empirical record nor our best theories of the firm suggest that is true. Intrinsic defenses hold that workers have a right to participate in systems that govern them. But that view faces a distinctly liberal cost, imposing a perfectionist, paternalist, and anti-pluralist picture of how we should relate to firms. I close by granting that more say at work can materially improve workers' situations. But the levers that might secure that mostly operate at levels other than firm governance.

With Garzino Demo

The way we're thinking about generative AI right now is fundamentally individual. We see this not just in how users interact with models but also in how models are built, how they're benchmarked, and how commercial and research strategies using AI are defined. We argue that we should abandon this approach if we're hoping for AI to support groundbreaking innovation and scientific discovery. Drawing on research and formal results in complex systems, organizational behavior, and philosophy of science, we show why we should expect deep intellectual breakthroughs to come from epistemically diverse groups of AI agents working together rather than singular superintelligent agents. Having a diverse team broadens the search for solutions, delays premature consensus, and allows for the pursuit of unconventional approaches. Developing diverse AI teams also addresses AI critics' concerns that current models are constrained by past data and lack the creative insight required for innovation. The upshot, we argue, is that the future of transformative transformer-based AI is fundamentally many, not one.

Read preprint →

With Berger, Grim, Bramson, Holman, and Jung

It is widely thought that deliberative quality improves with the number of participants: the more voices in the room, the better the collective judgment. This “wisdom of the crowds” intuition suggests that representative deliberation — in which a subset of deliberators acts on behalf of the larger group — should be epistemically inferior to full plenary deliberation. We test this using a computational agent-based model in which deliberators exchange evidence for and against a proposition and are evaluated on how accurately their collective beliefs track an objective truth. Varying four conditions — the length of deliberation, the distribution of available evidence, problem difficulty, and agents’ memory capacity — we find that representative deliberation frequently matches or outperforms full deliberation. This advantage does not stem from any superior epistemic ability of the representatives themselves. Rather, it emerges from two structural features of the two-tier process: the selective triage of the strongest available evidence, and the periodic resetting of polarized or entrenched beliefs that a second phase of deliberation enables. Our findings suggest that representative structure can be an epistemic asset rather than a liability — not despite, but because of the constraints it imposes.

The standard view is that gaslighting — a form of deception that makes victims doubt their own perspectives — is invariably wrong. This paper challenges the consensus by arguing that it's sometimes permissible for an employer to gaslight an employee. Drawing on three examples of management relationships, I show that managers can be justified in gaslighting employees to fulfill supervening obligations, such as protecting the firm from significant risks or fostering an employee's professional development. The permissibility of gaslighting in these situations stems from the inherently asymmetric structure of employment relationships, which does not always require the same kind of mutual respect demanded in symmetric relationships like friendships, I argue. This teaches us something directly about managerial ethics. But it also shows us something deeper about the ethics of asymmetric relationships (e.g., teacher-student and parent-child relationships), and it suggests a need for further theorizing about the broader normative structure of business and management.

With Grim, Bramson, Holman, Jung, and Berger

Abstract unavailable

With Grim, Bramson, Holman, Jung, and Berger

Many factors contribute to whether juries reach right verdicts. Here we focus on the role of diversity. Direct empirical studies of the effect of altering factors in jury deliberation are severely limited for conceptual, practical, and ethical reasons. Using an agent-based model to avoid these difficulties, we argue that diversity can play at least four importantly different roles in affecting jury verdicts: equal representation can strengthen epistemic jury success when subgroups have differential access to evidence; participation by a subgroup with access to stronger evidence becomes essential for accurate verdicts; diversity can reduce the redundancy of information on which a jury focuses; and limiting cross-group communication can enhance epistemic outcomes by allowing extended deliberation before consensus.

Read the paper →

With Berger, Bramson, Grim, Holman, and Jung

This chapter illustrates how philosophy and political science can inform one another by providing an overview of philosophical contributions the authors have made on the topic of political polarization. The authors outline three contributions they have made to understanding political polarization, particularly of the epistemic kind, discussing work that gives clearer terminology for and ways of measuring polarization, precise mechanistic accounts of polarization, and a novel normative view about a possible source of polarization that casts polarization as a possible outcome of rational, but limited, agents interacting. This last contribution illustrates, contra recent work, how dynamics akin to epistemic bubbles and echo chambers can develop without associated epistemic vices. Taken together, these projects can serve as a guide for producing philosophical work which both contributes to a mainstream disciplinary literature as well as informs cross-disciplinary, empirical literatures in the social sciences.

Read the paper →

With Grim, Bramson, Holman, Jung, and Berger

Epistemic network models are tools for understanding how information and beliefs move in groups of connected epistemic agents. While these tools have been used in philosophy mostly by philosophers of science and social epistemologists, this chapter discusses how epistemic network models can be used to investigate political polarization. The goal of this chapter is to give the reader a good idea of how epistemic network models can be used to approach complex questions in political epistemology about groups of agents sharing information and beliefs. The chapter starts by introducing the idea of an epistemic network model and the general approach to understanding complex epistemic phenomena in which these models play a role. We then discuss a series of epistemic network models of political polarization including ones from Hegselmann and Krause, Singer et al., and Weatherall and O’Connor. We conclude with brief thoughts about how epistemic network models can be used to answer other questions in political epistemology.

Read the paper →

With Grim, Bramson, Holman, Jung, Kovaka, and Berger

We motivate a picture of social epistemology that sees forgetting as subject to epistemic evaluation. Using computer simulations of a simple agent-based model, we show that how agents forget can have as large an impact on group epistemic outcomes as how they share information. But, how we forget, unlike how we form beliefs, isn’t typically taken to be the sort of thing that can be epistemically rational or justified. We consider what we take to be the most promising argument for this claim and find it lacking. We conclude that understanding how agents forget should be as central to social epistemology as understanding how agents form beliefs and share information with others.

Read the paper →

Abstract unavailable

Read the review →

With Sara Aronowitz

The standard view says that epistemic normativity is normativity of belief. If you’re an evidentialist, for example, you’ll think that all epistemic reasons are reasons to believe what your evidence supports. Here we present a line of argument that pushes back against this standard view. If the argument is right, there are epistemic reasons for things other than belief. The argument starts with evidentialist commitments and proceeds by a series of cases, each containing a reason. As the cases progress, the reasons change from counting in favor of things like having a belief to things like performing ordinary actions. We argue that each of those reasons is epistemic. If the argument succeeds, we should think there are epistemic reasons to consider hypotheses, conduct thought and physical experiments, extend one’s evidence, and perform mundane tasks like eating a sandwich, just as there are epistemic reasons to believe what one’s evidence supports.

Download preprint →

With Patrick Grim

Computational philosophy is the use of mechanized computational techniques to instantiate, extend, and amplify philosophical research. Computational philosophy is not philosophy of computers or computational techniques; it is rather philosophy using computers and computational techniques. The idea is simply to apply advances in computer technology and techniques to advance discovery, exploration and argument within any philosophical area.

After touching on historical precursors, this article discusses contemporary computational philosophy across a variety of fields: epistemology, metaphysics, philosophy of science, ethics and social philosophy, philosophy of language and philosophy of mind, often with examples of operating software. Far short of any attempt at an exhaustive treatment, the intention is to introduce the spirit of each application by using some representative examples.

Read the entry →

With Grim, Bramson, Berger, Jung, and Page

Epistemic justifications for democracy have been offered in terms of two different forms of information aggregation and decision-making. The Condorcet Jury Theorem is appealed to as a justification in terms of votes, and the Hong–Page ‘diversity trumps ability’ result is appealed to as a justification in terms of deliberation in the form of collaborative search. Both results, however, are models of full and direct participation across a population. In this paper, we contrast how these results hold up within the familiar structure of a representative hierarchy. We first consider extant analytic work that shows that representation inevitably weakens the voting results of the Condorcet Jury Theorem. We then go on to show that collaborative search, as modeled by Hong and Page, holds its own within hierarchical representation. In a variation on the dynamics of group search, representation even shows a slight edge over direct participation. This contrast illustrates how models of information aggregation vary when put into a representative structure. While some of the epistemic merits of democracy are lost when voting is done hierarchically, modeling results show that representation can preserve and even slightly amplify the epistemic virtues of collaborative search.

Read the paper →

Recent rejections of epistemic consequentialism—from Firth, Jenkins, Berker, and Greaves—have argued that consequentialism is committed to objectionable trade-offs between epistemic values. This paper argues that these objections fail to undermine epistemic consequentialism properly understood. By distinguishing the consequentialist account of what we should believe from accounts of rational and justified belief, and by adopting a sophisticated structure parallel to Railton's act consequentialism in ethics, epistemic consequentialism can deflect the objections while retaining its core commitments.

Read the paper →

With Jung, Grim, Bramson, Berger, Holman, and Kovaka

This article aims to describe the last 10 years of the collaborative scientific endeavors on polarization in particular and collective problem-solving in general by our multidisciplinary research team. We describe the team's disciplinary composition-social psychology, political science, social philosophy/epistemology, and complex systems science-highlighting the shared and unique skill sets of our group members and how each discipline contributes to studying polarization and collective problem-solving. With an eye to the literature on team dynamics, we describe team logistics and processes that we believe make our multidisciplinary team persistent and productive. We emphasize challenges and difficulties caused by disciplinary differences in terms of terminology, units/levels of analysis, methodology, and theoretical assumptions. We then explain how work disambiguating the concepts of polarization and developing an integrative theoretical and methodological framework with complex systems perspectives has helped us overcome these challenges. We summarize the major findings that our research has produced over the past decade, and describe our current research and future directions. Last, we discuss lessons we have learned, including difficulties in a "three models" project and how we addressed them, with suggestions for effective multidisciplinary team research.

Read the paper →

A number of formal models, including a highly influential model from Hong and Page, purport to show that functionally diverse groups often beat groups of individually high-performing agents in solving problems. Thompson argues that in Hong and Page’s model, that the diverse groups are created by a random process explains their success, not the diversity. Here, I defend the diversity interpretation of the Hong and Page result. The failure of Thompson’s argument shows that to understand the value of functional diversity, we should be clearer about how we conceive of and measure that diversity.

Read the paper →

With Bramson, Grim, Holman, Jung, Kovaka, Ranginani, and Berger

Public discussions of political and social issues are often characterized by deep and persistent polarization. In social psychology, it’s standard to treat belief polarization as the product of epistemic irrationality. In contrast, we argue that the persistent disagreement that grounds political and social polarization can be produced by epistemically rational agents, when those agents have limited cognitive resources. Using an agent-based model of group deliberation, we show that groups of deliberating agents using coherence-based strategies for managing their limited resources tend to polarize into different subgroups. We argue that using that strategy is epistemically rational for limited agents. So even though group polarization looks like it must be the product of human irrationality, polarization can be the result of fully rational deliberation with natural human limitations.

Read the paper →

With Grim, Bramson, Holman, McGeehan, and Berger

The Hong and Page ‘diversity trumps ability’ result has been used to argue for the more general claim that a diverse set of agents is epistemically superior to a comparable group of experts. Here we extend Hong and Page’s model to landscapes of different degrees of randomness and demonstrate the sensitivity of the ‘diversity trumps ability’ result. This analysis offers a more nuanced picture of how diversity, ability, and expertise may relate. Although models of this sort can indeed be suggestive for diversity policies, we advise against interpreting such results overly broadly.

Read the paper →

Recent rejections of epistemic consequentialism, like those from Firth, Jenkins, Berker, and Greaves, have argued that consequentialism is committed to objectionable trade-offs and suggest that consequentialism's propensity for trade-offs hints at a larger problem. Here I argue that sanctioning trade-offs isn't a fault of a theory of epistemic normativity, because there are permissible epistemic trade-offs. I give examples of permissible epistemic trade-offs in pedagogy, in changes of worldview, and in indirect epistemic decisions. I also show that views that sanction trade-offs have an easier time than their rivals in explaining both why we ought to be open-minded and how arguments with suppositions get their argumentative force. These considerations don't eliminate the consequentialist's burden to respond to the objectionable cases, but they do undermine the idea that no correct theory of epistemic normativity properly sanctions trade-offs.

Read the paper → Download PDF →

With Holman, Berger, Grim, and Bramson

Agent-based models have played a prominent role in recent debates about the merits of democracy. In particular, the formal model of Lu Hong and Scott Page and the associated “diversity trumps ability” result has typically been seen to support the epistemic virtues of democracy over epistocracy (i.e., governance by experts). In this paper we first identify the modeling choices embodied in the original formal model and then critique the application of the Hong-Page results to philosophical debates on the relative merits of democracy. In particular we argue that the “best-performing agents” in the Hong-Page model should not be interpreted as experts. We next explore a closely related model in which best-performing agents are more plausibly seen as experts and show that the diversity trumps ability result fails to hold. However, with changes in other parameters (such as the deliberation dynamic) the diversity trumps ability result is restored. The sensitivity of this result to parameter choices illustrates the complexity of the link between formal modeling and more general philosophical claims; we use this debate as a platform for a more general discussion of when and how agent-based models can contribute to philosophical discussions.

Read the paper →

Epistemic consequentialists think that epistemic norms are about believing the truth and avoiding error. Recently, a number of authors have rejected epistemic consequentialism on the basis that it incorrectly sanctions tradeoffs of epistemic goodness. Here, I argue that epistemic consequentialists should borrow two lessons from ethical consequentialists to respond to these worries. Epistemic consequentialists should construe their view as an account of right belief, which they distinguish from other notions like rational and justified belief. Epistemic consequentialists should also make their view ‘sophisticated,’ in the sense of Railton. Epistemic consequentialism, I conclude, is best construed as sharing much of its structure with prominent act-consequentialist views in ethics. Epistemic consequentialism has an advantage over its ethical counterpart though: the key claim of the view is practically universally accepted, which gives us an additional reason to think it’s true.

Read the paper →

With Grim, Bramson, Berger, Sack, Fisher, and Flocken

Polarization is a topic of intense interest among social scientists, but there is significant disagreement regarding the character of the phenomenon and little understanding of underlying mechanics. A first problem, we argue, is that polarization appears in the literature as not one concept but many. In the first part of the article, we distinguish nine phenomena that may be considered polarization, with suggestions of appropriate measures for each. In the second part of the article, we apply this analysis to evaluate the types of polarization generated by the three major families of computational models proposing specific mechanisms of opinion polarization.

Read the paper →

With Bramson, Grim, Fisher, Sack, Berger, and Flocken

This article distinguishes nine senses of polarization and provides formal measures for each one to refine the methodology used to describe polarization in distributions of attitudes. Each distinct concept is explained through a definition, formal measures, examples, and references. We then apply these measures to GSS data regarding political views, opinions on abortion, and religiosity—topics described as revealing social polarization. Previous breakdowns of polarization include domain-specific assumptions and focus on a subset of the distribution’s features. This has conflated multiple, independent features of attitude distributions. The current work aims to extract the distinct senses of polarization and demonstrate that by becoming clearer on these distinctions we can better focus our efforts on substantive issues in social phenomena.

Read the paper →

The is-ought gap is Hume’s claim that we can’t get an ‘ought’ from just ‘is’s. Prior (“The Autonomy of Ethics,” 1960) showed that its most straightforward formulation, a staple of introductory philosophy classes, fails. Many authors attempt to resurrect the claim by restricting its domain syntactically or by reformulating it in terms of models of deontic logic. Those attempts prove to be complex, incomplete, or incorrect. I provide a simple reformulation of the is-ought gap that closely fits Hume’s description of it. My formulation of the gap avoids the proposed counterexamples from Prior and offers a natural explanation of why they seem compelling. Moreover, I show that my formulation of the gap is guaranteed by standard theories of the semantics of normative terms, and that provides a more general reason to accept it.

Read the paper →

With Grim, Fisher, and Reade

Understanding the dynamics of information is crucial to many areas of research, both inside and outside of philosophy. Using computer simulations of three kinds of information, germs, genes, and memes, we show that the mechanism of information transfer often swamps network structure in terms of its effects on both the dynamics and the fitness of the information. This insight has both obvious and subtle implications for a number of questions in philosophy, including questions about the nature of information, whether there is genetic information, and how to arrange scientific communities.

Read the paper →

The traditional solutions to the Sleeping Beauty problem say that Beauty should have either a sharp 1/3 or sharp 1/2 credence that the coin flip was heads when she wakes. But Beauty's evidence is incomplete so that it doesn't warrant a precise credence. Instead, Beauty ought to have a properly imprecise credence when she wakes. In particular, her representor ought to assign R(Heads) = [0, 1/2]. The solution can account for many of the intuitions that motivate the traditional solutions. The paper also offers a new objection to Elga's restricted version of the principle of indifference, which an opponent may try to use to collapse the imprecision.

Read the paper →

With Grim, Fisher, Bramson, Berger, Reade, Flocken, and Sales

A scientific community can be modeled as a collection of epistemic agents attempting to answer questions, in part by communicating about their hypotheses and results. We can treat the pathways of scientific communication as a network. When we do, it becomes clear that the interaction between the structure of the network and the nature of the question under investigation affects epistemic desiderata, including accuracy and speed to community consensus. Here we build on previous work, both our own and others’, in order to get a firmer grasp on precisely which features of scientific communities interact with which features of scientific questions in order to influence epistemic outcomes.

Here we introduce a measure on the landscape meant to capture some aspects of the difficulty of answering an empirical question. We then investigate both how different communication networks affect whether the community finds the best answer and the time it takes for the community to reach consensus on an answer. We measure these two epistemic desiderata on a continuum of networks sampled from the Watts–Strogatz spectrum. It turns out that finding the best answer and reaching consensus exhibit radically different patterns. The time it takes for a community to reach a consensus in these models roughly tracks mean path length in the network. Whether a scientific community finds the best answer, on the other hand, tracks neither mean path length nor clustering coefficient.

Read the paper →

With Grim, Bramson, Fisher, Flocken, and Berger

Abstract unavailable

Updated and reprinted in Youngman & Hadzikadic (eds.), Complexity and the Human Experience, Pan Stanford Press, 2014.

Read the paper →

With Grim, Thomas, Reade, Fisher, Garza, Fryer, and Chatman

Public health care interventions—regarding vaccination, obesity, and HIV, for example—standardly take the form of information dissemination across a community. But information networks can vary importantly between different ethnic communities, as can levels of trust in information from different sources. We use data from the Greater Pittsburgh Random Household Health Survey to construct models of information networks for White and Black communities - models which reflect the degree of information contact between individuals, with degrees of trust in information from various sources correlated with positions in that social network. With simple assumptions regarding belief change and social reinforcement, we use those modeled networks to build dynamic agent-based models of how information can be expected to flow and how beliefs can be expected to change across each community. With contrasting information from governmental and religious sources, the results show importantly different dynamic patterns of belief polarization within the two communities.

Read the paper →

With Grim, Reade, and Fisher

Beyond belief change and meme adoption, both genetics and infection have been spoken of in terms of information transfer. What we examine here, concentrating on the specific case of transfer between sub-networks, are the differences in network dynamics in these cases: the different network dynamics of germs, genes, and memes.

Germs and memes, it turns out, exhibit a very different dynamics across networks. For infection, measured in terms of time to total infection, it is network type rather than degree of linkage between sub-networks that is of primary importance. For belief transfer, measured in terms of time to consensus, it is degree of linkage rather than network type that is crucial.

Genes model each of these other dynamics in part, but match neither in full. For genetics, like belief transfer and unlike infection, network type makes little difference. Like infection and unlike belief, on the other hand, the dynamics of genetic information transfer within single and between linked networks are much the same. In ways both surprising and intriguing, transfer of genetic information seems to be robust across network differences crucial for the other two.

Download PDF →

With Grim, Reade, Fisher, and Majewicz

In order to understand the transmission of a disease across a population we will have to understand not only the dynamics of contact infection but the transfer of health-care beliefs and resulting health-care behaviors across that population. This paper is a first step in that direction, focusing on the contrasting role of linkage or isolation between sub-networks in (a) contact infection and (b) belief transfer. Using both analytical tools and agent-based simulations we show that it is the structure of a network that is primary for predicting contact infection—whether the networks or sub-networks at issue are distributed ring networks or total networks (hubs, wheels, small world, random, or scale-free for example). Measured in terms of time to total infection, degree of linkage between sub-networks plays a minor role. The case of belief is importantly different. Using a simplified model of belief reinforcement, and measuring belief transfer in terms of time to community consensus, we show that degree of linkage between sub-networks plays a major role in social communication of beliefs. Here, in contrast to the case of contract infection, network type turns out to be of relatively minor importance. What you believe travels differently. In a final section we show that the pattern of belief transfer exhibits a classic power law regardless of the type of network involved.

Read the paper →

With Grim, Reade, Fisher, and Majewicz

Abstract unavailable

Read the paper →