David Freeborn

I propose a model of systematic understanding, suitable for machine learning systems. On this account, an agent understands a property of a target system when it contains an adequate internal model that tracks real regularities, is coupled to the target by stable bridge principles, and supports reliable prediction. I argue that contemporary deep learning systems often can and do achieve such understanding. However they generally fall short of the ideal of scientific understanding: the understanding is symbolically misaligned with the target system, not explicitly reductive, and only weakly unifying. I label this the Fractured Understanding Hypothesis.

This paper explores what computational methodologies can tell us about philosophical education, particularly in the context of artificial intelligence (AI) ethics. Taking the readings on our AI ethics and responsible AI syllabi as a corpus of AI ethics literature, we conduct an analysis of the content of these courses through a variety of methods: word frequency analysis, term frequency–inverse document frequency (TF–IDF) scoring, document vectorization via SciBERT, clustering via k-means, and topic modelling using latent Dirichlet allocation (LDA). We reflect on the findings of these analyses, and more broadly on what computational approaches can offer to the practice of philosophical education. Finally, we compare our approach to previous computational approaches in philosophy, and more broadly in the digital humanities. This project offers a proof of concept for how contemporary natural-language processing techniques can be used to support philosophical pedagogy: not only to reflect critically on what we teach, but to discover new materials, explore conceptual gaps, and make our courses more accessible to students from a range of disciplinary backgrounds.

There are myriad techniques industry actors use to shape the public understanding of science. While a naive view might assume these techniques typically involve fraud or outright deception, the truth is more nuanced. This paper analyzes industrial distraction, a common technique where industry actors fund and share research that is accurate, often high-quality, but nonetheless misleading on important matters of fact. This involves reshaping causal understanding of phenomena with distracting information. Using case studies and causal models, we illustrate how this impacts belief and decision making even for rational learners, informing science policy and debates about misleading content.

Receivers in standard signaling game models struggle with learning compositional information. Even when the signalers send compositional messages, the receivers do not interpret them compositionally. When information from one message component is lost or forgotten, the information from other components is also erased. In this paper I construct signaling game models in which genuine compositional understanding evolves. I present two new models: a minimalist receiver who only learns from the atomic messages of a signal, and a generalist receiver who learns from all of the available information. These models are in many ways simpler than previous alternatives, and allow the receivers to learn from the atomic components of messages.

Manifold learning and effective model building are generally viewed as fundamentally different types of procedure. After all, in one we build a simplified model of the data, in the other, we construct a simplified model of the another model. Nonetheless, I argue that certain kinds of high-dimensional effective model building, and effective field theory construction in quantum field theory, can be viewed as special cases of manifold learning. I argue that this helps to shed light on all of these techniques. First, it suggests that the effective model building procedure depends upon a certain kind of algorithmic compressibility requirement. All three approaches assume that real-world systems exhibit certain redundancies, due to regularities. The use of these regularities to build simplified models is essential for scientific progress in many different domains.

General epistemic polarization arises when the beliefs of a population grow further apart, in particular when all agents update on the same evidence. Epistemic factionalization arises when the beliefs grow further apart, but different beliefs also become correlated across the population. I present a model of how factionalization can emerge in a population of ideally rational agents. This kind of factionalization is driven by probabilistic relations between beliefs, with background beliefs shaping how the agents’ beliefs evolve in the light of new evidence. Moreover, I show that in such a model, the only possible outcomes from updating on identical evidence are general convergence or factionalization. Beliefs cannot spread out in all directions: if the beliefs overall polarize, then it must result in factionalization.

Haag’s theorem cries out for explanation and critical assessment: It sounds the alarm that something is (perhaps) not right in one of the standard ways of constructing interacting fields to be used in generating predictions for scattering experiments. Viewpoints as to the precise nature of the problem, the appropriate solution, and subsequently-called-for developments in areas of physics, mathematics, and philosophy differ widely. In this paper, we develop and deploy a conceptual framework for critically assessing these disparate responses to Haag’s theorem. Doing so reveals the driving force of more general questions as to the nature and purpose of foundational work in physics.

Bargaining games have played a prominent role in modeling the evolution of social conventions. Previous models assumed that agents must choose from a predetermined set of strategies. I present a new model of two agents learning in bargaining games in which new strategies must be invented and reinforced. I study the efficiency and fairness of the model outcomes. The outcomes are somewhat efficient, but a significant part of the resource is wasted nonetheless. I implement two forms of forgetting, and restrictions to the set of strategies that can be invented.

Recent work on the evolution of social contracts and conventions has often used models of bargaining games, with reinforcement learning. A recent innovation is the requirement that every strategy must be invented either through through learning or reinforcement. However, agents frequently get stuck in highly-reinforced “traps” that prevent them from arriving at outcomes that are efficient or fair to the both players. Agents face a trade-off between exploration and exploitation, i.e. between continuing to invent new strategies and reinforcing strategies that have already become highly reinforced by yielding high rewards. In this paper I systematically study the relationship between rates of invention and the efficiency and fairness of outcomes in two-player, repeated bargaining games. I use a basic reinforcement learning model with invention, and five variations of this model, designed introduce various forms of forgetting, to prioritize more recent reinforcement, or to maintain a higher rate of invention. I use computer simulations to investigate the outcomes of each model. Each models shows qualitative similarities in the relationship between the efficiency and fairness of outcomes, and the relative amount of exploration or exploitation that takes place. Surprisingly, there are often trade-offs between the efficiency and the fairness of the outcomes.

The “sloppy models” program originated in systems biology, but has seen applications across a range of fields. Sloppy models are dependent on a large number of parameters, but highly insensitive to the vast majority of parameter combinations. Sloppy models proponents claim that the program may explain the success of science. I argue that the sloppy models program can at best provide a very partial explanation. Drawing a parallel with renormalization group realism, I argue that it would only give us grounds for a minimal kind of scientific realism. Nonetheless, the program can offer certain epistemic virtues.