Unsupervised learning algorithms are widely used for many important statistical tasks with numerous applications in science and industry. Yet despite their prevalence, they have attracted remarkably little philosophical scrutiny to date. This stands in stark contrast to supervised and reinforcement learning algorithms, which have been widely studied and critically evaluated, often with an emphasis on ethical concerns. In this article, I analyze three canonical unsupervised learning problems: clustering, abstraction, and generative modeling. I argue that these methods raise unique epistemological and (...) ontological questions, providing data-driven tools for discovering natural kinds and distinguishing essence from contingency. This analysis goes some way toward filling the lacuna in contemporary philosophical discourse on unsupervised learning, as well as bringing conceptual unity to a heterogeneous field more often described by what it isnot(i.e., supervised or reinforcement learning) than by what itis. I submit that unsupervised learning is not just a legitimate subject of philosophical inquiry but perhaps the most fundamental branch of all AI. However, an uncritical overreliance on unsupervised methods poses major epistemic and ethical risks. I conclude by advocating for a pragmatic, error-statistical approach that embraces the opportunities and mitigates the challenges posed by this powerful class of algorithms. (shrink)

Contemporary societies are often “polarized”, in the sense that sub-groups within these societies hold stably opposing beliefs, even when there is a fact of the matter. Extant models of polarization do not capture the idea that some beliefs are true and others false. Here we present a model, based on the network epistemology framework of Bala and Goyal, 784–811 1998), in which polarization emerges even though agents gather evidence about their beliefs, and true belief yields a pay-off advantage. As we (...) discuss, these results are especially relevant to polarization in scientific communities, for these reasons. The key mechanism that generates polarization involves treating evidence generated by other agents as uncertain when their beliefs are relatively different from one’s own. (shrink)

I consider how complex logical operations might self-assemble in a signalling-game context via composition of simpler underlying dispositions. On the one hand, agents may take advantage of pre-evolved dispositions; on the other hand, they may co-evolve dispositions as they simultaneously learn to combine them to display more complex behaviour. In either case, the evolution of complex logical operations can be more efficient than evolving such capacities from scratch. Showing how complex phenomena like these might evolve provides an additional path to (...) the possibility of evolving more or less rich notions of compositionality. This helps provide another facet of the evolutionary story of how sufficiently rich, human-level cognitive or linguistic capacities may arise from simpler precursors. 1Signalling and Self-assembly2Simple Unary Logic Games3Composing Unary Functions for Binary Inputs 3.1Utilizing pre-evolved dispositions3.2Co-evolving logical dispositions3.3Learning appropriate outputs3.4Taking account of the full state-space of unary games3.5Role-free composition4Discussion 4.1Efficacy and efficiency of learning complex dispositions4.2Other binary operations4.3To infinity and beyond5Conclusion. (shrink)

In this paper, we explore interaction history as a particular source of pressure for achieving emergent compositional communication in multi-agent systems. We propose a training regime implementing template transfer, the idea of carrying over learned biases across contexts. In the presented method, a sender-receiver dyad is first trained with a disentangled pair of objectives, and then the receiver is transferred to train a new sender with a standard objective. Unlike other methods, the template transfer approach does not require imposing inductive (...) biases on the architecture of the agents. We experimentally show the emergence of compositional communication using topographical similarity, zero-shot generalization and context-independence as evaluation metrics. The presented approach is connected to an important line of work in semiotics and developmental psycholinguistics: it supports a conjecture that compositional communication is scaffolded on simpler communication protocols. (shrink)

In the literature on expert trust, it is often assumed that track records are the gold standard for evaluating expertise, and the difficulty of expert identification arises from either the lack of access to track records, or the inability to assess them. I show, using a computational model, that even in an idealized environment where agents have a God’s eye view on track records, they may fail to identify experts. Under plausible conditions, selecting testimony based on track records ends up (...) reducing overall accuracy, and preventing the community from identifying the real experts. (shrink)

This article considers how a generalized signalling game may self-assemble as the saliences of the agents evolve by reinforcement on those sources of information that in fact lead to successful action. On the present account, generalized signalling games self-assemble even as the agents co-evolve meaningful representations and successful dispositions for using those representations. We will see how reinforcement on successful information sources also provides a mechanism whereby simpler games might compose to form more complex games. Along the way, I consider (...) how an old game might be appropriated to a new context by reinforcement on successful associations between old and new saliences. (shrink)

We present here a hierarchical model for the evolution of compositional language. The model has the structure of a two-sender/one-receiver Lewis signaling game augmented with executive agents who m...

Wittgenstein used the notion of a language game to illustrate how language is interwoven with action. Here we consider how successful linguistic discourse of the sort he described might emerge in the context of a self-assembling evolutionary game. More specifically, we consider how discourse and coordinated action might self-assemble in the context of two generalized signaling games. The first game shows how prospective language users might learn to initiate meaningful discourse. The second shows how more subtle varieties of discourse might (...) co-emerge with a meaningful language. (shrink)

Typically, public discussions of questions of social import exhibit two important properties: they are influenced by conformity bias, and the influence of conformity is expressed via social networks. We examine how social learning on networks proceeds under the influence of conformity bias. In our model, heterogeneous agents express public opinions where those expressions are driven by the competing priorities of accuracy and of conformity to one’s peers. Agents learn, by Bayesian conditionalization, from private evidence from nature, and from the public (...) declarations of other agents. Our key findings are that networks that produce configurations of social relationships that sustain a diversity of opinions empower honest communication and reliable acquisition of true beliefs, and that the networks that do this best turn out to be those which are both less centralized and less connected. (shrink)