Prediction, Explanation, and Control: The Use of Mental Models in Dynamic Environments.

The abilities to predict, explain, and control might arise out of operations on a common underlying representation or, conversely, from independent cognitive processes. We developed a novel experimental paradigm to explore how individuals might use probabilistic mental models in these three tasks, under varying levels of complexity and uncertainty. Participants interacted with a simple chatbot defined by a finite-state machine, and were then tested on their ability to predict, explain, and control the chatbot's responses. When full information was available, performance varied significantly across the tasks, with control proving most robust to increased complexity, and explanation being the most challenging. In the presence of hidden information, however, performance across tasks equalized, and participants demonstrated an alternative neglect bias, i.e., a tendency to ignore less likely possibilities. A second, within-subject experimental design then looked for correlations between abilities. We did not find strong correlations, but the challenges of the task for the subjects limited our statistical power. To understand these effects better, a final experiment investigated the possibility of cross-training, skill transfer, or "zero-shot" performance: how well a participant, explicitly trained on one of the three tasks, could perform on the others without additional training. Here we found strong asymmetries: participants trained to control gained generalizable abilities to both predict and explain, while training on either prediction or explanation did not lead to transfer. This cross-training experiment also revealed correlations in performance; most notably between control and prediction. Our findings highlight the complex role of mental models, in contrast to task-specific heuristics, when information is partially hidden, and suggest new avenues for research into situations where the acquisition of general purpose mental models may provide a unifying explanation for a variety of cognitive abilities.

Inferring Cultural Landscapes with the Inverse Ising Model.

The space of possible human cultures is vast, but some cultural configurations are more consistent with cognitive and social constraints than others. This leads to a "landscape" of possibilities that our species has explored over millennia of cultural evolution. However, what does this fitness landscape, which constrains and guides cultural evolution, look like? The machine-learning algorithms that can answer these questions are typically developed for large-scale datasets. Applications to the sparse, inconsistent, and incomplete data found in the historical record have received less attention, and standard recommendations can lead to bias against marginalized, under-studied, or minority cultures. We show how to adapt the minimum probability flow algorithm and the Inverse Ising model, a physics-inspired workhorse of machine learning, to the challenge. A series of natural extensions-including dynamical estimation of missing data, and cross-validation with regularization-enables reliable reconstruction of the underlying constraints. We demonstrate our methods on a curated subset of the Database of Religious History: records from 407 religious groups throughout human history, ranging from the Bronze Age to the present day. This reveals a complex, rugged, landscape, with both sharp, well-defined peaks where state-endorsed religions tend to concentrate, and diffuse cultural floodplains where evangelical religions, non-state spiritual practices, and mystery religions can be found.

The cultural transmission of tacit knowledge.

A wide variety of cultural practices have a 'tacit' dimension, whose principles are neither obvious to an observer, nor known explicitly by experts. This poses a problem for cultural evolution: if beginners cannot spot the principles to imitate, and experts cannot say what they are doing, how can tacit knowledge pass from generation to generation? We present a domain-general model of 'tacit teaching', drawn from statistical physics, that shows how high-accuracy transmission of tacit knowledge is possible. It applies when the practice's underlying features are subject to interacting and competing constraints. Our model makes predictions for key features of the teaching process. It predicts a tell-tale distribution of teaching outcomes, with some students near-perfect performers while others receiving the same instruction are disastrously bad. This differs from standard cultural evolution models that rely on direct, high-fidelity copying, which lead to a much narrower distribution of mostly mediocre outcomes. The model also predicts generic features of the cultural evolution of tacit knowledge. The evolution of tacit knowledge is expected to be bursty, with long periods of stability interspersed with brief periods of dramatic change, and where tacit knowledge, once lost, becomes essentially impossible to recover.

Private versus public: A dual model for resource-constrained conflict representations.

Pietraszewski's representation scheme is parsimonious and intuitive. However, internal mental representations may be subject to resource constraints that prefer more unusual systems such as sparse coding or compressed sensing. Pietraszewski's scheme may be most useful for understanding how agents communicate. Conflict may be driven in part by the complex interplay between parsimonious public representations and more resource-efficient internal ones.

Epistemic phase transitions in mathematical proofs.

Mathematical proofs are both paradigms of certainty and some of the most explicitly-justified arguments that we have in the cultural record. Their very explicitness, however, leads to a paradox, because the probability of error grows exponentially as the argument expands. When a mathematician encounters a proof, how does she come to believe it? Here we show that, under a cognitively-plausible belief formation mechanism combining deductive and abductive reasoning, belief in mathematical arguments can undergo what we call an epistemic phase transition: a dramatic and rapidly-propagating jump from uncertainty to near-complete confidence at reasonable levels of claim-to-claim error rates. To show this, we analyze an unusual dataset of forty-eight machine-aided proofs from the formalized reasoning system Coq, including major theorems ranging from ancient to 21st Century mathematics, along with five hand-constructed cases including Euclid, Apollonius, Hernstein's Topics in Algebra, and Andrew Wiles's proof of Fermat's Last Theorem. Our results bear both on recent work in the history and philosophy of mathematics on how we understand proofs, and on a question, basic to cognitive science, of how we justify complex beliefs.

Learning Communicative Acts in Children's Conversations: A Hidden Topic Markov Model Analysis of the CHILDES Corpora.

Over their first years of life, children learn not just the words of their native languages, but how to use them to communicate. Because manual annotation of communicative intent does not scale to large corpora, our understanding of communicative act development is limited to case studies of a few children at a few time points. We present an approach to automatic identification of communicative acts using a hidden topic Markov model, applying it to the conversations of English-learning children in the CHILDES database. We first describe qualitative changes in parent-child communication over development, and then use our method to demonstrate two large-scale features of communicative development: (a) children develop a parent-like repertoire of our model's communicative acts rapidly, their learning rate peaking around 14 months of age, and (b) this period of steep repertoire change coincides with the highest predictability between parents' acts and children's, suggesting that structured interactions play a role in learning to communicate.

Aggression heuristics underlie animal dominance hierarchies and provide evidence of group-level social information.

Members of a social species need to make appropriate decisions about who, how, and when to interact with others in their group. However, it has been difficult for researchers to detect the inputs to these decisions and, in particular, how much information individuals actually have about their social context. We present a method that can serve as a social assay to quantify how patterns of aggression depend upon information about the ranks of individuals within social dominance hierarchies. Applied to existing data on aggression in 172 social groups across 85 species in 23 orders, it reveals three main patterns of rank-dependent social dominance: the downward heuristic (aggress uniformly against lower-ranked opponents), close competitors (aggress against opponents ranked slightly below self), and bullying (aggress against opponents ranked much lower than self). The majority of the groups (133 groups, 77%) follow a downward heuristic, but a significant minority (38 groups, 22%) show more complex social dominance patterns (close competitors or bullying) consistent with higher levels of social information use. These patterns are not phylogenetically constrained and different groups within the same species can use different patterns, suggesting that heuristic use may depend on context and the structuring of aggression by social information should not be considered a fixed characteristic of a species. Our approach provides opportunities to study the use of social information within and across species and the evolution of social complexity and cognition.

Can we detect conditioned variation in political speech? two kinds of discussion and types of conversation.

Previous work has demonstrated that certain speech patterns vary systematically between sociodemographic groups, so that in some cases the way a person speaks is a valid cue to group membership. Our work addresses whether or not participants use these linguistic cues when assessing a speaker's likely political identity. We use a database of speeches by U.S. Congressional representatives to isolate words that are statistically diagnostic of a speaker's party identity. In a series of four studies, we demonstrate that participants' judgments track variation in word usage between the two parties more often than chance, and that this effect persists even when potentially interfering cues such as the meaning of the word are controlled for. Our results are consistent with a body of literature suggesting that humans' language-related judgments reflect the statistical distributions of our environment.

From Probability to Consilience: How Explanatory Values Implement Bayesian Reasoning.

Recent work in cognitive science has uncovered a diversity of explanatory values, or dimensions along which we judge explanations as better or worse. We propose a Bayesian account of these values that clarifies their function and shows how they fit together to guide explanation-making. The resulting taxonomy shows that core values from psychology, statistics, and the philosophy of science emerge from a common mathematical framework and provide insight into why people adopt the explanations they do. This framework not only operationalizes the explanatory virtues associated with, for example, scientific argument-making, but also enables us to reinterpret the explanatory vices that drive phenomena such as conspiracy theories, delusions, and extremist ideologies.

How We Do Things With Words: Analyzing Text as Social and Cultural Data.

In this article we describe our experiences with computational text analysis involving rich social and cultural concepts. We hope to achieve three primary goals. First, we aim to shed light on thorny issues not always at the forefront of discussions about computational text analysis methods. Second, we hope to provide a set of key questions that can guide work in this area. Our guidance is based on our own experiences and is therefore inherently imperfect. Still, given our diversity of disciplinary backgrounds and research practices, we hope to capture a range of ideas and identify commonalities that resonate for many. This leads to our final goal: to help promote interdisciplinary collaborations. Interdisciplinary insights and partnerships are essential for realizing the full potential of any computational text analysis involving social and cultural concepts, and the more we bridge these divides, the more fruitful we believe our work will be.

Opacity, obscurity, and the geometry of question-asking.

Asking questions is a pervasive human activity, but little is understood about what makes them difficult to answer. An analysis of a pair of large databases, New York Times crosswords and questions from the quiz-show Jeopardy, establishes two orthogonal dimensions of question difficulty: obscurity (the rarity of the answer) and opacity (the indirectness of question cues, operationalized with word2vec). The importance of opacity, and the role of synergistic information in resolving it, suggests that accounts of difficulty in terms of prior expectations captures only a part of the question-asking process. A further regression analysis shows the presence of additional dimensions to question-asking: question complexity, the answer's local network density, cue intersection, and the presence of signal words. Our work shows how question-askers can help their interlocutors by using contextual cues, or, conversely, how a particular kind of unfamiliarity with the domain in question can make it harder for individuals to learn from others. Taken together, these results suggest how Bayesian models of question difficulty can be supplemented by process models and accounts of the heuristics individuals use to navigate conceptual spaces.

Individuals, institutions, and innovation in the debates of the French Revolution.

The French Revolution brought principles of "liberty, equality, fraternity" to bear on the day-to-day challenges of governing what was then the largest country in Europe. Its experiments provided a model for future revolutions and democracies across the globe, but this first modern revolution had no model to follow. Using reconstructed transcripts of debates held in the Revolution's first parliament, we present a quantitative analysis of how this body managed innovation. We use information theory to track the creation, transmission, and destruction of word-use patterns across over 40,000 speeches and a thousand speakers. The parliament as a whole was biased toward the adoption of new patterns, but speakers' individual qualities could break these overall trends. Speakers on the left innovated at higher rates, while speakers on the right acted to preserve prior patterns. Key players such as Robespierre (on the left) and Abbé Maury (on the right) played information-processing roles emblematic of their politics. Newly created organizational functions-such as the Assembly president and committee chairs-had significant effects on debate outcomes, and a distinct transition appears midway through the parliament when committees, external to the debate process, gained new powers to "propose and dispose." Taken together, these quantitative results align with existing qualitative interpretations, but also reveal crucial information-processing dynamics that have hitherto been overlooked. Great orators had the public's attention, but deputies (mostly on the political left) who mastered the committee system gained new powers to shape revolutionary legislation.

The evolution of lossy compression.

In complex environments, there are costs to both ignorance and perception. An organism needs to track fitness-relevant information about its world, but the more information it tracks, the more resources it must devote to perception. As a first step towards a general understanding of this trade-off, we use a tool from information theory, rate-distortion theory, to study large, unstructured environments with fixed, randomly drawn penalties for stimuli confusion ('distortions'). We identify two distinct regimes for organisms in these environments: a high-fidelity regime where perceptual costs grow linearly with environmental complexity, and a low-fidelity regime where perceptual costs are, remarkably, independent of the number of environmental states. This suggests that in environments of rapidly increasing complexity, well-adapted organisms will find themselves able to make, just barely, the most subtle distinctions in their environment.

Exploration and exploitation of Victorian science in Darwin's reading notebooks.

Search in an environment with an uncertain distribution of resources involves a trade-off between exploitation of past discoveries and further exploration. This extends to information foraging, where a knowledge-seeker shifts between reading in depth and studying new domains. To study this decision-making process, we examine the reading choices made by one of the most celebrated scientists of the modern era: Charles Darwin. From the full-text of books listed in his chronologically-organized reading journals, we generate topic models to quantify his local (text-to-text) and global (text-to-past) reading decisions using Kullback-Liebler Divergence, a cognitively-validated, information-theoretic measure of relative surprise. Rather than a pattern of surprise-minimization, corresponding to a pure exploitation strategy, Darwin's behavior shifts from early exploitation to later exploration, seeking unusually high levels of cognitive surprise relative to previous eras. These shifts, detected by an unsupervised Bayesian model, correlate with major intellectual epochs of his career as identified both by qualitative scholarship and Darwin's own self-commentary. Our methods allow us to compare his consumption of texts with their publication order. We find Darwin's consumption more exploratory than the culture's production, suggesting that underneath gradual societal changes are the explorations of individual synthesis and discovery. Our quantitative methods advance the study of cognitive search through a framework for testing interactions between individual and collective behavior and between short- and long-term consumption choices. This novel application of topic modeling to characterize individual reading complements widespread studies of collective scientific behavior.

Weak universality in sensory tradeoffs.

For many organisms, the number of sensory neurons is largely determined during development, before strong environmental cues are present. This is despite the fact that environments can fluctuate drastically both from generation to generation and within an organism's lifetime. How can organisms get by by hard coding the number of sensory neurons? We approach this question using rate-distortion theory. A combination of simulation and theory suggests that when environments are large, the rate-distortion function-a proxy for material costs, timing delays, and energy requirements-depends only on coarse-grained environmental statistics that are expected to change on evolutionary, rather than ontogenetic, time scales.

Social Feedback and the Emergence of Rank in Animal Society.

Dominance hierarchies are group-level properties that emerge from the aggression of individuals. Although individuals can gain critical benefits from their position in a hierarchy, we do not understand how real-world hierarchies form. Nor do we understand what signals and decision-rules individuals use to construct and maintain hierarchies in the absence of simple cues such as size or spatial location. A study of conflict in two groups of captive monk parakeets (Myiopsitta monachus) found that a transition to large-scale order in aggression occurred in newly-formed groups after one week, with individuals thereafter preferring to direct aggression more frequently against those nearby in rank. We consider two cognitive mechanisms underlying the emergence of this order: inference based on overall levels of aggression, or on subsets of the aggression network. Both mechanisms were predictive of individual decisions to aggress, but observed patterns were better explained by rank inference through subsets of the aggression network. Based on these results, we present a new theory, of a feedback loop between knowledge of rank and consequent behavior. This loop explains the transition to strategic aggression and the formation and persistence of dominance hierarchies in groups capable of both social memory and inference.

The civilizing process in London's Old Bailey.

The jury trial is a critical point where the state and its citizens come together to define the limits of acceptable behavior. Here we present a large-scale quantitative analysis of trial transcripts from the Old Bailey that reveal a major transition in the nature of this defining moment. By coarse-graining the spoken word testimony into synonym sets and dividing the trials based on indictment, we demonstrate the emergence of semantically distinct violent and nonviolent trial genres. We show that although in the late 18th century the semantic content of trials for violent offenses is functionally indistinguishable from that for nonviolent ones, a long-term, secular trend drives the system toward increasingly clear distinctions between violent and nonviolent acts. We separate this process into the shifting patterns that drive it, determine the relative effects of bureaucratic change and broader cultural shifts, and identify the synonym sets most responsible for the eventual genre distinguishability. This work provides a new window onto the cultural and institutional changes that accompany the monopolization of violence by the state, described in qualitative historical analysis as the civilizing process.

Collective phenomena and non-finite state computation in a human social system.

We investigate the computational structure of a paradigmatic example of distributed social interaction: that of the open-source Wikipedia community. We examine the statistical properties of its cooperative behavior, and perform model selection to determine whether this aspect of the system can be described by a finite-state process, or whether reference to an effectively unbounded resource allows for a more parsimonious description. We find strong evidence, in a majority of the most-edited pages, in favor of a collective-state model, where the probability of a "revert" action declines as the square root of the number of non-revert actions seen since the last revert. We provide evidence that the emergence of this social counter is driven by collective interaction effects, rather than properties of individual users.