Joshua C. Peterson

Raja Marjieh, Nori Jacoby, Joshua C. Peterson et al. · princeton

Shepard's universal law of generalization is a remarkable hypothesis about how intelligent organisms should perceive similarity. In its broadest form, the universal law states that the level of perceived similarity between a pair of stimuli should decay as a concave function of their distance when embedded in an appropriate psychological space. While extensively studied, evidence in support of the universal law has relied on low-dimensional stimuli and small stimulus sets that are very different from their real-world counterparts. This is largely because pairwise comparisons -- as required for similarity judgments -- scale quadratically in the number of stimuli. We provide direct evidence for the universal law in a naturalistic high-dimensional regime by analyzing an existing dataset of 214,200 human similarity judgments and a newly collected dataset of 390,819 human generalization judgments (N=2406 US participants) across three sets of natural images.

Ilia Sucholutsky, Ruairidh M. Battleday, Katherine M. Collins et al.

Supervised learning typically focuses on learning transferable representations from training examples annotated by humans. While rich annotations (like soft labels) carry more information than sparse annotations (like hard labels), they are also more expensive to collect. For example, while hard labels only provide information about the closest class an object belongs to (e.g., "this is a dog"), soft labels provide information about the object's relationship with multiple classes (e.g., "this is most likely a dog, but it could also be a wolf or a coyote"). We use information theory to compare how a number of commonly-used supervision signals contribute to representation-learning performance, as well as how their capacity is affected by factors such as the number of labels, classes, dimensions, and noise. Our framework provides theoretical justification for using hard labels in the big-data regime, but richer supervision signals for few-shot learning and out-of-distribution generalization. We validate these results empirically in a series of experiments with over 1 million crowdsourced image annotations and conduct a cost-benefit analysis to establish a tradeoff curve that enables users to optimize the cost of supervising representation learning on their own datasets.

Jian-Qiao Zhu, Joshua C. Peterson, Benjamin Enke et al. · princeton

Understanding how people behave in strategic settings--where they make decisions based on their expectations about the behavior of others--is a long-standing problem in the behavioral sciences. We conduct the largest study to date of strategic decision-making in the context of initial play in two-player matrix games, analyzing over 90,000 human decisions across more than 2,400 procedurally generated games that span a much wider space than previous datasets. We show that a deep neural network trained on these data predicts people's choices better than leading theories of strategic behavior, indicating that there is systematic variation that is not explained by those theories. We then modify the network to produce a new, interpretable behavioral model, revealing what the original network learned about people: their ability to optimally respond and their capacity to reason about others are dependent on the complexity of individual games. This context-dependence is critical in explaining deviations from the rational Nash equilibrium, response times, and uncertainty in strategic decisions. More broadly, our results demonstrate how machine learning can be applied beyond prediction to further help generate novel explanations of complex human behavior.

Marcel Binz, Elif Akata, Matthias Bethge et al. · princeton

Establishing a unified theory of cognition has been a major goal of psychology. While there have been previous attempts to instantiate such theories by building computational models, we currently do not have one model that captures the human mind in its entirety. A first step in this direction is to create a model that can predict human behavior in a wide range of settings. Here we introduce Centaur, a computational model that can predict and simulate human behavior in any experiment expressible in natural language. We derived Centaur by finetuning a state-of-the-art language model on a novel, large-scale data set called Psych-101. Psych-101 reaches an unprecedented scale, covering trial-by-trial data from over 60,000 participants performing over 10,000,000 choices in 160 experiments. Centaur not only captures the behavior of held-out participants better than existing cognitive models, but also generalizes to new cover stories, structural task modifications, and entirely new domains. Furthermore, we find that the model's internal representations become more aligned with human neural activity after finetuning. Taken together, our results demonstrate that it is possible to discover computational models that capture human behavior across a wide range of domains. We believe that such models provide tremendous potential for guiding the development of cognitive theories and present a case study to demonstrate this.

Ryan Liu, Jiayi Geng, Joshua C. Peterson et al.

In order for AI systems to communicate effectively with people, they must understand how we make decisions. However, people's decisions are not always rational, so the implicit internal models of human decision-making in Large Language Models (LLMs) must account for this. Previous empirical evidence seems to suggest that these implicit models are accurate -- LLMs offer believable proxies of human behavior, acting how we expect humans would in everyday interactions. However, by comparing LLM behavior and predictions to a large dataset of human decisions, we find that this is actually not the case: when both simulating and predicting people's choices, a suite of cutting-edge LLMs (GPT-4o & 4-Turbo, Llama-3-8B & 70B, Claude 3 Opus) assume that people are more rational than we really are. Specifically, these models deviate from human behavior and align more closely with a classic model of rational choice -- expected value theory. Interestingly, people also tend to assume that other people are rational when interpreting their behavior. As a consequence, when we compare the inferences that LLMs and people draw from the decisions of others using another psychological dataset, we find that these inferences are highly correlated. Thus, the implicit decision-making models of LLMs appear to be aligned with the human expectation that other people will act rationally, rather than with how people actually act.

Pulkit Singh, Joshua C. Peterson, Ruairidh M. Battleday et al.

Traditional models of category learning in psychology focus on representation at the category level as opposed to the stimulus level, even though the two are likely to interact. The stimulus representations employed in such models are either hand-designed by the experimenter, inferred circuitously from human judgments, or borrowed from pretrained deep neural networks that are themselves competing models of category learning. In this work, we extend classic prototype and exemplar models to learn both stimulus and category representations jointly from raw input. This new class of models can be parameterized by deep neural networks (DNN) and trained end-to-end. Following their namesakes, we refer to them as Deep Prototype Models, Deep Exemplar Models, and Deep Gaussian Mixture Models. Compared to typical DNNs, we find that their cognitively inspired counterparts both provide better intrinsic fit to human behavior and improve ground-truth classification.

Mayank Agrawal, Joshua C. Peterson, Thomas L. Griffiths

Do large datasets provide value to psychologists? Without a systematic methodology for working with such datasets, there is a valid concern that analyses will produce noise artifacts rather than true effects. In this paper, we offer a way to enable researchers to systematically build models and identify novel phenomena in large datasets. One traditional approach is to analyze the residuals of models---the biggest errors they make in predicting the data---to discover what might be missing from those models. However, once a dataset is sufficiently large, machine learning algorithms approximate the true underlying function better than the data, suggesting instead that the predictions of these data-driven models should be used to guide model-building. We call this approach "Scientific Regret Minimization" (SRM) as it focuses on minimizing errors for cases that we know should have been predictable. We demonstrate this methodology on a subset of the Moral Machine dataset, a public collection of roughly forty million moral decisions. Using SRM, we found that incorporating a set of deontological principles that capture dimensions along which groups of agents can vary (e.g. sex and age) improves a computational model of human moral judgment. Furthermore, we were able to identify and independently validate three interesting moral phenomena: criminal dehumanization, age of responsibility, and asymmetric notions of responsibility.

Joshua C. Peterson, Ruairidh M. Battleday, Thomas L. Griffiths et al.

The classification performance of deep neural networks has begun to asymptote at near-perfect levels. However, their ability to generalize outside the training set and their robustness to adversarial attacks have not. In this paper, we make progress on this problem by training with full label distributions that reflect human perceptual uncertainty. We first present a new benchmark dataset which we call CIFAR10H, containing a full distribution of human labels for each image of the CIFAR10 test set. We then show that, while contemporary classifiers fail to exhibit human-like uncertainty on their own, explicit training on our dataset closes this gap, supports improved generalization to increasingly out-of-training-distribution test datasets, and confers robustness to adversarial attacks.

David D. Bourgin, Joshua C. Peterson, Daniel Reichman et al.

Human decision-making underlies all economic behavior. For the past four decades, human decision-making under uncertainty has continued to be explained by theoretical models based on prospect theory, a framework that was awarded the Nobel Prize in Economic Sciences. However, theoretical models of this kind have developed slowly, and robust, high-precision predictive models of human decisions remain a challenge. While machine learning is a natural candidate for solving these problems, it is currently unclear to what extent it can improve predictions obtained by current theories. We argue that this is mainly due to data scarcity, since noisy human behavior requires massive sample sizes to be accurately captured by off-the-shelf machine learning methods. To solve this problem, what is needed are machine learning models with appropriate inductive biases for capturing human behavior, and larger datasets. We offer two contributions towards this end: first, we construct "cognitive model priors" by pretraining neural networks with synthetic data generated by cognitive models (i.e., theoretical models developed by cognitive psychologists). We find that fine-tuning these networks on small datasets of real human decisions results in unprecedented state-of-the-art improvements on two benchmark datasets. Second, we present the first large-scale dataset for human decision-making, containing over 240,000 human judgments across over 13,000 decision problems. This dataset reveals the circumstances where cognitive model priors are useful, and provides a new standard for benchmarking prediction of human decisions under uncertainty.

Ruairidh M. Battleday, Joshua C. Peterson, Thomas L. Griffiths

Human categorization is one of the most important and successful targets of cognitive modeling in psychology, yet decades of development and assessment of competing models have been contingent on small sets of simple, artificial experimental stimuli. Here we extend this modeling paradigm to the domain of natural images, revealing the crucial role that stimulus representation plays in categorization and its implications for conclusions about how people form categories. Applying psychological models of categorization to natural images required two significant advances. First, we conducted the first large-scale experimental study of human categorization, involving over 500,000 human categorization judgments of 10,000 natural images from ten non-overlapping object categories. Second, we addressed the traditional bottleneck of representing high-dimensional images in cognitive models by exploring the best of current supervised and unsupervised deep and shallow machine learning methods. We find that selecting sufficiently expressive, data-driven representations is crucial to capturing human categorization, and using these representations allows simple models that represent categories with abstract prototypes to outperform the more complex memory-based exemplar accounts of categorization that have dominated in studies using less naturalistic stimuli.

Ori Plonsky, Reut Apel, Eyal Ert et al.

Predicting human decisions under risk and uncertainty remains a fundamental challenge across disciplines. Existing models often struggle even in highly stylized tasks like choice between lotteries. We introduce BEAST Gradient Boosting (BEAST-GB), a hybrid model integrating behavioral theory (BEAST) with machine learning. We first present CPC18, a competition for predicting risky choice, in which BEAST-GB won. Then, using two large datasets, we demonstrate BEAST-GB predicts more accurately than neural networks trained on extensive data and dozens of existing behavioral models. BEAST-GB also generalizes robustly across unseen experimental contexts, surpassing direct empirical generalization, and helps refine and improve the behavioral theory itself. Our analyses highlight the potential of anchoring predictions on behavioral theory even in data-rich settings and even when the theory alone falters. Our results underscore how integrating machine learning with theoretical frameworks, especially those-like BEAST-designed for prediction, can improve our ability to predict and understand human behavior.

Mayank Agrawal, Joshua C. Peterson, Thomas L. Griffiths

Large-scale behavioral datasets enable researchers to use complex machine learning algorithms to better predict human behavior, yet this increased predictive power does not always lead to a better understanding of the behavior in question. In this paper, we outline a data-driven, iterative procedure that allows cognitive scientists to use machine learning to generate models that are both interpretable and accurate. We demonstrate this method in the domain of moral decision-making, where standard experimental approaches often identify relevant principles that influence human judgments, but fail to generalize these findings to "real world" situations that place these principles in conflict. The recently released Moral Machine dataset allows us to build a powerful model that can predict the outcomes of these conflicts while remaining simple enough to explain the basis behind human decisions.

Jordan W. Suchow, Joshua C. Peterson, Thomas L. Griffiths

Generative models of human identity and appearance have broad applicability to behavioral science and technology, but the exquisite sensitivity of human face perception means that their utility hinges on the alignment of the model's representation to human psychological representations and the photorealism of the generated images. Meeting these requirements is an exacting task, and existing models of human identity and appearance are often unworkably abstract, artificial, uncanny, or biased. Here, we use a variational autoencoder with an autoregressive decoder to learn a face space from a uniquely diverse dataset of portraits that control much of the variation irrelevant to human identity and appearance. Our method generates photorealistic portraits of fictive identities with a smooth, navigable latent space. We validate our model's alignment with human sensitivities by introducing a psychophysical Turing test for images, which humans mostly fail. Lastly, we demonstrate an initial application of our model to the problem of fast search in mental space to obtain detailed "police sketches" in a small number of trials.

Joshua C. Peterson, Paul Soulos, Aida Nematzadeh et al.

Modern convolutional neural networks (CNNs) are able to achieve human-level object classification accuracy on specific tasks, and currently outperform competing models in explaining complex human visual representations. However, the categorization problem is posed differently for these networks than for humans: the accuracy of these networks is evaluated by their ability to identify single labels assigned to each image. These labels often cut arbitrarily across natural psychological taxonomies (e.g., dogs are separated into breeds, but never jointly categorized as "dogs"), and bias the resulting representations. By contrast, it is common for children to hear both "dog" and "Dalmatian" to describe the same stimulus, helping to group perceptually disparate objects (e.g., breeds) into a common mental class. In this work, we train CNN classifiers with multiple labels for each image that correspond to different levels of abstraction, and use this framework to reproduce classic patterns that appear in human generalization behavior.

Joshua C. Peterson, Jordan W. Suchow, Krisha Aghi et al.

Understanding how people represent categories is a core problem in cognitive science. Decades of research have yielded a variety of formal theories of categories, but validating them with naturalistic stimuli is difficult. The challenge is that human category representations cannot be directly observed and running informative experiments with naturalistic stimuli such as images requires a workable representation of these stimuli. Deep neural networks have recently been successful in solving a range of computer vision tasks and provide a way to compactly represent image features. Here, we introduce a method to estimate the structure of human categories that combines ideas from cognitive science and machine learning, blending human-based algorithms with state-of-the-art deep image generators. We provide qualitative and quantitative results as a proof-of-concept for the method's feasibility. Samples drawn from human distributions rival those from state-of-the-art generative models in quality and outperform alternative methods for estimating the structure of human categories.

Ruairidh M. Battleday, Joshua C. Peterson, Thomas L. Griffiths

Over the last few decades, psychologists have developed sophisticated formal models of human categorization using simple artificial stimuli. In this paper, we use modern machine learning methods to extend this work into the realm of naturalistic stimuli, enabling human categorization to be studied over the complex visual domain in which it evolved and developed. We show that representations derived from a convolutional neural network can be used to model behavior over a database of >300,000 human natural image classifications, and find that a group of models based on these representations perform well, near the reliability of human judgments. Interestingly, this group includes both exemplar and prototype models, contrasting with the dominance of exemplar models in previous work. We are able to improve the performance of the remaining models by preprocessing neural network representations to more closely capture human similarity judgments.

Joshua C. Peterson, Joshua T. Abbott, Thomas L. Griffiths

Decades of psychological research have been aimed at modeling how people learn features and categories. The empirical validation of these theories is often based on artificial stimuli with simple representations. Recently, deep neural networks have reached or surpassed human accuracy on tasks such as identifying objects in natural images. These networks learn representations of real-world stimuli that can potentially be leveraged to capture psychological representations. We find that state-of-the-art object classification networks provide surprisingly accurate predictions of human similarity judgments for natural images, but fail to capture some of the structure represented by people. We show that a simple transformation that corrects these discrepancies can be obtained through convex optimization. We use the resulting representations to predict the difficulty of learning novel categories of natural images. Our results extend the scope of psychological experiments and computational modeling by enabling tractable use of large natural stimulus sets.

Dawn Chen, Joshua C. Peterson, Thomas L. Griffiths

Vector-space representations provide geometric tools for reasoning about the similarity of a set of objects and their relationships. Recent machine learning methods for deriving vector-space embeddings of words (e.g., word2vec) have achieved considerable success in natural language processing. These vector spaces have also been shown to exhibit a surprising capacity to capture verbal analogies, with similar results for natural images, giving new life to a classic model of analogies as parallelograms that was first proposed by cognitive scientists. We evaluate the parallelogram model of analogy as applied to modern word embeddings, providing a detailed analysis of the extent to which this approach captures human relational similarity judgments in a large benchmark dataset. We find that that some semantic relationships are better captured than others. We then provide evidence for deeper limitations of the parallelogram model based on the intrinsic geometric constraints of vector spaces, paralleling classic results for first-order similarity.

Joshua C. Peterson, Thomas L. Griffiths

Shepard's Universal Law of Generalization offered a compelling case for the first physics-like law in cognitive science that should hold for all intelligent agents in the universe. Shepard's account is based on a rational Bayesian model of generalization, providing an answer to the question of why such a law should emerge. Extending this account to explain how humans use multiple examples to make better generalizations requires an additional assumption, called the size principle: hypotheses that pick out fewer objects should make a larger contribution to generalization. The degree to which this principle warrants similarly law-like status is far from conclusive. Typically, evaluating this principle has not been straightforward, requiring additional assumptions. We present a new method for evaluating the size principle that is more direct, and apply this method to a diverse array of datasets. Our results provide support for the broad applicability of the size principle.

Steven Tang, Joshua C. Peterson, Zachary A. Pardos

Digital learning environments generate a precise record of the actions learners take as they interact with learning materials and complete exercises towards comprehension. With this high quantity of sequential data comes the potential to apply time series models to learn about underlying behavioral patterns and trends that characterize successful learning based on the granular record of student actions. There exist several methods for looking at longitudinal, sequential data like those recorded from learning environments. In the field of language modelling, traditional n-gram techniques and modern recurrent neural network (RNN) approaches have been applied to algorithmically find structure in language and predict the next word given the previous words in the sentence or paragraph as input. In this paper, we draw an analogy to this work by treating student sequences of resource views and interactions in a MOOC as the inputs and predicting students' next interaction as outputs. In this study, we train only on students who received a certificate of completion. In doing so, the model could potentially be used for recommendation of sequences eventually leading to success, as opposed to perpetuating unproductive behavior. Given that the MOOC used in our study had over 3,500 unique resources, predicting the exact resource that a student will interact with next might appear to be a difficult classification problem. We find that simply following the syllabus (built-in structure of the course) gives on average 23% accuracy in making this prediction, followed by the n-gram method with 70.4%, and RNN based methods with 72.2%. This research lays the ground work for recommendation in a MOOC and other digital learning environments where high volumes of sequential data exist.

Joshua C. Peterson, Joshua T. Abbott, Thomas L. Griffiths

Deep neural networks have become increasingly successful at solving classic perception problems such as object recognition, semantic segmentation, and scene understanding, often reaching or surpassing human-level accuracy. This success is due in part to the ability of DNNs to learn useful representations of high-dimensional inputs, a problem that humans must also solve. We examine the relationship between the representations learned by these networks and human psychological representations recovered from similarity judgments. We find that deep features learned in service of object classification account for a significant amount of the variance in human similarity judgments for a set of animal images. However, these features do not capture some qualitative distinctions that are a key part of human representations. To remedy this, we develop a method for adapting deep features to align with human similarity judgments, resulting in image representations that can potentially be used to extend the scope of psychological experiments.