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.

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.

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.

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.

Deep neural networks have become increasingly successful at solving classic perception problems (e.g., recognizing objects), often reaching or surpassing human-level accuracy. In this abridged report of Peterson et al. [2016], we examine the relationship between the image representations learned by these networks and those of humans. 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 appear to capture some key qualitative aspects of human representations. To close this gap, we present 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 and inform human-centric AI.

Modern deep neural networks have achieved impressive results in a variety of automated tasks, such as text generation, grammar learning, and speech recognition. This paper discusses how education research might leverage recurrent neural network architectures in two small case studies. Specifically, we train a two-layer Long Short-Term Memory (LSTM) network on two distinct forms of education data: (1) essays written by students in a summative environment, and (2) MOOC clickstream data. Without any features specified beforehand, the network attempts to learn the underlying structure of the input sequences. After training, the model can be used generatively to produce new sequences with the same underlying patterns exhibited by the input distribution. These early explorations demonstrate the potential for applying deep learning techniques to large education data sets.

Recent studies on image memorability have shed light on what distinguishes the memorability of different images and the intrinsic and extrinsic properties that make those images memorable. However, a clear understanding of the memorability of specific objects inside an image remains elusive. In this paper, we provide the first attempt to answer the question: what exactly is remembered about an image? We augment both the images and object segmentations from the PASCAL-S dataset with ground truth memorability scores and shed light on the various factors and properties that make an object memorable (or forgettable) to humans. We analyze various visual factors that may influence object memorability (e.g. color, visual saliency, and object categories). We also study the correlation between object and image memorability and find that image memorability is greatly affected by the memorability of its most memorable object. Lastly, we explore the effectiveness of deep learning and other computational approaches in predicting object memorability in images. Our efforts offer a deeper understanding of memorability in general thereby opening up avenues for a wide variety of applications.