Large language models generate complex, open-ended outputs: instead of outputting a class label they write summaries, generate dialogue, or produce working code. In order to asses the reliability of these open-ended generation systems, we aim to identify qualitative categories of erroneous behavior, beyond identifying individual errors. To hypothesize and test for such qualitative errors, we draw inspiration from human cognitive biases--systematic patterns of deviation from rational judgement. Specifically, we use cognitive biases as motivation to (i) generate hypotheses for problems that models may have, and (ii) develop experiments that elicit these problems. Using code generation as a case study, we find that OpenAI's Codex errs predictably based on how the input prompt is framed, adjusts outputs towards anchors, and is biased towards outputs that mimic frequent training examples. We then use our framework to elicit high-impact errors such as incorrectly deleting files. Our results indicate that experimental methodology from cognitive science can help characterize how machine learning systems behave.

Computational cognitive models are a fundamental tool in behavioral neuroscience. They embody in software precise hypotheses about the cognitive mechanisms underlying a particular behavior. Constructing these models is typically a difficult iterative process that requires both inspiration from the literature and the creativity of an individual researcher. Here, we adopt an alternative approach to learn parsimonious cognitive models directly from data. We fit behavior data using a recurrent neural network that is penalized for carrying excess information between timesteps, leading to sparse, interpretable representations and dynamics.

Humans learn from visual inputs at multiple timescales, both rapidly and flexibly acquiring visual knowledge over short periods, and robustly accumulating online learning progress over longer periods. Modeling these powerful learning capabilities is an important problem for computational visual cognitive science, and models that could replicate them would be of substantial utility in real-world computer vision settings. In this work, we establish benchmarks for both real-time and life-long continual visual learning. Our real-time learning benchmark measures a model's ability to match the rapid visual behavior changes of real humans over the course of minutes and hours, given a stream of visual inputs. Our life-long learning benchmark evaluates the performance of models in a purely online learning curriculum obtained directly from child visual experience over the course of years of development.

Contaminant observations and outliers often cause problems when estimating the parameters of cognitive models, which are statistical models representing cognitive processes. In this study, we test and improve the robustness of parameter estimation using amortized Bayesian inference (ABI) with neural networks. To this end, we conduct systematic analyses on a toy example and analyze both synthetic and real data using a popular cognitive model, the Drift Diffusion Models (DDM). First, we study the sensitivity of ABI to contaminants with tools from robust statistics: the empirical influence function and the breakdown point. Next, we propose a data augmentation or noise injection approach that incorporates a contamination distribution into the data-generating process during training. We examine several candidate distributions and evaluate their performance and cost in terms of accuracy and efficiency loss relative to a standard estimator. Introducing contaminants from a Cauchy distribution during training considerably increases the robustness of the neural density estimator as measured by bounded influence functions and a much higher breakdown point. Overall, the proposed method is straightforward and practical to implement and has a broad applicability in fields where outlier detection or removal is challenging.

How do the synthetic neurons in language models create "thought categories" to segment and analyze their informational environment? What are the cognitive characteristics, at the very level of formal neurons, of this artificial categorical thought? Based on the mathematical nature of algebraic operations inherent to neuronal aggregation functions, we attempt to identify mathematico-cognitive factors that genetically shape the categorical reconstruction of the informational world faced by artificial cognition. This study explores these concepts through the notions of priming, attention, and categorical phasing.

Word-by-word language model surprisal is often used to model the incremental processing of human readers, which raises questions about how various choices in language modeling influence its predictive power. One factor that has been overlooked in cognitive modeling is the granularity of subword tokens, which explicitly encodes information about word length and frequency, and ultimately influences the quality of vector representations that are learned. This paper presents experiments that manipulate the token granularity and evaluate its impact on the ability of surprisal to account for processing difficulty of naturalistic text and garden-path constructions. Experiments with naturalistic reading times reveal a substantial influence of token granularity on surprisal, with tokens defined by a vocabulary size of 8,000 resulting in surprisal that is most predictive. In contrast, on garden-path constructions, language models trained on coarser-grained tokens generally assigned higher surprisal to critical regions, suggesting their increased sensitivity to syntax. Taken together, these results suggest a large role of token granularity on the quality of language model surprisal for cognitive modeling.

As AI systems become increasingly powerful, the need for safe AI has become more pressing. Humans are an attractive model for AI safety: as the only known agents capable of general intelligence, they perform robustly even under conditions that deviate significantly from prior experiences, explore the world safely, understand pragmatics, and can cooperate to meet their intrinsic goals. Intelligence, when coupled with cooperation and safety mechanisms, can drive sustained progress and well-being. These properties are a function of the architecture of the brain and the learning algorithms it implements. Neuroscience may thus hold important keys to technical AI safety that are currently underexplored and underutilized. In this roadmap, we highlight and critically evaluate several paths toward AI safety inspired by neuroscience: emulating the brain's representations, information processing, and architecture; building robust sensory and motor systems from imitating brain data and bodies; fine-tuning AI systems on brain data; advancing interpretability using neuroscience methods; and scaling up cognitively-inspired architectures. We make several concrete recommendations for how neuroscience can positively impact AI safety.

When we speak, write or listen, we continuously make predictions based on our knowledge of a language's grammar. Remarkably, children acquire this grammatical knowledge within just a few years, enabling them to understand and generalise to novel constructions that have never been uttered before. Language models are powerful tools that create representations of language by incrementally predicting the next word in a sentence, and they have had a tremendous societal impact in recent years. The central research question of this thesis is whether these models possess a deep understanding of grammatical structure similar to that of humans. This question lies at the intersection of natural language processing, linguistics, and interpretability. To address it, we will develop novel interpretability techniques that enhance our understanding of the complex nature of large-scale language models. We approach our research question from three directions. First, we explore the presence of abstract linguistic information through structural priming, a key paradigm in psycholinguistics for uncovering grammatical structure in human language processing. Next, we examine various linguistic phenomena, such as adjective order and negative polarity items, and connect a model's comprehension of these phenomena to the data distribution on which it was trained. Finally, we introduce a controlled testbed for studying hierarchical structure in language models using various synthetic languages of increasing complexity and examine the role of feature interactions in modelling this structure. Our findings offer a detailed account of the grammatical knowledge embedded in language model representations and provide several directions for investigating fundamental linguistic questions using computational methods.

Training agents that can coordinate zero-shot with humans is a key mission in multi-agent reinforcement learning (MARL). Current algorithms focus on training simulated human partner policies which are then used to train a Cooperator agent. The simulated human is produced either through behavior cloning over a dataset of human cooperation behavior, or by using MARL to create a population of simulated agents. However, these approaches often struggle to produce a Cooperator that can coordinate well with real humans, since the simulated humans fail to cover the diverse strategies and styles employed by people in the real world. We show \emph{learning a generative model of human partners} can effectively address this issue. Our model learns a latent variable representation of the human that can be regarded as encoding the human's unique strategy, intention, experience, or style. This generative model can be flexibly trained from any (human or neural policy) agent interaction data. By sampling from the latent space, we can use the generative model to produce different partners to train Cooperator agents. We evaluate our method -- \textbf{G}enerative \textbf{A}gent \textbf{M}odeling for \textbf{M}ulti-agent \textbf{A}daptation (GAMMA) -- on Overcooked, a challenging cooperative cooking game that has become a standard benchmark for zero-shot coordination. We conduct an evaluation with real human teammates, and the results show that GAMMA consistently improves performance, whether the generative model is trained on simulated populations or human datasets. Further, we propose a method for posterior sampling from the generative model that is biased towards the human data, enabling us to efficiently improve performance with only a small amount of expensive human interaction data.

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. 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, Centaur is the first real candidate for a unified model of human cognition. We anticipate that it will have a disruptive impact on the cognitive sciences, challenging the existing paradigm for developing computational models.