Declan Campbell

Eghbal A. Hosseini, Yuxuan Li, Yasaman Bahri et al. · deepmind, stanford

Large Language Models (LLMs) have been shown to organize the representations of input sequences into straighter neural trajectories in their deep layers, which has been hypothesized to facilitate next-token prediction via linear extrapolation. Language models can also adapt to diverse tasks and learn new structure in context, and recent work has shown that this in-context learning (ICL) can be reflected in representational changes. Here we bring these two lines of research together to explore whether representation straightening occurs \emph{within} a context during ICL. We measure representational straightening in Gemma 2 models across a diverse set of in-context tasks, and uncover a dichotomy in how LLMs' representations change in context. In continual prediction settings (e.g., natural language, grid world traversal tasks) we observe that increasing context increases the straightness of neural sequence trajectories, which is correlated with improvement in model prediction. Conversely, in structured prediction settings (e.g., few-shot tasks), straightening is inconsistent -- it is only present in phases of the task with explicit structure (e.g., repeating a template), but vanishes elsewhere. These results suggest that ICL is not a monolithic process. Instead, we propose that LLMs function like a Swiss Army knife: depending on task structure, the LLM dynamically selects between strategies, only some of which yield representational straightening.

Udith Haputhanthri, Declan Campbell, Rim Assouel et al.

Despite success on standard benchmarks, vision language models display persistent failures on tasks involving processing of multi-object scenes, including many tasks that are relatively easy for humans. Recent work has found that these failures may stem from a basic inability to accurately bind object features in-context, a challenge that is referred to as the "binding problem" in cognitive science and neuroscience. The human visual system is thought to solve this binding problem via serial processing, attending to individual objects one at a time so as to avoid interference from other objects. Recent work has proposed "pointing" -- the use of explicit spatial coordinates to refer to objects -- as an analogous solution for vision language models, and found that it improves performance on challenging multi-object tasks. However, it is unclear $\textit{why}$ (i.e., on a mechanistic or representational level) this approach improves performance, and how directly this relates to serial processing in human vision. Here, we investigate this question. We find that learning to point-via-text induces an internal visual search routine, and we characterize the mechanisms that support this procedure. We also find that pointing behavior can be generalized to new tasks via fine-tuning, and that doing so eliminates binding errors and enables compositional generalization. These results provide a proof-of-principle that serial processing can solve the binding problem for vision language models just as it does for biological vision.

Taylor W. Webb, Steven M. Frankland, Awni Altabaa et al.

A central challenge for cognitive science is to explain how abstract concepts are acquired from limited experience. This has often been framed in terms of a dichotomy between connectionist and symbolic cognitive models. Here, we highlight a recently emerging line of work that suggests a novel reconciliation of these approaches, by exploiting an inductive bias that we term the relational bottleneck. In that approach, neural networks are constrained via their architecture to focus on relations between perceptual inputs, rather than the attributes of individual inputs. We review a family of models that employ this approach to induce abstractions in a data-efficient manner, emphasizing their potential as candidate models for the acquisition of abstract concepts in the human mind and brain.

Declan Campbell, Sunayana Rane, Tyler Giallanza et al.

Recent work has documented striking heterogeneity in the performance of state-of-the-art vision language models (VLMs), including both multimodal language models and text-to-image models. These models are able to describe and generate a diverse array of complex, naturalistic images, yet they exhibit surprising failures on basic multi-object reasoning tasks -- such as counting, localization, and simple forms of visual analogy -- that humans perform with near perfect accuracy. To better understand this puzzling pattern of successes and failures, we turn to theoretical accounts of the binding problem in cognitive science and neuroscience, a fundamental problem that arises when a shared set of representational resources must be used to represent distinct entities (e.g., to represent multiple objects in an image), necessitating the use of serial processing to avoid interference. We find that many of the puzzling failures of state-of-the-art VLMs can be explained as arising due to the binding problem, and that these failure modes are strikingly similar to the limitations exhibited by rapid, feedforward processing in the human brain.

Yukang Yang, Declan Campbell, Kaixuan Huang et al.

Many recent studies have found evidence for emergent reasoning capabilities in large language models (LLMs), but debate persists concerning the robustness of these capabilities, and the extent to which they depend on structured reasoning mechanisms. To shed light on these issues, we study the internal mechanisms that support abstract reasoning in LLMs. We identify an emergent symbolic architecture that implements abstract reasoning via a series of three computations. In early layers, symbol abstraction heads convert input tokens to abstract variables based on the relations between those tokens. In intermediate layers, symbolic induction heads perform sequence induction over these abstract variables. Finally, in later layers, retrieval heads predict the next token by retrieving the value associated with the predicted abstract variable. These results point toward a resolution of the longstanding debate between symbolic and neural network approaches, suggesting that emergent reasoning in neural networks depends on the emergence of symbolic mechanisms.

Declan Campbell, Sreejan Kumar, Tyler Giallanza et al.

Humans possess a remarkable capacity to recognize and manipulate abstract structure, which is especially apparent in the domain of geometry. Recent research in cognitive science suggests neural networks do not share this capacity, concluding that human geometric abilities come from discrete symbolic structure in human mental representations. However, progress in artificial intelligence (AI) suggests that neural networks begin to demonstrate more human-like reasoning after scaling up standard architectures in both model size and amount of training data. In this study, we revisit empirical results in cognitive science on geometric visual processing and identify three key biases in geometric visual processing: a sensitivity towards complexity, regularity, and the perception of parts and relations. We test tasks from the literature that probe these biases in humans and find that large pre-trained neural network models used in AI demonstrate more human-like abstract geometric processing.

Alexander Ku, Declan Campbell, Xuechunzi Bai et al.

Modern artificial intelligence systems, such as large language models, are increasingly powerful but also increasingly hard to understand. Recognizing this problem as analogous to the historical difficulties in understanding the human mind, we argue that methods developed in cognitive science can be useful for understanding large language models. We propose a framework for applying these methods based on the levels of analysis that David Marr proposed for studying information processing systems. By revisiting established cognitive science techniques relevant to each level and illustrating their potential to yield insights into the behavior and internal organization of large language models, we aim to provide a toolkit for making sense of these new kinds of minds.

Sreejan Kumar, Raja Marjieh, Byron Zhang et al.

Humans extract useful abstractions of the world from noisy sensory data. Serial reproduction allows us to study how people construe the world through a paradigm similar to the game of telephone, where one person observes a stimulus and reproduces it for the next to form a chain of reproductions. Past serial reproduction experiments typically employ a single sensory modality, but humans often communicate abstractions of the world to each other through language. To investigate the effect language on the formation of abstractions, we implement a novel multimodal serial reproduction framework by asking people who receive a visual stimulus to reproduce it in a linguistic format, and vice versa. We ran unimodal and multimodal chains with both humans and GPT-4 and find that adding language as a modality has a larger effect on human reproductions than GPT-4's. This suggests human visual and linguistic representations are more dissociable than those of GPT-4.

Nicholas Budny, Kia Ghods, Declan Campbell et al.

Why do Vision Language Models (VLMs), despite success on standard benchmarks, often fail to match human performance on surprisingly simple visual reasoning tasks? While the underlying computational principles are still debated, we hypothesize that a crucial factor is a deficit in visually-grounded serial processing. To test this hypothesis, we compared human and VLM performance across tasks designed to vary serial processing demands in three distinct domains: geometric reasoning, perceptual enumeration, and mental rotation. Tasks within each domain varied serial processing load by manipulating factors such as geometric concept complexity, perceptual individuation load, and transformation difficulty. Across all domains, our results revealed a consistent pattern: decreased VLM accuracy was strongly correlated with increased human reaction time (used as a proxy for serial processing load). As tasks require more demanding serial processing -- whether composing concepts, enumerating items, or performing mental transformations -- the VLM-human performance gap widens reliably. These findings support our hypothesis, indicating that limitations in serial, visually grounded reasoning represent a fundamental bottleneck that distinguishes current VLMs from humans.

Rim Assouel, Declan Campbell, Taylor Webb

To accurately process a visual scene, observers must bind features together to represent individual objects. This capacity is necessary, for instance, to distinguish an image containing a red square and a blue circle from an image containing a blue square and a red circle. Recent work has found that language models solve this 'binding problem' via a set of symbol-like, content-independent indices, but it is unclear whether similar mechanisms are employed by vision language models (VLMs). This question is especially relevant, given the persistent failures of VLMs on tasks that require binding. Here, we identify a set of emergent symbolic mechanisms that support binding in VLMs via a content-independent, spatial indexing scheme. Moreover, we find that binding errors can be traced directly to failures in these mechanisms. Taken together, these results shed light on the mechanisms that support symbol-like processing in VLMs, and suggest possible avenues for addressing the persistent binding failures exhibited by these models.

Declan Campbell, Jonathan D. Cohen

The human cognitive system exhibits remarkable flexibility and generalization capabilities, partly due to its ability to form low-dimensional, compositional representations of the environment. In contrast, standard neural network architectures often struggle with abstract reasoning tasks, overfitting, and requiring extensive data for training. This paper investigates the impact of the relational bottleneck -- a mechanism that focuses processing on relations among inputs -- on the learning of factorized representations conducive to compositional coding and the attendant flexibility of processing. We demonstrate that such a bottleneck not only improves generalization and learning efficiency, but also aligns network performance with human-like behavioral biases. Networks trained with the relational bottleneck developed orthogonal representations of feature dimensions latent in the dataset, reflecting the factorized structure thought to underlie human cognitive flexibility. Moreover, the relational network mimics human biases towards regularity without pre-specified symbolic primitives, suggesting that the bottleneck fosters the emergence of abstract representations that confer flexibility akin to symbols.

Yuxuan Li, Declan Campbell, Stephanie C. Y. Chan et al. · deepmind, stanford

Many of language models' impressive capabilities originate from their in-context learning: based on instructions or examples, they can infer and perform new tasks without weight updates. In this work, we investigate when representations for new tasks are formed in language models, and how these representations change over the course of context. We focus on ''transferrable'' task representations -- vector representations that can restore task contexts in another instance of the model, even without the full prompt. We show that these representations evolve in non-monotonic and sporadic ways, and are distinct from a more inert representation of high-level task categories that persists throughout the context. Specifically, when more examples are provided in the context, transferrable task representations successfully condense evidence. This allows better transfer of task contexts and aligns well with the performance improvement. However, this evidence accrual process exhibits strong locality along the sequence dimension, coming online only at certain tokens -- despite task identity being reliably decodable throughout the context. Moreover, these local but transferrable task representations tend to capture minimal ''task scopes'', such as a semantically-independent subtask. For longer and composite tasks, models rely on more temporally-distributed representations. This two-fold locality (temporal and semantic) underscores a kind of just-in-time computational process that language models use to perform new tasks on the fly.

Marco Nurisso, Jesseba Fernando, Raj Deshpande et al.

Intelligent systems must deploy internal representations that are simultaneously structured -- to support broad generalization -- and selective -- to preserve input identity. We expose a fundamental limit on this tradeoff. For any model whose representational similarity between inputs decays with finite semantic resolution $\varepsilon$, we derive closed-form expressions that pin its probability of correct generalization $p_S$ and identification $p_I$ to a universal Pareto front independent of input space geometry. Extending the analysis to noisy, heterogeneous spaces and to $n>2$ inputs predicts a sharp $1/n$ collapse of multi-input processing capacity and a non-monotonic optimum for $p_S$. A minimal ReLU network trained end-to-end reproduces these laws: during learning a resolution boundary self-organizes and empirical $(p_S,p_I)$ trajectories closely follow theoretical curves for linearly decaying similarity. Finally, we demonstrate that the same limits persist in two markedly more complex settings -- a convolutional neural network and state-of-the-art vision-language models -- confirming that finite-resolution similarity is a fundamental emergent informational constraint, not merely a toy-model artifact. Together, these results provide an exact theory of the generalization-identification trade-off and clarify how semantic resolution shapes the representational capacity of deep networks and brains alike.

Andrew Nam, Declan Campbell, Thomas Griffiths et al.

Neural networks are powerful tools for cognitive modeling due to their flexibility and emergent properties. However, interpreting their learned representations remains challenging due to their sub-symbolic semantics. In this work, we introduce a novel probabilistic framework for interpreting latent task representations in neural networks. Inspired by Bayesian inference, our approach defines a distribution over representational units to infer their causal contributions to task performance. Using ideas from information theory, we propose a suite of tools and metrics to illuminate key model properties, including representational distributedness, manifold complexity, and polysemanticity.