The ability to acquire abstract knowledge is a hallmark of human intelligence and is believed by many to be one of the core differences between humans and neural network models. Agents can be endowed with an inductive bias towards abstraction through meta-learning, where they are trained on a distribution of tasks that share some abstract structure that can be learned and applied. However, because neural networks are hard to interpret, it can be difficult to tell whether agents have learned the underlying abstraction, or alternatively statistical patterns that are characteristic of that abstraction. In this work, we compare the performance of humans and agents in a meta-reinforcement learning paradigm in which tasks are generated from abstract rules. We define a novel methodology for building "task metamers" that closely match the statistics of the abstract tasks but use a different underlying generative process, and evaluate performance on both abstract and metamer tasks. We find that humans perform better at abstract tasks than metamer tasks whereas common neural network architectures typically perform worse on the abstract tasks than the matched metamers. This work provides a foundation for characterizing differences between humans and machine learning that can be used in future work towards developing machines with more human-like behavior.

Strong inductive biases give humans the ability to quickly learn to perform a variety of tasks. Although meta-learning is a method to endow neural networks with useful inductive biases, agents trained by meta-learning may sometimes acquire very different strategies from humans. We show that co-training these agents on predicting representations from natural language task descriptions and programs induced to generate such tasks guides them toward more human-like inductive biases. Human-generated language descriptions and program induction models that add new learned primitives both contain abstract concepts that can compress description length. Co-training on these representations result in more human-like behavior in downstream meta-reinforcement learning agents than less abstract controls (synthetic language descriptions, program induction without learned primitives), suggesting that the abstraction supported by these representations is key.

The extent to which brain functions are localized or distributed is a foundational question in neuroscience. In the human brain, common fMRI methods such as cluster correction, atlas parcellation, and anatomical searchlight are biased by design toward finding localized representations. Here we introduce the functional searchlight approach as an alternative to anatomical searchlight analysis, the most commonly used exploratory multivariate fMRI technique. Functional searchlight removes any anatomical bias by grouping voxels based only on functional similarity and ignoring anatomical proximity. We report evidence that visual and auditory features from deep neural networks and semantic features from a natural language processing model, as well as object representations, are more widely distributed across the brain than previously acknowledged and that functional searchlight can improve model-based similarity and decoding accuracy. This approach provides a new way to evaluate and constrain computational models with brain activity and pushes our understanding of human brain function further along the spectrum from strict modularity toward distributed representation.