Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning

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.Author summary: There has been a recent explosion of progress in artificial intelligence models, in the form of neural network models. As these models achieve human-level performance in a variety of task domains, one may ask what exactly is the difference between these models and human intelligence. Many researchers have hypothesized that neural networks often learn to solve problems through simple pattern matching while humans can often understand a problem’s underlying abstract concepts or causal mechanisms and solve it using reasoning. Because it is difficult to tell which of these two strategies is being employed in problem solving, this work develops a method to disentangle the two from a human’s or neural network model’s behavior. The findings confirm that humans typically use abstraction to solve problems whereas neural networks typically use pattern matching instead.