Kay, Kenneth, Biderman, Natalie, Khajeh, Ramin, Beiran, Manuel, Cueva, Christopher J., Shohamy, Daphna, Jensen, Greg, Wei, Xue-Xin, Ferrera, Vincent P., and Abbott, LF
Relational cognition—the ability to infer relationships that generalize to novel combinations of objects—is fundamental to human and animal intelligence. Despite this importance, it remains unclear how relational cognition is implemented in the brain due in part to a lack of hypotheses and predictions at the levels of collective neural activity and behavior. Here we discovered, analyzed, and experimentally tested neural networks (NNs) that perform transitive inference (TI), a classic relational task (if A > B and B > C, then A > C). We found NNs that (i) generalized perfectly, despite lacking overt transitive structure prior to training, (ii) generalized when the task required working memory (WM), a capacity thought to be essential to inference in the brain, (iii) emergently expressed behaviors long observed in living subjects, in addition to a novel order-dependent behavior, and (iv) expressed different task solutions yielding alternative behavioral and neural predictions. Further, in a large-scale experiment, we found that human subjects performing WM-based TI showed behavior inconsistent with a class of NNs that characteristically expressed an intuitive task solution. These findings provide neural insights into a classical relational ability, with wider implications for how the brain realizes relational cognition. Author summary: The ability to infer underlying relationships in the world—or "relational cognition"—is fundamental to both human and animal intelligence. Despite this importance, it remains generally unknown how relational abilities are realized in the brain. Addressing this lack of knowledge requires generating testable hypotheses regarding the activity of neurons when relationships are inferred. In this study, we discovered artificial neural networks that perform transitive inference (TI), a classic relational task that has the form "if A > B and B > C, then A > C." We found that these neural networks show behavioral patterns that match those seen in humans and animals performing TI, and, further, performed TI in different ways, thereby providing different testable hypotheses regarding neural activity. We also conducted a large-scale behavioral study in humans, which suggested systematic differences in biological accuracy across the neural networks. These findings provide new hypotheses and insights regarding how TI, and relational abilities more generally, can be realized in the brain. [ABSTRACT FROM AUTHOR]