Collective ant transport outperforms gravity-based solvers on complex puzzles.

Abstract

Collective cognition, in which groups display enhanced problem-solving abilities compared with individuals, is a hallmark of ant behaviour. For instance, in navigation tasks, such as the piano movers' problem, a short-term memory-like directional persistence emerges in large enough ant groups and aids them in implementing an effective wall-following heuristic. Here, we investigate the collective problem-solving abilities of Paratrechina longicornis by presenting groups of varying sizes with an array of piano movers' puzzles. We find that increasingly complex puzzles reveal performance differences between small and large groups. To benchmark these results, we compare the performance of the ants to a simulated physics-based null model incorporating gravity and noise. Although the null model performs comparably with ants in simple puzzles, it fails in more challenging ones. Introducing ant-inspired mechanisms, including attachment along edges or transient leadership, enables the simulated solver to match ant performance across group sizes and tasks. Beyond these features, we show that ants solve a broad spectrum of puzzles without prior knowledge of geometry, whereas the simulated solver requires parameter adjustments tailored to each puzzle. Our findings highlight the flexibility and robustness of collective cognition in ant groups and provide a framework for integrating biological strategies into artificial problem-solving systems.