While previously thought to be uniquely human, cumulative cultural evolution continues to be found in non-human animals. It occurs when an adaptive innovation from an individual is repeatedly passed onto consecutive generations through social learning. For example, pigeons who fly alone or in stable pairs show relatively rigid sub-optimal routes, but gradually improve route efficiency over generations of pairs in which experienced members are swapped for naive ones. This raises the question of what the minimally required cognitive architecture is for cumulative cultural evolution to emerge. Here, I aimed to answer this question in artificial agents who employ three main functions: goal-direction, social proximity, and route memory. At the optima for efficiency and generational efficiency improvement, agents replicated cumulative culture observed in pigeons. At each optimum, paths were determined primarily by memory, and to a lesser extent by social proximity and goal-direction. Because of their need for social proximity, each naive agent stayed close to their experienced counterpart as that followed its memorised path. However, unhindered by route memory, the naive agent's heading was more likely to err towards the goal. This subtly biased pairs' routes, and the resulting efficiency improvement is thus regression to the goal. The resulting incremental improvements over generations meet all core criteria in current frameworks of cumulative cultural evolution, suggesting that rudimentary cumulative optimisation is an evolutionary mechanism that emerges even in simple systems that prefer social proximity and have a memory capacity.

本研究旨在回答什么是孕育文化演化所需的最小认知结构，通过人工智能代理的实验结果表明，具备目标导向、社交亲近和路线记忆的个体，在社交学习中存在类似于鸽子的文化演化和优化的现象。

社交和记忆为指导的人工导航器自发地形成累积文化