Even some researchers aware of the historical importance of neuroscience in shaping the field of AI often argue that it has lost its relevance. “Engineers do not study birds to build better planes” is the usual refrain. But the analogy fails, in part because pioneers of aviation did indeed study birds, and some still do. Moreover, the analogy fails also at a more fundamental level: The goal of modern aeronautical engineering is not to achieve “bird-level” flight, whereas a major goal of AI is indeed to achieve (or exceed) “human-level” intelligence. Just as computers exceed humans in many respects，such as the ability to compute prime numbers, so too do planes exceed birds in characteristics such as speed, range and cargo capacity. But if the goal of aeronautical engineers were indeed to build a machine with the "bird-level" ability to fly through dense forest foliage and alight gently on a branch, they would be well-advised to pay very close attention to how birds do it. Similarly, if Al aims to achieve animal-level common-sense sensorimotor intelligence, researchers would be well-advised to learn from animals and the solutions they evolved to behave in an unpredictable world.

There are, however, significant methodological and cultural gaps between machine learning (ML) research and control/systems research. The most prominent difference is in the role of mathematical models. Much of the current mainstream ML research is (proudly) model free, whereas most of modern control is model based. Despite enormous interest and serious research efforts, it has been very difficult to provide strong mathematical foundations (guarantees) for state-of-the-art ML techniques. It is not unfair to say that the implementations of the latest and most powerful ML techniques do not come with a deep understanding of how and why they work as well as they appear to and when they might fail. This situation is in sharp contrast to the traditions in control systems that insist on hard mathematical theorems and proofs for supporting research claims and progress. While the learning for control research community is grappling with the methodological and cultural gaps, they remain very serious obstacles to truly leveraging the best of these two research streams. 

不囿旧隅拘浅见，勇开新境拓遐研。本工作（Wang et al, PRL, 2026）与课题组此前的系列论文从不同维度分别研究了不确定性对于博弈动力学的影响。这些研究共同构成了研究不确定性如何驱动博弈动力学的系统性理论框架——鲁棒博弈理论(Robust Game Theory)。

第一个维度是交互不确定性（Wang et al, PNAS, 2024）。该工作关注个体在感知收益时引入的观察噪声，并采用期望效用理论来刻画个体对不确定性的差异化响应——风险厌恶者倾向于选择低方差策略，风险追逐者反之。第二个维度是群体规模不确定性（Wang et al, PNAS, 2023）。该工作聚焦于策略传播速率的随机性导致的群体规模波动，通过引入随机微分方程，揭示了有限群体中噪声如何逆转演化动力学的方向。第三个维度是本文研究的环境全局不确定性（Wang et al, PRL, 2026）。本文关注作用于整个群体的全局噪声，与个体层面收益观测的噪声和策略传播过程的噪声形成互补。

分途渐探弈中理，累论终成至道诠。三个维度共同指向一个核心结论：不确定性不是演化系统中可忽略的随机扰动，而是能够改变系统动力学结构的基本要素。这一系列工作表明，在不确定性普遍存在的现实世界中，演化过程不会简单地“平均掉”噪声，而是可能利用噪声创造新的稳定结构。这为理解复杂系统中的秩序涌现提供了新的理论视角，也为群体演化过程和智能集群系统提供了可操作的调控手段。

文章链接：https://doi.org/10.1103/3yby-qq2n

相关研究：

[1] G. Wang, Q. Su, L. Wang, J. B. Plotkin, Reproductive variance can drive behavioral dynamics. PNAS. 120, e2216218120 (2023).

[2] G. Wang, Q. Su, L. Wang, J. B. Plotkin, The evolution of social behaviors and risk preferences in settings with uncertainty. PNAS, e2406993121 (2024).

支持单位：