中文导读 人们常说好奇心来自复杂的人脑，一项新的建模研究却把这个因果顺序反过来想：寻求新信息的驱力，可能本身就在帮助大脑变得更复杂。剑桥大学 Francesco Poli 团队先用 275 名 8 至 15 个月婴儿的眼动数据观察学习节奏，再把“学习变慢就换任务”的规则写进人工神经网络；高好奇心模型在未见过的测试上更灵活，并在类似人脑的三维连线约束下出现模块结构和连接先密后疏的轨迹。文章最后保留外部专家的质疑，并写到对早期教育中“强制长时间只做一件事”的提醒。阅读时可留意 however、when、if、while 如何连接发现、对照与边界，以及 may、suggests、argues、still early 如何表达不确定。
正文 People often say that curiosity comes from a complex human brain . A new modelling study turns that idea around . It suggests that the drive to seek new information may help brains become complex in the first place .
本段译文 人们常说好奇心来自复杂的人脑。一项新的建模研究把这个想法反过来：寻求新信息的驱力，可能首先就在帮助大脑变得复杂。
Francesco Poli of the University of Cambridge and his colleagues started with babies . They recorded eye movements from 275 infants aged between eight and 15 months . The babies watched sequences of colourful shapes . The patterns were not random , so each shape carried different amounts of useful information about where the next one would appear . When the infants rate of learning fell below average , they looked away and searched for something new .
本段译文 剑桥大学的 Francesco Poli 与同事从婴儿入手。他们记录了 275 名 8 至 15 个月大婴儿的眼动。婴儿观看彩色形状组成的序列。这些模式并非随机，因此每个形状携带多少关于下一位置的有用信息并不相同。当婴儿的学习速度降到平均水平以下时，他们会把目光移开，去寻找新的东西。
The team built that rule into artificial neural networks Models with high curiosity switched tasks as soon as their learning rate dropped below a set level Models with low curiosity switched at random . All of the networks became skilled at the 20 training tasks . After training , however , the curious models did much better on tests they had never seen . That flexible mixing of skills is often treated as a mark of complex human thinking .
本段译文 研究团队把这条规则写进人工神经网络。高好奇心模型一旦学习速度降到设定阈值以下就切换任务；低好奇心模型则随机切换。所有网络都熟练掌握了 20 项训练任务。但训练结束后，好奇模型在从未见过的测试上表现好得多。这种灵活组合技能的能力，常被看作复杂人类思维的标志。
The researchers also placed the networks in a virtual three-dimensional map shaped like a human brain and made links over a long distance costly Curiosity plus that wiring cost produced nested modules that look more like a real brain Connection density rose and then fell , a pattern similar to the early rise and later pruning of synapses in children . Mapped against a human growth curve , the most curious networks looked about 20 years old in developmental terms , while less curious ones stayed nearer to age nine .
本段译文 研究人员还把网络放进一个形似人脑的虚拟三维地图，并让跨越较长距离的连接付出更高成本。好奇心加上这种连线成本，产生了更像真实大脑的嵌套模块。连接密度先升后降，类似于儿童早期突触先增多、再修剪的模式。对照人类生长曲线，最好奇的网络在发育意义上大约相当于 20 岁，而好奇心较低的网络更接近 9 岁。
Outside experts note limits . Alireza Modirshanechi of the University of G öttingen said it is no surprise that curiosity makes learning efficient , but the claim that it shapes structure like a brain is strong and needs more checks . He asked whether networks without curiosity might catch up with much longer training . Poli argues they already matched curious models on the basic tasks , so the gap is not only about speed
本段译文 外部专家指出局限。德国哥廷根大学的 Alireza Modirshanechi 说，好奇心提高学习效率并不奇怪，但它塑造像大脑一样的结构的说法很强，还需要更多核对。他追问：没有好奇心的网络若训练更久，会不会最终赶上。Poli 认为它们在基础任务上已经与好奇模型相当，因此差距不只是速度问题。
The preprint on bioRxiv , covered by New Scientist on 2 November 2026, also speaks to classrooms . Poli warns that forcing young children to stay with one activity for a long time may work against curiosity . If the goal is flexible thinking , he says , children may need more choice of tasks and freer exploration within clear bounds . The work is still early modelling , not a final map of the human brain
本段译文 这篇发表在 bioRxiv 上的预印本，由《新科学家》于 2026 年 11 月 2 日报道，也谈到课堂。Poli 提醒，强迫幼儿长时间只做一件事，可能不利于好奇心。若目标是灵活思维，他说，孩子可能需要更多任务选择，并在明确边界内更自由地探索。这项工作仍是早期建模，并不是人类大脑的最终地图。
新闻来源