新研究在乌干达Ngamba孤儿黑猩猩（每项实验15–23只参与者）中显示其可权衡证据强度、更新信念并识别自身知识缺口。实验通过透明窗口或摇动容器提供视觉与听觉线索，并加入增删证据与“假新闻”干扰；黑猩猩能舍弃虚假线索、回忆弱证据并做出正确选择，表现出元认知能力。该结果反驳长期将理性视为人类独享的观点，并得到Duke大学研究者的高度评价。

对比之下，《Science》报道ChatGPT等系统在训练中被系统性削弱元认知，因训练机制奖励自信回答、惩罚不确定性，使其常给出自信但错误的结论。文章指出此类AI虽然具备超越人类的语言与知识检索能力，但缺乏好奇心与用于稳健判断的概率逻辑。动物智能的评估亦具有主观性：猫在某些任务上优于狗，反之亦然，人类长期以等级思维试图以单一尺度衡量不同物种与AI。

研究者强调智能并非神秘先天量，而是具体可测能力的集合。黑猩猩展示的推理结构表明其与人类共享关键认知机制。文章将人类痴迷排序的倾向归因于灵长类的等级本能，导致公众推断若AI在“排名”上超过人类便将统治甚至消灭人类。然而，人类自身几近消灭黑猩猩的历史说明，以支配视角理解智能并不明智。

The new study on orphaned chimpanzees in Uganda’s Ngamba sanctuary (15–23 participants per task) shows they can weigh evidence strength, update beliefs, and identify gaps in their own knowledge. Using visual or auditory cues, altered evidence, and even “fake news,” chimps discarded false information, recalled weak clues, and chose correctly, demonstrating metacognition. The findings challenge the long-held view that reasoning is uniquely human and were praised by Duke researchers.

By contrast, Science reported that systems like ChatGPT show weak metacognition because training rewards confident answers and penalizes uncertainty, producing confident but incorrect responses. Such AI may have superhuman language and vast information access but lacks curiosity and probabilistic reasoning crucial for judgment. Intelligence comparisons remain subjective: cats excel in some tasks, dogs in others, and humans attempt to impose a single hierarchy across species and machines.

Researchers argue intelligence is not a mystical innate quantity but a set of measurable abilities. Chimp reasoning suggests shared cognitive mechanisms with humans. The article links the human drive to rank animals, people, and AIs to primate hierarchical instincts, fueling assumptions that if AI rises “above” us it will dominate. Yet humanity’s near-eradication of chimpanzees illustrates that using dominance as a framework for understanding intelligence is itself irrational.