Bumblebees can spontaneously solve problems, study finds

研究发现：大黄蜂能够自发解决问题

Despite having tiny brains, bumblebees have demonstrated a remarkable ability to socially learn how to use tools, solve simple puzzles, and cooperate to achieve a goal. It seems they can also solve object-manipulation tasks without any previous training, according to a new paper published in the journal Science. According to the authors, it’s the first time this kind of spontaneous problem-solving has been demonstrated in an insect.

尽管大脑微小，大黄蜂却展现出了惊人的能力，能够通过社交学习来使用工具、解决简单的谜题并协作达成目标。根据发表在《科学》杂志上的一篇新论文，大黄蜂似乎还可以在没有任何预先训练的情况下解决物体操纵任务。作者表示，这是首次在昆虫身上观察到这种自发的问题解决能力。

In 2024, Olli Loukola of the University of Finland co-authored a study demonstrating that bumblebees could cooperate to solve complex challenges. It’s the kind of cognitive task scientists had previously only observed in large-brained mammals like humans and chimpanzees. Loukola et al. trained pairs of bees to push a Lego block to the middle of a mini-arena or push against a door at the end of a tunnel to get a reward. The team noticed that the bees were more likely to engage in the tasks if their partners also participated, compared to untrained control groups. They concluded that bees can learn to solve novel cooperative tasks outside the hive and may even be intentionally working together, although the researchers cautioned that more detailed monitoring of the behavior was needed to fully understand the partners’ roles.

2024年，芬兰大学的奥利·卢科拉（Olli Loukola）参与合著了一项研究，证明了大黄蜂可以通过协作来解决复杂的挑战。这种认知任务此前仅在人类和黑猩猩等大脑发达的哺乳动物身上被观察到。卢科拉及其团队训练成对的蜜蜂将乐高积木推到迷你竞技场的中央，或推开隧道尽头的门以获取奖励。研究小组发现，与未经训练的对照组相比，如果同伴也参与其中，蜜蜂更有可能投入到任务中。他们得出结论，蜜蜂可以在蜂巢外学习解决新颖的协作任务，甚至可能是有意地进行合作，尽管研究人员提醒，需要对这种行为进行更详细的监测，才能充分理解同伴各自的角色。

For this latest study, Loukola was interested in whether bees could spontaneously solve problems. The first experiment featured an artificial flower placed above a pit in the floor so that there was insufficient space for a bee to hover to reach the flower. The bee would have to roll a small ball into the pit and climb on top to reach the flower. “This is essentially an insect version of the classic ‘box-and-banana’ problem,” said Loukola. “The animal must realize that an object can be repositioned and then used as a tool to reach an otherwise inaccessible goal.”

在最新的这项研究中，卢科拉关注的是蜜蜂是否能自发地解决问题。第一个实验设置了一朵人造花，放置在地面上的一个坑上方，蜜蜂无法通过悬停来触及花朵。蜜蜂必须将一个小球滚入坑中，并爬到球上才能触及花朵。“这本质上是经典‘箱子与香蕉’问题的昆虫版本，”卢科拉说，“动物必须意识到物体可以被重新定位，并作为工具来达到原本无法触及的目标。”

One set of bees was trained to recognize the flower as a source of sugary reward and that the ball could be moved into the pit, but they were not trained to solve the experimental conundrum. “They only learned the properties of the individual elements and success would therefore reflect spontaneous problem-solving rather than gradual reinforcement learning,” the authors wrote. A second group was trained that the flower was a source of reward but not that the ball was movable. And a third group received no training at all. Bees in the first group solved the problem at a much higher rate than those in the other two groups, whose poorer performances were similar. The first group also made more attempts at working the problem, and the bees interacted with the ball more efficiently and in a more structured way than those in the other two groups.

第一组蜜蜂接受了训练，认识到花朵是糖分奖励的来源，并且球可以被移动到坑里，但它们并未接受过解决该实验难题的训练。“它们只是学习了各个要素的属性，因此成功解决问题反映的是自发的问题解决能力，而非渐进式的强化学习，”作者写道。第二组蜜蜂被训练认识到花朵是奖励来源，但不知道球可以移动。第三组则完全没有接受训练。第一组蜜蜂解决问题的成功率远高于其他两组，而后两组的表现均较差且相似。第一组蜜蜂在尝试解决问题时也表现得更为积极，且与球的互动比其他两组更高效、更有条理。

Those initial results were interesting, but Loukola et al. wanted to rule out the possibility that bees might have an inherent preference for rolling balls, such that perceptual feedback may influence their actions, i.e., rolling the ball might be rewarding on its own. So the team performed a second version of the experiment in which a barrier with a small opening blocked the bees’ view of the flower. The bees had to roll the ball through the opening to climb on top and reach the flower. “This design assessed whether bees could solve the task without continuous perceptual feedback,” the authors wrote. All told, 16 of 22 bees succeed in this task. Granted, the bees could still potentially catch a glimpse of the flower once the ball was near the opening, so the team repeated the experiment with three openings in the barrier to further limit visual feedback. This time, there were no significant differences in performance between trained and untrained (control group) bees.

这些初步结果很有趣，但卢科拉及其团队希望排除蜜蜂可能天生喜欢滚球的可能性，即感知反馈可能会影响它们的行为——也就是说，滚球本身可能就是一种奖励。因此，研究小组进行了第二个版本的实验，设置了一个带有小开口的障碍物，挡住了蜜蜂观察花朵的视线。蜜蜂必须将球滚过开口，才能爬上去触及花朵。“这一设计旨在评估蜜蜂是否能在没有持续感知反馈的情况下解决任务，”作者写道。总共有22只蜜蜂中有16只成功完成了任务。诚然，当球靠近开口时，蜜蜂仍有可能瞥见花朵，因此研究小组在障碍物上设置了三个开口重复了实验，以进一步限制视觉反馈。这一次，受过训练的蜜蜂与未受训练的（对照组）蜜蜂在表现上没有显著差异。

In one last experiment, Loukola et al. sought to isolate the bees’ goal-directed performance from accidental success and from visual feedback cues. This time, the testing apparatus featured a rectangular arena with two compartments, both invisible to the bees. During pretraining, 30 bees were shown the flower positioned above one of those compartments. For the actual test, the flower was not visible from the ball’s starting location, and the bees had to move the ball into the correct compartment. The results: 23 of the 30 bees succeeded at the task, and 16 of the successful 23 bees did so without first moving the ball to the incorrect compartment.

在最后一次实验中，卢科拉及其团队试图将蜜蜂的目标导向行为与偶然成功及视觉反馈线索区分开来。这一次，测试装置是一个矩形竞技场，包含两个蜜蜂无法看到的隔间。在预训练期间，30只蜜蜂被展示了放置在其中一个隔间上方的花朵。在实际测试中，从球的起始位置无法看到花朵，蜜蜂必须将球移动到正确的隔间。结果显示：30只蜜蜂中有23只成功完成了任务，其中16只在没有先将球移到错误隔间的情况下就直接成功了。

The team acknowledged that the experimental setups had no way to track the bees’ gaze, posture, or other behavioral cues that might have let them pinpoint the precise “Eureka!” moment when the bees “understood” the problem. Further experiments should test how well bees grasp causal relationships. “Nonetheless, the present design provides the clearest evidence to date that bumblebees are capable of generating novel, goal-directed solutions, establishing a foundation for future studies to further investigate the cognitive processes underlying insight in insects,” the authors concluded.

研究团队承认，目前的实验设置无法追踪蜜蜂的视线、姿态或其他行为线索，从而无法精准定位蜜蜂“理解”问题的“顿悟”时刻。未来的实验应测试蜜蜂对因果关系的掌握程度。“尽管如此，目前的设计提供了迄今为止最明确的证据，证明大黄蜂有能力产生新颖的、目标导向的解决方案，为未来进一步研究昆虫洞察力背后的认知过程奠定了基础，”作者总结道。