Simply by watching other bees, bumblebees can learn to use a novel tool to obtain a reward, a new study reveals. The results demonstrate the capacity of bees to learn how to solve a complex, goal-directed problem, and even improve upon the original process -- a capability long-known possible in humans, primates, marine mammals and birds, and here extended to invertebrates. Olli J. Loukola and colleagues placed bumblebees on a platform and trained them to move a yellow ball from the edge of the platform to the center, in order to access a sugar solution as a reward. Plastic bees were used to demonstrate the behavior. Once trained, all bees successfully moved their balls to the center of the platform to access the sugar. Next, three groups of bees were exposed to different scenarios. In a "social demonstration" scenario, trained bees moved the ball in front of untrained bees. In a "ghost demonstration," a magnet was used to move the ball. For the third group, a ball was already placed in the center. During training for these first two groups, three balls were present and the two closest balls were glued in place, so the farthest ball had to be moved to gain the reward. When the three groups were tested, the observer bees in the social group were more successful and took less time to solve the task than did the ghost demonstration group, and the ghost demonstration group was on average more successful than the no demonstration group. These results suggest that observation of a moving ball was enough for bees to solve the task. What's more, the observer bees went directly for the closest balls - or, in another experiment, a differently colored ball - indicating that they did not simply copy the behavior of the demonstrator but rather improved on the observed behavior by using a better route.