Honey bees collect their food from different types of flowers, each of which contains different amounts of nectar. The quality of the nectar and the likelihood of finding it at all also vary from flower type to flower type. But bees are very effective at selecting flowers to maximize their yield. So far, however, it was unclear how they would master this complex task with a wide range of options. Now researchers have found that the clever insects use an astonishingly simple tactic.
Although honey bees only have a tiny brain, they have extraordinary cognitive abilities: They live in complex social structures, communicate in a sophisticated way and are extremely capable of learning. According to previous studies, they are even able to distinguish numbers and solve arithmetic problems. The task of choosing the most promising flowers when looking for food is also a challenge that should not be underestimated. How the bees deal with it has now been examined by researchers led by HaDi MaBouDi from the University of Sheffield.
Colored discs with a reward
In addition, they presented bees with colorful discs in five different colors. Depending on the color, the slices often contained a drop of sugar water as a reward or a drop of a bitter liquid that is unpleasant for the foraging bees. In a training phase, the researchers gave the bees the opportunity to get to know the reward probabilities of the respective colors. To do this, they had them explore two colors with different success rates at the same time. How would the bees proceed? One possibility would be to choose only the color with the highest probability of success. For this, the bees would have to compare the colors based on their experience and rank them. The other option would be to choose the most promising color most often, but also to continue to use other colors. In this case, the bees would not need to compare the different colors with one another, but would have to remember the probability of success for each color.
In fact, the results of MaBouDi and his colleagues show that bees follow the second strategy. This applied both to color combinations that the insects already knew from training and to colors that the bees had only experienced in combination with other colors during training, but not with each other. The researchers conclude from this that the bees do not create a mental ranking of the colors, but only evaluate the chance of a reward on the basis of the probability for each individual color. This strategy is considered to be cognitively less complex. However, depending on the context, it can produce better results.
If the probabilities are known in advance and cannot be changed, it is most promising to always choose the type of flower that is most likely to offer a reward. The situation is different if it is only necessary to find out during the collection process which option is particularly successful. This is exactly the case with honey bees. So it makes sense for them to combine collecting particularly promising flowers with researching new sources.
Neural network mimics bee brain
To substantiate their theory, the researchers modeled a computer neural network that mimics the learning center in the bees’ brain. They trained this neural network with the same combination of positive and negative reinforcement that they had previously used in the bees with the help of sugar water and bitter liquid. It should then solve the same tasks.
And indeed: the results were very similar to those of real bees. A ranking of the various stimuli is therefore actually not necessary in order to simulate the behavior of the insects. It is sufficient to consider the respective probabilities in isolation on the basis of reinforcement learning. With very simple cognitive methods, honey bees are able to behave optimally in complex decision-making situations. “Of course, that doesn’t prove that honeybees are not able to set up rankings,” the researchers write in their publication. “The effectiveness of the simple strategy for most foraging tasks makes more complex strategies superfluous in most cases.”
Source: HaDi MaBouDi (University of Sheffield) et al., Proceedings of the Royal Society B, doi: 10.1098 / rspb.2020.1525