Humboldtians in Focus
Through the Eyes of a Bee
Interview with Adrian Dyer
A bee’s brain is about ten thousand times smaller than that of a human being and still accomplishes amazing things. The Australian bee researcher Adrian Dyer investigates how bees learn complex tasks and can even recognise human faces. The processes going on in the bee’s brain could become the model for computer systems for facial recognition, at airports for example.
Kosmos: What do researchers find so interesting about a bee’s brain?
Dyer: The bee brain can learn tasks like navigating mazes, which were previously thought to require a much larger brain. Although bees are phylogenetically distant and shared the last common ancestor with humans about 600 million years ago, the publication of the bee genome two years ago in “Nature” suggests some interesting commonalities as to how ‘brains’ might have evolved to solve problems.
|Computer simulation: How bees perceive
A - Flower seen through the human eye.
B - Bees can see ultra-violet rays.
C - Bees have different trichromatic
colour vision. Image seen through
a bee’s simulated compound eye.
D - A bee’s hypothetical colour perception.
Bees have worse eye-sight than humans
Kosmos: Actually one would expect it to be comparatively difficult to train a bee and make it learn a task …
Dyer: On the contrary. Bees are an ideal model because they are altruistic and collect food for distribution to the entire colony. This means that it is possible to train an individual bee to do a complex task for seven to eight hours a day by rewarding it for correct decisions with a small drink of sucrose solution. When the bee’s stomach is full, it simply returns to its hive and then, two minutes later it is back to participate in the experiment again. So it is possible to condition the bee for a long time and collect sufficient data to understand how its brain is solving complex visual tasks.
Kosmos: You trained honeybees using black-and-white photographs of human faces and showed that their miniature brain can learn this task. But does this mean that they could also recognise a three-dimensional face on a living person with this training?
Dyer: What we have recently observed in honeybees is an amazing capacity to learn complex novel tasks and use tricks like image interpolation (assessing a novel representation from previously learnt representations) to solve visual problems like recognising three-dimensional faces. This is not to say that all bees can do this; actually, the exciting point is the opposite; individual bees have the neural flexibility to learn how to solve visual problems if they are provided with the correct conditioning. It teaches us a lot about how brains learn and neural flexibility.
Kosmos: Previous research did not observe such fine capabilities. Why is it that we are now realising what miniature brains can do?
Dyer: It has been a very stepwise progression. Initially, I think people thought a tiny brain has to be simple and possesses only hardwired solutions for solving tasks, but over the past couple of decades, a number of studies from different groups have progressively been showing that bees have a remarkable ability to learn novel tasks, depending upon the conditioning regimes used.
Kosmos: How might these biological solutions be useful for technological improvements in face recognition applications, for example?
Dyer: There have been a lot of difficulties in producing algorithms that can reliably recognise faces when there is a change in viewpoint. For example, a field test of an artificial intelligence face recognition system at a major airport in the United States produced so many false positives that the system was practically not useable in crowded situations – which is when it is needed to work best. Knowing how biological systems cope with these types of visual challenges can potentially provide some insights for software developers about new or novel types of approach. The central idea relates back to the miniature brain possibly having very efficient solutions that are easier to model than biological solutions we might derive from primate brains which are amazingly complex to understand and model.
Kosmos: Bees don’t see the world like we do. Most humans can’t imagine how they would perceive things if they had compound eyes instead of two mammalian eyes. You have worked on simulations of insect vision. How do you turn your knowledge into photographs which give humans an impression?
Dyer: It is possible to make inferences about how bees are likely to view their world by carefully conducting experiments and inspecting the physiological and psychophysics data. We know that bees can see ultraviolet wavelengths of light. Many flowers have evolved flower patterns that suit the visual capabilities of bees, for example, some flowers have patterns that can only be seen with ultraviolet sensitive vision. We also know that bees have a different trichromatic visual system than humans have, and we know that their acuity is considerably poorer than human vision. This makes it possible to produce optics and/or computer simulations of how bees are likely to visually interpret their world.
Kosmos: Why did you choose Germany for your collaboration?
Dyer: I enjoy working on bee science in Germany because the country has such a strong background in this field – since the times of Nobel laureate Karl von Frisch. In Australia, at least in Melbourne, research on bee vision sometimes seems to be regarded as a bit of a novelty, which is a little surprising as we are fortunate in Australia to have a number of strong groups working on the topic.
Comment on article
If you are an Humboldtian and have logged in, you have the option of commenting on this article or other Humboldtians' comments. (Please read the comment guidelines first)
After logging in, Humboldtians have the option of participating in discussion of articles in Humboldt Kosmos and contributing comments of up to 1,000 characters for publication in German or English. If the comment is published it will appear under your name.
Every comment will be checked by the editors and published as soon as possible unless there are objections on legal or content grounds. The editors reserve the right to abridge and revise comments where necessary. Please bear in mind that published comments can be accessed by anyone on the Internet and may be located by search engines.