��«Ӱҵ

Honeybees can solve a type of maths test without any need for numbers - a discovery that could be used to develop smarter artificial intelligence - according to new research from the University of ��«Ӱҵ. 

In the study, by researchers in the University’s Department of Computer Science and their international collaborators, published in the journal Proceedings of the Royal Society B, honeybees were found to be using continuous, non-numerical cues to solve a maths problem. 

The insight into how honeybees solve numerical problems could be used to design more sophisticated machines based on the brains of animals, which have evolved to find the simplest, most efficient way to carry out certain tasks.

The methods used in the study could also provide a new, alternative blueprint for testing numerical cognition in animals. Most previous studies attempt to control for at least one non-numerical cue, whereas the ��«Ӱҵ study is the first to propose a new method that carefully assesses non-numerical continuous cues. 

Determining how different brains, especially those with a miniature brain, solve numerical tasks also provides a valuable insight into the evolutionary roots of cognition.

The ability to use numbers is a powerful cognitive capacity of humans, but replicating their full sophistication via AI is not possible with current approaches. With animals often being very good at finding efficient and sometimes unexpected solutions to problems, it is worth finding out which of their shortcuts could be useful for developing autonomous machines 

Through a task that is commonly used to test numerical cognition in bees and a variety of other animals, the ��«Ӱҵ team discovered that honeybees can discriminate between placards displaying different numbers of elements without their brains having to process numerical data.

Honeybees were individually trained to identify placards showing different numbers of shapes. Some bees learned to find a sugary reward at the placards that had the most shapes on display while others learned to find the sugary treat at the placards showing the fewest number of shapes.

Once the bees learned this rule, they were able to quickly identify the placard with the highest or lowest number of shapes on them in order to find the sugary treat.

Since these visual cues are computationally simpler to process by a bee’s brain, it is a more efficient way for bees to use them in solving the task, instead of solving a complex cognitive task of numerosity.

To determine if the bees used non-numerical clues, they were then tested with two pairs of placards that all contained the same number of shapes but differed in edge length, convex hull and spatial frequency. None of the placards had a sugary treat and if the bees used numerosity they should have flown to each placard equally in search of a reward.

The study found that bees trained to find the placards with the highest number of shapes still flew to signs with the highest level of continuous variables and bees that were trained to find the placards with the lowest number shapes still flew to the signs with the lowest level of continuous variables, and ignored numbers. This suggests the honeybees responded to continuous cues on the shapes and not the number of elements.

Dr HaDi MaBouDi, the lead author on the paper who is based at the University of ��«Ӱҵ, said: “The results of our study show that animals are incredibly clever and can solve tasks in effective and unexpected ways. This will be very practical in the future of artificial intelligence for designing smart machines based on animals that have evolved for some particular tasks.”

To further test the hypothesis, the bees were shown placards with visual cues that were opposite to the number of shapes on the placards. For example, placards that had three shapes on display but had a lower spatial frequency than the placards with two shapes.

In these tests the bees behaved in the reverse manner to which we would expect if they had learned numerosity. Bees that were trained to associate a sugary reward with placards with the highest number of shapes still flew to the displays with the highest levels of continuous variables, even though they had fewer shapes. Bees that were trained to associate fewer shapes with a sugary treat were flying to placards with the fewest level of continuous cues even though they displayed the highest number of shapes.

Dr Cwyn Solvi, a co-author at Macquarie University said: “The results of our work suggest that the ability to distinguish items based on continuous magnitudes, i.e. properties such as density or total area of items, might be more basic and perhaps evolutionarily older than the ability to process numbers.”

To determine how efficient this non-numeric strategy was for bees, the researchers also created a model based on known details of the honeybee brain. This bee brain model was able to solve several other similar numeric-based tasks without any need for number processing - similar to the real honeybees. 

Professor James Marshall from the University of ��«Ӱҵ, who leads the Brains on Board research project that is seeking to reverse-engineer honeybee brains to develop new approaches to artificial intelligence, said: “Showing that animals are clever enough to do things we assume only humans can do always generates a lot of interest, but it is worth bearing in mind that nature usually finds the simplest possible solution to a problem.”

Dr MaBouDi cautions: “This doesn’t mean that bees or other non-verbal animals can’t understand numbers, but it does suggest that animals use non-numeric properties to solve the math problems they often face if such information is available. However, we hope that our study provides insight into better methods of exploring mathematical cognition in animals.” 

The research paper, Non-numerical strategies used by bees to solve numerical cognition tasks, is published in the journal Proceedings of the Royal Society B. To view the paper, visit:  

The paper is authored by HaDi MaBouDi, Andrew B Barron, Sun Li, Maria Honkanen, Olli Loukola, Fei Peng, Wenfeng Li, James A. R. Marshall, Alex Cope, Eleni Vasilaki, and Cwyn Solvi. 

Contact

For further information please contact: