Researchers have discovered that bees can do basic mathematics, from understanding the concept of zero, to performing addition and subtraction. This could have surprising implications for future Artificial Intelligence.
The ability to solve a maths problem highlights a sophisticated level of cognition, which involves long-term rules, short-term working memory, and the mental management of numbers. The discovery that even the tiny brain of a bee has this level of cognition has implications for the future of improving rapid learning for Artificial Intelligence.
How were the bees taught maths?
The research trained honeybees by having them visit a Y-shaped maze. The bee would see a set of elements between 1 to 5 shapes. If the shape was blue, the bee was meant to add, and if the shape was yellow, the bee was supposed to subtract. The bee would then fly through a hole to choose the left or the right size of the maze, one had an incorrect solution and the other side a correct solution, or either plus or minus one. The correct answer was changed randomly throughout the experiment.
They were given a reward of sugar water when they made a correct choice, and a quinine solution with a bitter taste if their choice was not correct.
At the beginning, bees made random choices. But they eventually learned to solve the problem over 100 learning trials that took 4 to 7 hours.
They kept returning to the experiment to collect nutrition, because bees go back to the same location if they find a good source of food there.
The cognition involved in mathematical calculations
The Royal Melbourne Institute of Technology (RMIT), Australia’s Associate Professor Adrian Dyer explained that basic mathematical problems such as addition and subtraction are actual complex acts of cognition, because they require two levels of processing.
“You need to be able to hold the rules around adding and subtracting in your long-term memory, while mentally manipulating a set of given numbers in your short-term memory,” he said.
“On top of this, our bees also used their short-term memories to solve arithmetic problems, as they learned to recognise plus or minus as abstract concepts rather than being given visual aids…Our findings suggest that advanced numerical cognition may be found much more widely in nature among non-human animals than previously suspected.”
How could this help the development of AI?
This discovery could enable the development of rapid learning. Dyer added: “If maths doesn’t require a massive brain, there might also be new ways for us to incorporate interactions of both long-term rules and working memory into designs to improve rapid AI learning of new problems.”