Honeybees cheat on their 'math tests' — here’s how

Cheating on a math test is a rite of passage for some. It turns out, other animals also use shortcuts to help them problem solve.

A new study published in the journal Proceedings of the Royal Society B found that while honeybees (Apis mellifera) are capable of using numerical strategies on cognitive tests, they actually often deploy continuous (non-numerical) clues to "cheat" on these tests.

"Our study provides information on which cues and strategies bees used to solve a numeric-based task," co-author Olli Loukola, a behavioral ecologist at the University of Oulu, tells Inverse.

Some background — Researchers set out with one question: Do honeybees actually use "numerosity" or number processing to solve problems? Or do they use continuous (visual) shortcuts to complete tasks?

"Using continuous cues, such as density or total area of items, might be more basic and perhaps evolutionarily older than the ability to process numbers," Loukola says.

Moreso than numerical cues, these visual shortcuts can help bees survive in the real world.

"Bees can use landmarks in their spatial navigation. They probably use continuous cues, such as height or width of the landmarks [trees, rocks, and bushes] rather than discrete cues [the number of trees]," Loukola explains.

This question has stumped researchers designing studies in this field since both cues overlap, and it's hard to know when bees are using numerical or continuous cues.

According to the study: "This covariation makes it difficult to know whether animals actually used any sense of number to solve their tasks."

How they did it — The scientists presented a series of visual displays on acrylic paper to the honeybees with the goal of assessing whether — and how — they can distinguish shapes based on their relative size.

They trained one group of honeybees to associate a positive reward — a sweet sugar-based solution — with displays containing more shapes. Conversely, they trained the bees to associate a negative prize — an unpleasant quinine solution — with displays containing fewer shapes.

For comparison, the scientists taught another control group the exact opposite training.

Once the bees had selected the "correct" shape with 80 percent accuracy, they were moved onto the actual tests, which contained no reward.

The first test showed that honeybees were able to correctly distinguish shapes based on their training.

The second test asked whether honeybees used numerical or continuous cues. To test their hypothesis, researchers provided two pairs of displays that contained the same number of shapes or "elements."

If the honeybees were using numerical cues to solve the problem, they should have favored all displays equally.

Spoiler alert: That's not what happened.

What they found — The bees trained to favor displays with a greater number of shapes wound up choosing displays containing more continuous variables — and vice versa.

According to the study, "this suggests honey-bees responded to continuous cues in the stimuli and not the number of elements."

The researchers further confirmed the honeybee's preference for continuous cues by placing these visual shortcuts in direct opposition to the numerical cues.

The insects behaved in the "reverse manner to which we would expect if they had learned numerosity," the research team writes.

Why it matters — The study's findings have widespread implications for understanding the evolution of certain cognitive skills — specifically related to processing numbers — across the animal kingdom.

"This [study] will provide vital information for how, when, and why numerical cognition may have evolved, and how processing of numerosity compares between animals," Loukola says.

This study also offers a stepping stone to better understanding how animals generally solve problems, according to Loukola.

"To better understand how animals solve problems in general, I would challenge them in non-natural tasks," he explains.

These non-natural tasks could entail challenging bees to more cognitively complex activities, such as rolling a ball or pulling a string.

"Solving a non-natural task that is relatively far removed from animals’ natural daily activities shows cognitive flexibility," Loukola says.

What's next — Researchers might have figured out how honeybees trick them in this one experiment, but what happens when future studies want to capture true numerical learning in honeybees — or any other animal? How can scientists ensure that animals won't bypass their control methods using visual shortcuts?

Based on their experience with the honeybees, this team has a few tips: change up the kinds of cues used during animal training and make sure that you use all of these cues in one single, unrewarded test.

"In numerical cognition studies, we should not only use visual cues but instead combine different sensory modalities [visual, auditory, and vibrational cues] to be sure an animal has a sense of number," Loukola says.

Artificial intelligence, such as machine learning technology, may also naturally pave the way to improved research.

Ultimately, the study suggests "automated approaches combining machine vision and learning with computational behavioral analyses have the ability to discover behavioral features that humans cannot."

Abstract: We examined how bees solve a visual discrimination task that uses stimuli commonly found in numerical cognition studies. Bees performed well on the task, but additional tests showed that they had learned continuous(non-numerical) cues. A network model using biologically plausible visual feature filtering and a simple associative rule was capable of learning the task using only continuous cues inherent in the training stimuli, with no numerical processing. This model was also able to reproduce behaviours that have been considered in other studies indicative of numerical cognition. Our results support the idea that a sense of magnitude may be more primitive and basic than a sense of number. Our findings highlight how problematic inadvertent continuous cues can be for studies of numerical cognition. This remains a deep issue within the field that requires increased vigilance and cleverness from the experimenter. We suggest ways of better assessing numerical cognition in non-speaking animals, including assessing the use of all alternative cues in one test, using cross-modal cues, analysing behavioural responses to detect underlying strategies and finding the neural substrate.