A numerical bias in honeybees: Numerousness is more salient than space and size non-numerical cues during quantity discrimination.

This study demonstrates that free-flying honeybees exhibit a spontaneous numerical bias, preferring to rely on item count over non-numerical visual cues like area and space when discriminating quantities, while also revealing individual variation in strategy and evidence of an oriented mental number line.

The Gist

Imagine you are a honeybee flying out to find flowers. You see two patches of blooms: one has a few huge, bright flowers, and the other has a swarm of tiny, less obvious ones. Which one do you choose? Do you count the flowers, or do you just look at which patch looks "bigger" overall?

For a long time, scientists thought animals might just use the "bigger looks" trick (total area or space) because it's easier for the brain. But this new study suggests that honeybees are actually surprisingly good at counting, even when it's harder to do so.

Here is the story of the research, broken down simply:

The Big Question: Counting vs. Guessing

Scientists wanted to know: Do bees actually "count" numbers, or do they just cheat by looking at size and space?

In the wild, numbers and size usually go hand-in-hand. If you have four big rocks, they take up more space than two big rocks. It's hard to tell if an animal is counting the rocks or just noticing the big pile of space they occupy.

To solve this, the researchers set up a "brain trap" for the bees. They trained the bees to pick a specific number of dots (either 2 or 4) to get a sugary treat. But they did it in two tricky ways:

1. The Size Game: They made the 4-dot group take up twice as much space as the 2-dot group.
2. The Space Game: They made the 4-dot group spread out over a larger area than the 2-dot group.

In both cases, the bees could have just learned: "Pick the bigger shape" or "Pick the wider shape" to get the sugar. They didn't need to count.

The Test: The Magic Switch

Once the bees learned the rule, the scientists pulled the rug out from under them. They showed the bees new pictures where the "cheating" tricks didn't work anymore.

• The Test: They showed the bees a choice between 2 dots and 4 dots, but this time, both groups were the exact same size.
• The Result: The bees still picked the right number! They ignored the fact that the shapes were the same size and went for the correct count.

This is like teaching a child to pick the "bigger" pile of cookies, and then giving them two piles that are the same size but have different numbers of cookies. If the child still picks the pile with more cookies, they aren't just looking at size; they are actually counting.

The Plot Twist: Not All Bees Are the Same

Here is where it gets really interesting. When the scientists looked at the bees individually, they found two distinct "personalities" in how they solved the puzzle:

1. The "Math Whizzes" (Numerical Bias): About half the bees were like strict accountants. They learned the number, ignored the size, and stuck to the count even when the size changed. If the numbers were right, they got the sugar. If the numbers were wrong, they got a bitter taste (a punishment). They didn't care about the size at all.
2. The "Generalists" (The Cheaters): The other half of the bees were flexible. They learned both the number and the size. When the numbers and size agreed, they did great. But when the scientists tricked them (making the size wrong but the number right), these bees got confused. They tended to go with the size because it was the "easier" clue they had learned first.

It's like a classroom where some students memorize the formula (the number), while others just guess based on how the problem looks (the size).

The "Mental Number Line"

The study also found something cool about how the bees see numbers. It turns out, bees seem to have a mental map where small numbers are on the left and big numbers are on the right.

When the "smaller" number (2) was on the left and the "bigger" number (4) was on the right, the bees were faster and more accurate. It's as if they have an internal ruler stretching from left to right, just like humans do when we think about numbers!

Why Does This Matter?

This study is a big deal because bees have tiny brains—less than one million neurons. Humans have billions. For a long time, we thought only big-brained animals (like monkeys or humans) could truly understand "numbers" as an abstract concept.

This research shows that you don't need a giant brain to count. Even with a brain the size of a grain of rice, bees can separate "how many" from "how big." It suggests that the ability to understand numbers might be a fundamental tool that evolved independently in very different animals to help them survive.

In a nutshell: Honeybees are smarter than we thought. They don't just look at the size of a group; they actually count. And just like in a human classroom, some bees are natural mathematicians, while others prefer to take the easy way out!

Technical Summary

1. Problem Statement

While numerous studies confirm that animals, including invertebrates like honeybees, can discriminate quantities, a critical debate remains regarding the ecological relevance and cognitive mechanism of this ability.

• The Core Question: Do animals rely on abstract numerosity (the approximate number of items) or on computationally cheaper non-numerical cues (such as total area, perimeter, density, or convex hull) that naturally covary with number in the wild?
• The Gap: Previous studies often failed to distinguish between a "numerical bias" (a preference for number over other cues) and simple reliance on correlated non-numerical features. Furthermore, most research on numerical bias has been limited to vertebrates (primates, birds, fish), leaving the existence of such a bias in insects with miniature brains (<1 million neurons) unproven.
• Hypothesis: The authors hypothesize that honeybees possess a spontaneous numerical bias, prioritizing numerousness over Size (total area/perimeter) and Space (convex hull/density) cues, even when non-numerical cues are predictive.

2. Methodology

The study employed two complementary experiments using free-flying Apis mellifera (honeybees) in an appetitive-aversive differential conditioning paradigm within a Y-maze.

Experimental Design

• Task: Bees were trained to discriminate between 2 vs. 4 black dots.
• Conditioning: Correct choices yielded a 50% sucrose solution; incorrect choices yielded a bitter 60 mM quinine solution.
• Training Phase: 40 trials per bee. In both experiments, numerosity was perfectly correlated with a specific non-numerical cue during training, but other cues were controlled or randomized.
  • Experiment Size: Numerosity covaried with Size cues (Total Area and Total Perimeter). Space cues (convex hull, density, inter-distance) were partially controlled.
  • Experiment Space: Numerosity covaried with Space cues (Convex Hull). Size cues (Total Area, Perimeter, Item Size) were controlled.
• Testing Phase: Unreinforced transfer tests were conducted to dissociate cues:
  1. Learning Tests: Novel stimuli to ensure bees learned the rule, not just specific patterns.
  2. Numerical Transfer Test (TN): Non-numerical cues (e.g., Total Area in Exp Size) were homogenized (equalized), leaving only numerosity to distinguish the stimuli.
  3. Non-Numerical Transfer Test (TS/TSp): Numerosity was homogenized (3 dots in both stimuli), leaving only the non-numerical cue (e.g., Total Area) to distinguish them.
  4. Conflict Test (CNS): Numerosity and the non-numerical cue were pitted against each other (e.g., 2 dots with large area vs. 4 dots with small area).

Statistical Analysis

• Models: Generalized Linear Mixed Models (GLMMs) with binomial error distribution were used to analyze choice data.
• Clustering: Hierarchical clustering was applied to identify subgroups of bees based on their performance across transfer and conflict tests.
• Mental Number Line (MNL): Interaction effects between the rewarded numerosity group (2 vs. 4) and the spatial position (Left vs. Right) were analyzed to detect spatial-numerical associations.

3. Key Results

A. Spontaneous Encoding of Numerosity

• Numerical Transfer: In both experiments, bees performed significantly above chance when only numerosity varied (and non-numerical cues were equalized). This indicates they spontaneously encoded the number of items during training.
• Non-Numerical Transfer: When numerosity was equalized and only non-numerical cues varied, bees performed at chance level. This demonstrates that they did not rely primarily on Size or Space cues during training.

B. The "Numerical Bias"

• Bees favored numerousness over non-numerical cues. In the Conflict Test, the population-level performance hovered around chance, masking individual strategies. However, the data confirmed that bees could extract numerical information even when it conflicted with strong non-numerical signals.

C. Inter-Individual Variability (Two Distinct Strategies)

Cluster analysis revealed two stable strategies within the population:

1. The "Numerical Bias" Strategy (~50% of bees): These bees relied exclusively on numerosity. They succeeded in numerical transfer and conflict tests but dropped below chance in non-numerical transfer tests (suggesting they ignored the non-numerical cue entirely, even when it was the only available information).
2. The "Generalist" Strategy (~50% of bees): These bees encoded multiple cues. They performed well in both numerical and non-numerical transfer tests. However, in the conflict test, they tended to follow the non-numerical cue, suggesting a flexible switching mechanism where non-numerical cues dominate when cues conflict.

D. Mental Number Line (MNL) Effect

• A significant left-to-right spatial-numerical association was observed. Bees performed better when the smaller quantity (2) appeared on the left and the larger quantity (4) on the right. This effect was robust in Experiment Size and suggests an oriented mental number line in honeybees, similar to findings in vertebrates.

E. Learning Dynamics

• Bees initially relied on simple spatial heuristics (e.g., repeating the side of the previous reward or alternating sides) before shifting to stimulus-based rule learning in the final training blocks.

4. Key Contributions

1. First Evidence of Numerical Bias in Insects: The study provides the first rigorous evidence that honeybees exhibit a spontaneous numerical bias, prioritizing abstract numerosity over Size and Space cues, challenging the "last-resort" hypothesis (that animals only use number when other cues fail).
2. Inter-Individual Variability: It highlights that population-level averages can obscure distinct cognitive strategies. The coexistence of "specialist" (numerical) and "generalist" (multi-cue) strategies within a single species offers a new framework for understanding cognitive diversity in insects.
3. Convergent Evolution: The findings suggest that complex quantity evaluation mechanisms (like numerical bias and MNL) may have evolved independently in distantly related taxa (bees and primates), implying that such cognitive traits are not exclusive to large-brained vertebrates.
4. Methodological Refinement: The study refines the protocol for testing numerical bias by explicitly modeling the relative contributions of cues and using conflict tests to dissociate learned strategies.

5. Significance

• Cognitive Evolution: The results challenge the assumption that abstract numerical cognition requires a complex neocortex. The honeybee brain, with fewer than one million neurons, is capable of spontaneous numerical abstraction, suggesting that similar computational solutions for quantity evaluation can arise via convergent evolution.
• Ecological Relevance: The existence of a numerical bias suggests that foraging bees may prioritize counting discrete items (e.g., flowers, landmarks) over estimating continuous area, which has implications for understanding their foraging efficiency and navigation.
• Future Directions: The identification of distinct cognitive strategies (numerical vs. generalist) opens new avenues for research into the neural mechanisms of inhibition and latent inhibition in insects, potentially linking specific behavioral roles (e.g., scouts vs. recruits) to cognitive profiles.

In conclusion, this paper demonstrates that honeybees possess a robust, spontaneous numerical bias, utilizing an abstract representation of quantity that often overrides simpler perceptual cues, and that this ability is supported by diverse individual strategies within the population.