Bumblebees socially learn behaviour too complex to innovate alone

The study investigates whether bumblebees can socially learn a multi-step behaviour that they cannot individually innovate, addressing a key criterion often attributed uniquely to human cumulative culture. Prior work shows animals can socially transmit and modify behaviours, but typically these behaviours remain within the bounds of individual innovation. The authors designed a novel two-step puzzle box requiring an unrewarded initial action (pushing a blue tab) before a rewarded action (pushing a red tab) to test whether naive bees can acquire such a behaviour only via social learning from trained demonstrators. The rationale is that the temporal and spatial separation between the initial action and reward would prevent individual associative learning, thereby testing the threshold between individual innovation and socially learned complexity.

The paper reviews evidence for animal culture and potential cumulative cultural processes: long-term changes in bird and whale songs; the progression of sweet-potato washing in Japanese macaques; and diversification in New Caledonian crow pandanus tool designs. Experimental studies include pigeons improving homing routes through sequentially replaced partners, indicating cumulative improvements, though achievable by individuals. Prior bee studies demonstrated social learning of string-pulling and ball-rolling, and maintenance of arbitrary puzzle-box variants via social transmission, yet those tasks were individually innovable. Comparisons are drawn to great tits, which could recombine two previously learned steps to solve a two-step puzzle but did not socially learn the full two-step sequence de novo, suggesting multi-step social learning without direct reward for all steps may be particularly challenging across species.

• Apparatus: A custom two-step puzzle box required first pushing a blue tab to clear the path of a red tab, then pushing the red tab to rotate a lid and expose a sucrose reward (50% w/w) on a yellow target. The design incorporated a tail section and an opaque cover to conceal a temporary training reward under the blue tab during demonstrator training.
• Setting: Experiments were conducted in a flight arena connected to a colony nest box via an acrylic tunnel. Boxes could be removed/replaced via side flaps; bristles prevented escape.
• Control population experiments: Three colonies (C1, C2, C3) were exposed to pre-training with opened boxes for 30 min followed by closed boxes for 3 h per day. Two colonies ran for 12 days (36 h closed-box exposure), and one for 24 days (72 h). No demonstrators were introduced. Behaviours around closed boxes were monitored.
• Demonstrator training: Training took ~2 days per bee (vs several hours for prior single-step tasks). Bees readily learned to push the rewarded red tab but not the unrewarded blue tab. A temporary yellow target and reward were placed under the blue tab initially to establish the first action, then gradually removed while increasing the final reward to maintain motivation. Demonstrators sometimes needed coaxing with fully opened lids. Two opening techniques were observed: (1) staggered-pushing (distinct pushes of blue then red tab, with repositioning), and (2) squeezing (a continuous movement starting at the junction of tabs, squeezing past the shield to push red), often with looping around the red tab.
• Dyad experiments (social learning): 15 dyads (one trained demonstrator, one naive observer) foraged together on three closed boxes per session (each box filled with 20 µl 50% sucrose). Each dyad had 30–40 joint sessions, up to 20 min each (maximum total 800 min, ~13.3 h). Unrewarded learning tests for observers occurred in isolation after 30, 35, and 40 joint sessions. If an observer passed, it received 10 solo foraging sessions (up to 10 min each) with boxes holding 50 µl sucrose per box to assess proficiency (expected 2+ boxes per session based on foraging intake).
• Behavioural quantification: For dyads, the demonstrator’s box-opening incidence and session-normalized opening index were recorded. Observer ‘following’ behaviour was quantified as time spent within one bee length on the box surface while the demonstrator performed box-opening (from start of blue push to reward access). Following duration and a session-normalized following index were computed.
• Statistics: Group comparisons employed unpaired t-tests, one-way ANOVA, Mann–Whitney U, or Kruskal–Wallis tests as appropriate, with 95% CIs and effect sizes (Cohen’s d or η²). Correlations used Spearman’s rank coefficients.

• Individual innovation failed: Across control population experiments (C1, C2: 12 days/36 h; C3: 24 days/72 h of closed box exposure), no bee opened a two-step box; interest in closed boxes declined over time. Only a single full blue-tab push was observed once and not repeated.
• Social learning succeeded: In dyads, 5 of 15 observers (33%) passed unrewarded learning tests after exposure to demonstrators. Despite the task difficulty (unrewarded first step), some observers acquired the full sequence without ever being rewarded for the first step.
• Proficiency: All learners repeated two-step opening in solo sessions; two individuals were classified as proficient learners (≥10 boxes opened) under the study’s criterion.
• Demonstrator technique mattered: Two techniques were used by demonstrators—squeezing (n=10 dyads) and staggered-pushing (n=5). All observers paired with staggered-pushing demonstrators failed to learn (0/5), whereas 5/10 paired with squeezing demonstrators succeeded. Successful observers tended to adopt the squeezing technique.
• Demonstration amount did not explain success: When accounting for session number, there was no significant difference in demonstrator opening index between squeezing vs staggered-pushing demonstrators (unpaired t-test: t = −2.015, P = 0.065, df = 13; 95% CI −3.63 to 0.13). Although all-stagger dyads had higher raw opening incidence (P = 0.026), this did not translate to observer learning.
• Following behaviour: No significant correlation between demonstrator opening index and observer following index (Spearman r = 0.173, P = 0.537). Following indices tended to be higher in learners vs non-learners (34.82 vs 16.26) and with squeezing vs staggered demonstrators (25.78 vs 15.76), though group differences were not statistically significant at α=0.05. Following duration increased with number of joint sessions across groups, strongest in squeezing-pass dyads (r = 0.408, P < 0.001, df = 168); also observed in other groups (e.g., r = 0.275, P = 0.0001, df = 188; r = 0.153, P = 0.036, df = 188).
• Training difficulty underscores complexity: Demonstrators could not learn the unrewarded first step without an initial temporary reward, highlighting the associative challenge posed by temporal/spatial separation of steps.

The findings directly address whether bumblebees can acquire a complex, multi-step behaviour that they cannot individually innovate: observers learned to open a two-step puzzle box only after social exposure to demonstrators, while extensive individual exposure failed to yield innovation. This supports the idea that social learning can enable acquisition of behaviours that exceed individual associative learning capacities, challenging the view that such capability is unique to humans. Mechanistically, the squeezing technique likely reduced the temporal and spatial gap between the initial (unrewarded) action and the reward, facilitating associative linkage; observers that closely followed squeezing demonstrators effectively reproduced the critical action pattern, aiding learning without requiring high-fidelity imitation. Comparisons with great tits underscore the distinct achievement here: bees learned the full two-step sequence de novo via social exposure rather than recombining previously learned steps. These results inform debates on cumulative culture by demonstrating in an invertebrate that social learning can transmit behaviours beyond individual innovation, a key component often associated with human cultural complexity.

This study shows that Bombus terrestris can socially learn a novel, multi-step behaviour that they do not individually innovate, providing, to the authors’ knowledge, the first experimental evidence in a non-human animal for social acquisition of behaviour too complex to be re-innovated individually. Demonstrator technique critically influenced transmission, with the squeezing variant enabling successful learning and subsequent adoption by observers. These insights expand our understanding of social learning capacities in insects and contribute to the broader discourse on the prerequisites of cumulative culture. Future work could: (1) test maintenance and diffusion of two-step box-opening within larger groups over time; (2) explore whether similar multi-step behaviours can be socially learned in other species and contexts; (3) dissect the perceptual and motor features of demonstrator actions that facilitate observer learning; and (4) examine long-lived social insects (e.g., honeybees, tropical bumblebees, stingless bees) for natural instances of complex cultural traditions.

• The study did not assess long-term maintenance or population-level diffusion (cultural tradition) of the two-step behaviour.
• Sample sizes at the dyad level were modest (n=15), limiting statistical power for some comparisons (e.g., following indices).
• Although unlikely, the possibility that a bee could individually re-innovate the two-step solution over a lifetime cannot be entirely excluded.
• Demonstrator training required a temporary reward for the first step, potentially shaping demonstrator technique distribution; some demonstrators were reused across dyads.
• Learners’ solo exposure time was limited compared to prior single-step studies, possibly underestimating post-learning proficiency.
• Findings are from laboratory conditions; ecological relevance and occurrence in the wild remain unknown.