Uncertainty in risky environments: a high-risk phenotype interferes with social learning about risk and safety

• We tested whether uncertainty makes social information more persuasive.
• Uncertainty was induced via incomplete information about predation risk.
• Uncertain and certain background risk regimes elicited a high-risk phenotype.
• Neophobia was culturally transmitted to observers from risk-experienced models.
• Safe-experienced models had little effect on reducing fright of observers.

Uncertainty about risk is a pervasive problem for prey that must continuously manage risk in an ever-changing world. Prey can, however, minimize this uncertainty by learning from the information they sample in their environment. How uncertainty affects learning about risk has been the subject of recent attention, but no studies have examined how uncertainty affects learning about risk (or safety) via social information. Here, we sought to assess whether uncertainty would make social information about risk and safety more persuasive. We induced uncertainty in minnows, Pimephales promelas, by exposing them to a background regime of incomplete information about predation risk in the form of conspecific alarm cues without any information about a specific predator. For other fish, we paired the alarm cues with a novel odour to give minnows the opportunity to be certain about the predator's chemical signature. Then, uncertain and certain minnows were given an opportunity to use social information from live conspecifics (models) that were experienced with the odour as being either risky or safe. Compared to control fish (no background risk), minnows that were both certain and uncertain developed a high-risk phenotype. These individuals spent less time moving and foraging, were more likely to develop a behavioural stereotypy (rapid-loop swimming) and displayed neophobic responses, regardless of their social conditioning. Control fish also developed neophobic responses after interacting with risk-experienced models, presumably because the models' high level of background risk triggered high-risk behaviour that indicated the environment was risky. Thus, whether these observers learned specific information from risk-experienced models or only learned generalized fright remains unknown. In contrast, we found weak evidence that interacting with safe-experienced models can reduce fright in observers to a previously known threat. Alternative approaches to safety conditioning may be more influential.