Domestic Dogs' understanding of the accuracy of human social informants

• In Event: Poster Session 09
  In Poster Session: PS 09 Topic - Social Cognition

Abstract

Dogs exhibit an excellent understanding of human social cues, highlighted in dogs’ ability to learn from human communicative gestures such as pointing. Previous studies have shown that dogs trust selectively, distinguishing between human social informants who vary in characteristics such as knowledge or familiarity. Over 3 experiments, this study explores if dogs, like human children, can use human social informants’ past accuracy when deciding whom to trust. Experiment 1 investigated dogs’ ability to integrate their own observations about treat location with accurate or inaccurate social cues, using an object choice task. This experiment aimed to clarify previous research, which has resulted in differing conclusions about dogs’ behaviour in the presence of an inaccurate informant. In Experiment 1, dogs (N = 32) observed a person hide a treat in one of two locations, and then consistently point at the correct (accurate condition) or incorrect (inaccurate condition) location. Overall, dogs performed differently between the accurate and inaccurate informants in Experiment 1, t(17.57) = 5.16, p < .001. Dogs followed an accurate informant’s point significantly above chance (M = 11.63/12, SE = .20), t(15) = 27.91, p < .001. When presented with an inaccurate point, dogs were more likely to ignore it and choose the correct location (M = 7.94/12, SE = .69), t(15) = 2.82, p = .013, though they were less successful at finding the hidden treat than in the accurate condition.
Experiment 2 explored if dogs could selectively follow a previously accurate pointer, over a previously inaccurate one. Dogs (N = 16) experienced both an accurate and an inaccurate informant separately, followed by test trials where both informants simultaneously pointed at different locations (only one of which contained a treat). As in Experiment 1, dogs distinguished between accurate and inaccurate pointers when meeting them separately (M = 5.69, SE = .68), t(15) = 5.51, p < .001, but in the test phase, where dogs viewed conflicting points, dogs chose between the two informants at chance (M = 6.31, SE = .68), t(15) = .36, p = .72.
Experiment 3 aimed to see if dogs could succeed at using past informant accuracy as a choice metric if they were scaffolded on the non-social aspects of the task that may have caused dogs’ poor performance in Experiment 2. In particular we exposed dogs to potentially difficult non-social problems like hidden baiting, occlusion events prior to the critical test trials, and reduced the cognitive leap between informant history (where dogs could see the treat) and test (where dogs had no information about the treat) by having the informants point simultaneously throughout. Over two visits, dogs (N = 19) completed a-social scaffolding tasks (visit 1) and met both informants simultaneously (visit 2, see pre-registration for more details). In test trials, dogs chose to follow the accurate (vs. inaccurate) informant (M = 7, SE = .34), t(18) = 2.92, p = .009. This suggests that like children, dogs may be able to use informants’ past accuracy when choosing between information sources.

Experiment 3 pre-registration link: https://osf.io/vaq8w/?view_only=ab918173d74f4cdb8f989784197ad10c

Authors

• Madeline Helmer Pelgrim, Brown University
  Presenting Author

• Julia Espinosa, University of Toronto
  Non-Presenting Author

• Emma Tecwyn, University of Toronto
  Non-Presenting Author

• Sarah MacKay Marton, University of Toronto
  Non-Presenting Author

• Angie Johnston, Boston College
  Non-Presenting Author

• Daphna Buchsbaum, Brown University
  Non-Presenting Author