Welcome back to Science Sunday
Four legs, a tail, must be a dog. No wait maybe it's a cat or ferret. Dogs do not seem to have any problem recognizing members of their own species, notwithstanding the myriad of shapes and sizes that they come in. We know dogs are renowned for their olfactory competence, so is that how they do it? Anecdotal experience suggest that they have already made the species identification before the serious butt sniffing begins. So what cues are they using to identify a stranger as a dog or not a dog. For that matter how do we do it? Small children can discern between a dog and a cat even among broad morphological samples. What differentiates "catness" from "dogness". So are dogs able recognize their con-specifics solely by sight?
A team of researchers based in France took on this question, publishing their findings in Animal Cognition in 2013. Nine dogs were used in the study, a Lab, a Border Collie and seven mixed breeds, all living in homes and normally trained as pet dogs.
The dogs first went through a training phase where they were shown two pictures, one of a dog (always the same dog) the other screen was either all black, all blue, or had a picture of a cow’s face. The dogs were rewarded for selecting the picture of the dog by approaching the screen. All nine subjects learned to do this in three sessions.
Then came the test. Dogs were presented with a wide variety of never-before-seen dog faces paired against never-before-seen non-dog faces. As before, dogs had to approach the dog image and avoid the non-dog image to get a treat. The none dog faces included a wide variety of domestic and wild animals and even humans. All nine dogs in the study were able to group all the dog images, regardless of breed, into into a single category despite the diversity of breeds. We still do not know how they do this, that is to say what is the "dogness" of a dog that makes it recognizable by dogs or humans.
Recent developments in Artificial Intelligence (AI) have brought computers to a similar place in recognition. As I understand it (which is not very well), convolutional neural networks are shown huge samples of dog pictures tagged as dogs. The AI system learns to identify dogs from this process, the accuracy dependent on the size of the sample. Interestingly again, we do not understand what the system is identifying to discern "dogness" . Even more impressive these systems can now identify dog breeds.
So whether its dogs, computers or kids there's something about dogs that render them recognizable.