AI system predicts dog behavior by thinking like a dog
MANILA, Philippines - Over the years, artificial intelligence (AI) systems were trained to perform a host of tasks such as identifying objects and recognizing faces but have yet to reach the complexity to simulate the highly intelligent beings humanity sees as their best friend, dogs. This all changes thanks to a new study being worked on by Kiana Ehsani and colleagues from the University of Washington in Seattle, who are training AI systems to interpret and predict a dog’s behavior.
To do this, the researchers first collected data in the form of videos and motion information captured from a single dog. The dog went about its daily life with a GoPro camera equipped on its head and sensors on its leg and body. The Verge likened the setup to the motion capture technology currently being utilized by popular games and films that record actors playing computer-generated characters.
The researchers then used deep learning, a subset of AI that can identify patterns from data, to analyze the dog’s behavior. It matched the motion data from the limbs and the visual data from the captured video with various canine activities. This enabled the computer system to predict what a dog would do in specific situations. For instance, if a person asked the dog to play fetch and threw an object, the AI would know that the reaction of the dog would be to chase it.
Interestingly, the researchers found that the AI in a limited capacity picked up subtle things about the world without it being programmed to do so. The researchers tests yielded results that suggest the AI is capable of differentiating indoors from outdoors and recognizes places that can be walked on.
“Our intuition for this was that dogs are really good at finding where to walk – where they’re allowed to go and where they’re not,” Ehsani told The Verge. “This is a very hard task for a computer because it requires a lot of prior knowledge. It would tedious to program an AI with this knowledge, but a dog already knows them all. The AI essentially learned all this just by watching the dog’s behavior.
It’s important to note that the neural network or AI software created by Ehsani and her colleagues is in no way a model of a dog’s brain or consciousness.
The study itself believes that life is too varied for the AI to predict everything accurately. All the AI does is learn sets of rules by finding patterns in the given data, which is not new.
What’s new is that this may be the first time an AI is learning from a dog, opening up possibilities for AI systems to even use other highly-intelligent animals as sources for training data. An AI could learn intuitive things like avoiding cars and navigating stairs from animals like dogs for it to function better in a human environment.
One immediate thing that comes to my mind is making a robot dog. It’s already a hard task for a robot, to know how to move and where to go, or if they want to chase something,” Ehsani says. “This this would definitely help us building a more efficient and better robot dog.” – Rappler.com