Man’s best friend: artificial intelligence can think like a dog

39

Лучший друг человека: искусственный интеллект сможет думать как собакаTests showed that the AI is able to learn from animals.

Researchers from the University of Washington and the Institute of artificial intelligence Allen trained neural network to interpret and predict the behavior of dogs. Scientists came to the conclusion that animals can become a new source of data for training artificial intelligence systems, including those used in robots.

Recently they were able to teach the neural network to understand and predict the behavior of dogs. Meanwhile, the scientists found that the behavior of animals may become a new valuable resource of knowledge for artificial intelligence — including the one used in the robots.

To teach the AI to think like a dog, the researchers had to collect data. For this they used motion capture and videos with one dog, a Malamute named Kelp. Was recorded 380 videos. For their records on the head of a dog attached GoPro, and the legs and body — motion sensors. Scientists have recorded his daily actions: running, games with ball and a walk in the Park.

Armed with the information collected, the researchers analyzed the behavior of Kelp with deep learning — the technology to identify certain patterns. In this case, the algorithms for correlated data about the movement of body parts of Kelp and the camera image with different types of activity. In the end, the neural network learned to predict how it will behave with the dog in different situations. For example, if someone throws a ball, the system will know that, most likely, the dog will chase him.

Study leader Qian Ehsani explained that AI can predict behavior very accurately, but its capabilities are limited. For example, if the video stream will show the stairs, the system will guess that the dog can use it to climb. But in General, life is too varied to predict every reaction.

Teaching the AI to analyze a dog’s behavior, scientists have wondered whether the algorithm can learn from these animals something else. According to researchers, dogs “have excellent visual perception — they know how to recognize good food, obstacles, as well as people and other animals.”

It turned out that the neural network can learn from dogs this trait, though in a limited form. The researchers conducted two tests in which AI had to recognize the environment where it is (inside, outside, on the stairs or on the balcony) and define the surface on which you can walk. In both cases, the neural network coped with the task with good accuracy using the simple knowledge gained from the study of dog behavior.

The system needs to understand which surface is comfortable to walk and what not, and the dog immediately is the realization of all necessary factors. After studying the behavior of Kelp, a neural network learn these rules yourself.

Of course, it should be noted that the results of the study are not revolutionary, what might seem. Ehsani and her colleagues are unable to develop a counterpart of the brain or mind of the dog. AI has just learned the basic rules of conduct, based on limited data about where he likes to walk the dog. The system does not understand the reasons for the behavior, it just says templates.

But still this is the first time when artificial intelligence has learned anything from the dog. And the results of the experiment show that the knowledge of animals can be useful in learning algorithms. In the end, dogs are aware of the many things that will be useful to robots. For example, they understand how people look and can distinguish the adult from the child. Dogs know not to get under wheels of cars and to focus on the street — these skills will be useful robots working among the people.

Ehsani sure that the results of her research can be found a useful application. “First and foremost, I think about creating a dog robot, she said. Robots hardly understand how to move, where to go and how to behave if you need something to chase. I think our research will help to create the best version of dog-robot”.

LEAVE A REPLY

Please enter your comment!
Please enter your name here