If you can really let the robot such as hot drama"Western world"
AI Google program with reasoning ability
According to the British Guardian reported that Google scientists have recently developed a new artificial intelligence system in the experiment, it can use simple reasoning ability to conduct autonomous navigation of the subway line.
Navigation tasks, a simple app can be completed, but the reasoning? Maybe only primates can do it.
Bremen university computer scientist Herbert Jaegar said so:
Although the task itself seems relatively simple, after all, any one of the smart phone app can easily be in accordance with the subway map planning route. However, the way to get the results of this program is impressive. In fact, this is the world's first external memory devices (memory) and the depth of learning technology is one of the integration of the program. On the basis of the above, the system can learn how to finish the task independently without being set up by human beings.
The combination of neural network and deep learning
Although the ordinary computer can handle all kinds of complex forms of data, but it requires the developer to manually programming to perform these tasks. At the same time, the artificial neural network can learn from the data, although it can be like the brain, but its lack of storage structure can not be used to deal with structured data.
This is a foreign student who has been baffled taste in Chinese logic contest. Also hinted that the future of small assistants like Apple's Siri may be replaced by a more sophisticated Google virtual assistant.
Similarly, Graves also stressed that although this small task on the story for humans is simply insignificant, and even a little is not a problem, but the basic computer programs do not do this. And Mind Deep development of this new program in answer to the question of the accuracy rate is as high as 96%.
Does the robot with the memory become a new development trend of AI?
However, until now, a single application technology seems good, but if a task requires a kind of integrated solutions, the overall performance of these techniques is still unsatisfactory (such as navigation, such as unmanned).
In fact, this is a short-term resource library of a human brain in a working memory. When we are going to do a task, we will follow the previous experience (memory) to list the steps to do this. Of course, you can also take it as a recipe.
Jaegar said that from the recent AI R & D heat point of view, these are very cost DeepMind results should be regarded as a catalyst or a crucial straw, will lead to more follow-up effect. He stressed that the future of the computer's functions will no longer be bound by those more logical task, people may soon see it in the field of navigation and other widely used:
However, some scientists with sharp positions are always reminding us of the existence of the threat of AI. For example, the famous physicist Hocking even warned:
But DeepMind's founder, Hassabis Demis, last year had a deliberate weakening of the concerns of scientists:
I am 36Kr news group author Fu Bo (Yu Duotian), WeChat fudabo001