Wunda: There's no need to worry about robots controlling the world

January 22 News, yesterday, the future forum of the founding of the conference was successfully held in Beijing. Baidu chief scientist Wunda published a keynote speech titled "Artificial Intelligence and deep learning" at the forum, and received an interview with Sina technology after the meeting.

"We recently decided to look at the neural network instead of the traditional speech recognition software," Wunda said. The world moves toward mobile devices, and imagine that we are redesigning mobile products around the voice interface--and that voice will drive the phone revolution. I want to redesign everyone's phone ", it will also change a lot of household appliances, wearable equipment and so on," these devices have too many interfaces, if the use of voice recognition and equipment to communicate, it will be more convenient. ”

Technological development brings new opportunities for deep development

Wunda to build rockets as an example, expounded the reasons for the growing popularity of deep learning. "The rocket has two parts, the first part is the rocket launcher, is the big engine." The second is fuel, if the engine is small, a lot of fuel is impossible, if the engine is very small, the fuel is very little, the fuel burning is gone, so the rocket must have a large engine and a lot of fuel. Deep learning and this is similar, we now neural network is our engine, it drives the development of AI, fuel is the data we use, we now have so much data, social data to provide us with a lot of data, so input into our engine for us as fuel. ”

Now, with the rise of mobile internet, everyone's communication way slowly from the text to the image and video, based on this, more and more people began to search images and language, and through in-depth learning, can help people better understand the text and image, voice interaction between.

In Wunda's view, with the development of software and technology, new opportunities have been created so that software can analyze data through deep learning, so that we have more data, so that computer vision can do things that could not be done before.

At present, Baidu based on in-depth learning technology products have a lot of in-depth learning technology to help us better understand the language, pictures and voice, so that the computer better understand our communication. And hope to make better use of this technology in various ways.

One of the most important examples is the identification of objects. You can ask the computer to "look" at a picture and identify the object from it. In many well-known ratings, Baidu has achieved the best results. Some other teams have achieved the best results, and we continue to improve on this basis. This is achieved through deep learning techniques. This requires the construction of a huge neural network, input a lot of data, and Baidu's GPU technology makes it possible.

Robots don't control the world.

AI rapid development has contributed to some hype, we can not help but think that some day in the future, the computer may become a robot, control the whole world?

Wunda very seriously that they will work with us to create a very good world, and will not become an evil robot. Because, you can use technology to predict, technology constantly change, neural network is a very good identification tool, as well as predictive tools, when you enter data they can predict.

If we can stay away from hype, AI development is relatively healthy, the challenge of AI creation is not the machine will control the world, he is likely to help us replace labor, our leaders must seriously consider technology, let artificial intelligence technology to help us replace those complicated labor.

Baidu speech recognition is making good progress

Baidu's speech recognition technology is making good progress, the computer has saved a lot of voice data, through the training of neural networks, thus creating a very accurate speech recognition system. A big problem with the current speech recognition technology is that it can't be used in noisy environments, such as when you are driving in a noisy car. Baidu currently has the best voice recognition technology in noisy environments.

As to how to overcome the difficulties in the research and development of speech recognition technology, two tasks need to be done. On the one hand, Baidu needs to study how to build a large network. So Baidu has a very strong GPU team, and developed a powerful GPU software. On the other hand, Baidu needs to find enough data to be used as input to this large neural network.

Working in Baidu is exciting

Wunda has been working in Baidu for nearly a year, and is inevitably exposed to the company and its members. This is also an unavoidable problem.

Of course, Wunda was very happy when he mentioned the question, because it seemed to him that working in Baidu was exciting and enjoyable. The company is very agile and quick, so it can create something, and it's exciting to work with the research team.

He also exemplified it with speech recognition. Baidu has 3 laboratories, and these 3 laboratories often discuss and provide professional guidance to each other. Many people visit and communicate with each other. In the past, when Voice technology announced a breakthrough, the United States team made progress, the Beijing team will learn from the English environment experience, and instead in the Chinese environment to try. Baidu is a unified company, we strive to act as a team, the same company.

(Responsible editor: Mengyishan)