On October 27, Baidu held The fifth edition of The BIG Talk with The theme of robots Reshaping Future Life in Beijing. Wang Haifeng, vice president of Technology at Baidu, Woody Lipson, director of Creative Robotics Lab at Cornell University, and Daniela Ruth, director of THE MIT Center for Computer and Artificial Intelligence, discussed the development of artificial intelligence robots.
At the forum, Baidu technology vice president Wang Haifeng explained baidu’s thinking in the development of artificial intelligence to the outside world. He believes that robots should first understand human language, secondly understand human knowledge system, and also understand human itself. Once these conditions are satisfied, they will acquire comprehensive thinking ability.
Giiso Information, founded in 2013, is a leading technology provider in the field of “artificial intelligence + information” in China, with top technologies in big data mining, intelligent semantics, knowledge mapping and other fields. At the same time, its research and development products include information robot, editing robot, writing robot and other artificial intelligence products! With its strong technical strength, the company has received angel round investment at the beginning of its establishment, and received pre-A round investment of $5 million from GSR Venture Capital in August 2015.
The full text of Wang Haifeng’s speech is as follows:
Wang Haifeng: Good afternoon! It is my great pleasure to come to the BIG Talk of Baidu Hundred and share with you some of our work and thinking on intelligent robots. When we mention robots, there are many images of robots in everyone’s mind. Some of them work on the production line, some help people do housework at home, and some even appear in football or battlefield. It should be said that these mechanical technology and control technology of robots, has made a lot of achievements, how can make a robot more like a human? We need to make it as intelligent as a human, and the small robot that you just saw can even answer a lot of questions that most of us can’t answer, so it’s already very intelligent.
What kind of ability does a robot have that makes it human? We all know that people’s ability is reflected in many aspects, listening, speaking, seeing, walking, learning, and the most important point is thinking. For example, an animal can also have eyes to see, ears to hear and to act, and even it can train some abilities, such as what actions it will take when seeing a scene, these conditioned reflexes. And man’s most important ability is to think, which is unique to man. How do people think, how do we make robots think? I think need to have the ability of three aspects, the first is the understanding of the language, you know that language is peculiar to people an advanced intellectual activity, only a language, only can talk is not enough, we also need to know what to say, this is behind needs the support of knowledge, need to grasp of the knowledge, the use of knowledge, and knowledge reasoning, and so on. On the other hand, people live in a society, in addition to dealing with these objective knowledge, also need to deal with people. When we say the same word to different people, their feelings are different and their reactions are different, which requires robots to be able to model different people.
I’m going to tell you a story. You know spies are mysterious people with very strong abilities, very good at disguising themselves, infiltrating the enemy so that the enemy can’t find them. There was once a story about a female agent, who was very capable and did a very good job, but she finally revealed her identity, why? It was during childbirth that she spontaneously cried out her mother tongue and was discovered. What is the moral of this story? Language is a basic carrier of human thinking. For people with different mother tongues and languages, the characteristics of language will affect their thinking activities. When we learn English, we need to think in English when we learn a higher level. Thinking in English is equivalent to a change in your thinking mode. The same sentence is said differently in Chinese than in English, such as the one shown on the screen. The word order, diction, affirmation and negation are different in Chinese and English, which shows that language itself affects people’s thinking. When a person needs to communicate with the machine, for example, ask it some questions, ask it what the weather is like, or chat with it and so on, in fact, language is used for interaction.
How do we get machines to understand language? As we all know, language has been developed by human beings for many years. It fully carries the accumulation of human knowledge, thinking and emotion. Language itself is very complicated, to take The familiar Chinese for example. We all know that there is no space between Chinese words, so the first task of speaking Chinese is word segmentation. Please look at the screen, “Liu Qing Chu-moving”, you can easily know what the word is when you see the six characters, the front is a name Liu Qing, the back is Chu-moving. Similarly, we added a word, “Liu Hua knows this matter”, here is not “Liu Qing”, but “Liu Hua”. Change a little bit more, “Liu Hua clearly revisited”, is another participle result adds another word, “Liu Hua clearly remembered”, becomes Liu Hua again. Don’t worry, Don’t worry, don’t worry. But look at the next sentence, “Why are you in such a hurry?”, it is different again. The above should be said that there is a basic division can be divided, but this sentence we see, “ping-pong auction”, in the end is “ping-pong racket” sold out, or “ping-pong” “auction” finished? Both are true. That’s the complexity of language, and that’s what makes it so attractive. Chinese is a very attractive language. Let’s look at a real example in Baidu search. If you search in the Baidu search box for a freestyle athlete with height above 1.80, we will directly give these answers. Such a complicated search, we want to understand it well, we can find the corresponding answer. The tree on the left is the foundation of our natural language processing technology: semantic parsing, where we can analyze the relationships between words to better understand what the user is looking for. With the background knowledge base stored in a large number of knowledge retrieval, finally got such an answer.
You see actually “who is the son of Nicholas Tse” and “who is the son of Nicholas Tse”, the words of these two sentences are the same, but the order is not the same. If we use traditional search techniques, we would divide the word, calculate the term importance and so on, and we would find the same thing from the calculated correlation. Obviously the problem is different. How can we do that? It takes a deep analysis of the language, a deep understanding of the semantics, to know that the answers they’re looking for are different. As you can see on Baidu, “Who is The son of Nicholas Tse” can be given to “Xie Zhenxuan” and “Xie Zhennan”, which is based on a very profound language analysis and understanding technology.
What’s the weather like in Beijing today? “And then” What about Shanghai? The weather in Shanghai is very hot today. The weather in Shanghai is very hot today.
Giiso information, founded in 2013, is the first domestic high-tech enterprise focusing on the research and development of intelligent information processing technology and the development and operation of core software for writing robots. At the beginning of its establishment, the company received angel round investment, and in August 2015, GSR Venture Capital received $5 million pre-A round of investment.
“I plan to fly from Beijing to Hangzhou on Sunday. Help me book the tickets.” “Please help me book a hotel”, “Please book a hotel for me”, “Please book a hotel for me”, “Please book a hotel for me”, “Please book a hotel for me”, “Please book a hotel for me”, “Please book a hotel for me”, “Please book a hotel for me”, “Please book a hotel for me”, “please book a hotel for me”, “please book a hotel for me”, “please book a hotel for me”.
If there is not enough context between sentences, textual context can be used. For example, if we ask why the sky is blue, if we just give a classic search result, Baidu knows to find a result, like the result on the left. But you can see that the sentence that comes straight out doesn’t really explain why the sky is blue, and you can see the full explanation when you expand it. In a long paragraph, I find the part of the answer that is really the most relevant, and the summary is a better answer. If we do a better job, we can give very specific answers.