(Original title: Bamboo Jian Renjian: Breaking the stereotyped chatbot)
Lei Feng: According to the question-answering system, chat robots have existed for decades. But only in the past three to four years, how to make the machine smoothly and naturally communicate with people has become a hot spot for commercial companies, especially domestic Internet companies. From Microsoft Xiao Bing, Baidu Duwei to various vertical industry mature customer service robots, these chat robots come in all shapes and sizes.
Lei Fengnet tried to find the reasons for the formation of the domestic typical Chatbot company and the business opportunities they saw in a series of interviews. This article is an interview by Lei Feng. Prior to starting the business, Jian Renxian was one of the top leaders of Microsoft Asia Engineering Academy. The projects he participated in and reported include Bing Asia’s strategy and partnership, Microsoft’s virtual assistant Xiao Na and business realisation in Asia. . After the start of business, Kenji Kenji's bamboo intelligence has expanded to more than 100 people today. Jian Renxian believes that doing bamboo is roughly one thing: Let the machine read the context, read the emotions and emotions of the person, and then naturally engage in dialogue with people.
At approximately 8 pm on Monday night, Kan Renxian sat in an opposite chair and began an interview. After this interview, he also had a product meeting at 9:00. His low voice was interrupted by coughing from time to time, but his expression did not show obvious fatigue. After more than a year of starting a business, most of his answers were clear and powerful when it comes to chatbot Chatbot and emotional computing. These judgments in the above dialogue with Jane Kenji, an ordinary person can easily capture from the tone, action, expression and even the environment; but for the machine, this process is very complicated. And what he envisioned as a chatbot is more than just doing it.
First, the history of Chatbot
The chatbot Chatbot has been around for a long time, from my student's time. I wrote a robot quiz program myself. In fact, it is very simple, the question answering system is to find keywords, a word and two words together, you can find a sentence. Chatbot, it should be said that someone has been doing, but all as an exercise, never really popular.
I think Chatbot can do it. That's when I was still in the United States. It was the time when the search was the most red. In the middle of the 05-10 years, when the demand for search was greatest, I felt that Chatbot should have a future.
why? Because the search is always in the user's input of a keyword, it returns a scenario where multiple results are linked. In the past, there was a company in the United States called ASK.COM, which had already appeared in the early 2000s. It is the first public natural language question and answer system, not the search. The user finds the information he wants in the form of a sentence. Compared to keywords and links, this time, it is a bit more advanced than search, and it is natural.
But in the middle of many years, this kind of question and answer system is not popular, and no one can think Chatbot is useful. Because a robot, if people think it is useful: To make people use it, feel very smart, if you ask one, two, can not answer, people will not come back. When people define a thing as a small toy, they will not return.
So Chatbot has used it for an exercise in the past few decades, and some people use it for academic purposes. There is no ordinary person to use it again.
Second, technology is not a fundamental issue. Idea is the fundamental issue.
Let me want to regain Chatbot, it is quite a long time later. I came into contact with the movie "Her" in 2012-2013. I think the scene in the movie is achievable. She has a lot of human nature inside, not just a robot. A Chatbot can understand you and can be human. But before she has humanity, she must first understand emotions. I think there are many things that can be done here.
At that time, I thought that the definition of the robot should be rewritten. But even now, four years have passed, there are still many questions and answers on the market, and they are mechanical. They use keywords and keywords to find out. Many robots on the market today are not talking about AI. They are just tools for commercialization in traditional technology.
At that time, it was not technically mature. But in my career, technology is never difficult. Because we learn technology, we think that as long as we can say things, we can probably achieve it. It's just a matter of time.
So technology is not a fundamental problem. The problem is the concept, the idea, and your opinion of Chatbot. If you think that Chatbot is still a question and answer, very mechanized, then your technology is still high-end, and you still make a question or answer; if you can imagine Chatbot is emotional, then the robot can understand you.
It sounds like everyone has this vision. However, in reality, some people did not go in this direction. This is not a question of technology. It is a question of design concept. For example, when WeChat first came out, there weren't a few features. Like other chat platforms, technology is not a challenge in the development of chat platforms. But how many of them can be as successful as WeChat? If Chatbot is just a fusion of technologies, every Chatbot made is the same, and everyone's IM should be exactly the same as WeChat. no.
When we look at Chatbot again, it is very important that we design the ideas, the thinking of the architecture, and the thinking of the algorithm. Is it not in a humane way? If we say that the vision is to go for humanity, but the architecture, algorithms, development or take shortcuts, then do it with the Chatbot 20 or 30 years ago is no different.
Third, the emotional robot modeling
Be a humanoid robot that understands you and has emotions.
First, we must start from the emotional mechanism. How many teams do robots really invest in emotional computing? Everyone is investing in corpus, word vectors, and a search. Has anyone ever thought about how to do an emotional calculation? If someone wants to do a Chatbot, simulate human nature and simulate human behavior, including understanding human behavioral models, and allow Chatbot to interact humanly.
There are several ways to model human nature and emotions. One is the method of calculating cognitive science. It can also evolve from a logical model to a machine learning model. So one is logic to simulate human emotions and behaviors, one to use machine learning model to predict.
What should I do when you are dissatisfied with me? When you are bored and tell me that you are annoying, how should I respond? When there is nothing to say, what should I do?
First, we must be able to identify such a state. Second, we must be able to cope with such a situation. How do we adjust ourselves after an inappropriate state and become a good state? This requires a combination of logical models and machine learning models.
On the other hand, machine learning can create a lot of interactive data. This interaction data generates a good interaction model, and the interaction model allows the machine to respond to different emotional states. Combine these two to simulate human interaction and human behavior.
The important point is that your Chatbot is able to recognize and understand emotions, understand the intent of the conversation, understand the intent, understand the many different things in the entire language, and you have a way to make a real understanding. robot.
Fourth, Chatbot's business
Over the past few years, the popularity of information and communication platforms, from the previous QQ to the current WeChat, foreign LINE, Facebook Messenger, and other chat platforms are prevalent. In recent years, humans have become accustomed to chat. The behavior of people has changed. If people's behavior does not change, then the acceptance of today's chatbots can not be so high.
Basic technologies have also improved, including machine learning tools, platforms, and deep learning. You need a RNN model, Google wrote it for you, and TensorFlow wrote it for you. You don't need to code a code, just focus on your model. In the same way, in the tools and technologies, everyone’s experience is also rich. The technical theory of this part is iteratively very fast, so the technology is mature.
Third, there is a lot of data. The most precious data is not from the Internet. The most precious data is self-generated data. However, the data from the network is the key to cold start, data cleaning, data cleaning after cleaning, marking after finishing and so on. The cost of data storage has also been reduced. GPUs are getting cheaper, and GPUs can grow at a rate of 3 times a year. Without price increases, performance can be increased by 3 to 4 times, and the annual growth rate by 20 to 30 times.
Under these conditions, I think it is the best time to do a chat robot. Not even now, it was more than two years ago. For example, Microsoft Xiao Bing began in late 2013 and early 2014, and these preparation conditions have already existed.
The emotional robot I want to do is contrary to Xiao Bing and is completely customizable. Let everyone have their own robots, and each merchant can have its own robot. I want to make a robot factory that can make many emotional robots, not just one.
A robot is not just an IP, such as Disney's Mickey Mouse. It cannot be a spokesperson for other businesses because there is in fact only one Disney. Every business has its own brand, so every business should own its own robot.
If you really want to push the robot to various places, it needs emotions, knows you, and can "remember". The robot will remember "you" and the people around you, learning around you. This robot will grow up with you.
The construction of such a robot is divided into many steps and requires relatively large resources. We must start with the platform. It is actually a 10-year plan.
The first one must be able to prove that chatbots are sticky with people. If they are sticky, they must gain people's trust. In the past few decades, robots have not been able to trust people because they are not smart enough. Only chat, it is impossible to have stickiness. After Chatting, Chatbot can help people do something, so that there is a way to achieve greater stickiness, emotionally and functionally combined.
Secondly, the chat robot must be able to play a role in helping people in their lives. It must be integrated with the business. Otherwise, it can help you check the weather, and can help you do a search, but there is no way to help you in other business services, Chatbot has no way to stick with people. Only chat, it is impossible to have stickiness.
I think a complete robot with human understanding, this dream may take 10 years to reach. Next, it may have to eliminate search, or even replace some of the entrance of the App. This process must take place in the future evolution.
Flat Control Cable Also known as festoon cable, this flat cable fits in confined spaces such as overhead cranes and hoists. It sends signal and data to operate, measure, or regulate automated equipment.
Raw cable can be UL2651, UL20251 ,etc
Flat cable, IDC cable, flat cable wiring, ribbon cable harness, flexible flat cable
ETOP WIREHARNESS LIMITED , https://www.oemwireharness.com