Lin Dekang: Making voice assistants is the best way

After returning to China to start a business for a year and a half, Lin Dekang visits his mother almost every week. His mother often asks him how the company is doing now and whether he regrets quitting Google. Lin always smiles and says that everything is going well now.

Many people have asked similar questions. As a professor at the University of Alberta for 8 years and a scientist at Google Research for 12 years, Lin’s choices at that time were beyond his comfort zone both in terms of work and life.

This can be sensed when communicating with him. Lin always has strange pauses when speaking. Xing Jiayuan, a member of his team, said, "Teacher Dekang's punctuation is very obvious. He always thinks before speaking, and his inner thoughts must be 'how do you say this sentence in Chinese?'"

In short, in the eyes of many people, Lin originally had better choices.

Someone once asked on Zhihu, who are the great figures in the NLP (natural language processing) world? As one of the few ACL Fellows in the Chinese community, DeKang LIN's name was mentioned by many people. Someone also added, "DeKang LIN. After returning from Google, he did not choose to teach at Tsinghua University, nor did he accept the invitation from BAT, but instead started a small startup, Singularity Intelligence."

The person involved obviously does not think that this decision is uneconomical. Lin told Leifeng.com (Official Account: Leifeng.com) that he often had the idea of ​​starting a business when he was in Silicon Valley and talked to many people about it. Although leaving Google is a difficult decision, once he has an idea that makes him "exciting", he makes up his mind very quickly.

In 2014, Lin Dekang's good friend Wu Xiaoyun resigned from Google and made various preparations to start his own business. By the end of the year, he began to build his own team.

Lin and Wu met in 2006, when Wu had just joined Google and, like Lin, was also a scientist at Google Research. As both were Chinese and had similarities in work and life, the two became close. According to Scenny, a software engineer at Singularity Intelligence, "During the eight years that Mr. Dekang worked with Xiaoyun, he was the closest person in their team."

This seems to explain why Lin eventually chose to start a business with Wu. But in fact, when Lin joined Singularity Intelligence, it was already a year after the company was founded. Such a long decision-making process can to some extent explain that the two people joining forces was not a matter of course.

"Xiaoyun has a strong one-on-one persuasion ability," said Xing Jiayuan. Almost all the technical backbones of the company were persuaded by him to join one by one, and some even came by chatting with him without knowing him at all. But even so, in the eyes of some start-up employees, it was almost impossible for Wu to ask Lin to join.

Scenny was the first employee of Singularity Intelligence. At the end of 2014, he received offers from both Facebook and Google. When he found Wu Xiaoyun through a friend, he wanted to ask him to help him choose a team. At that time, Wu recommended Scenny to Lin's team and told him what he wanted to do.

"If he looked at my resume, Professor Dekang would definitely not choose me," Scenny said. "The resumes of people admitted by Google are put into a pool, and all team leaders can recruit people from here. Professor Dekang graduated from the research institute and basically only recruits PhDs, so according to his standards, I don't think he would recruit a Peking University undergraduate."

Although Wu kept Scenny updated on the progress of recruiting Lin, and said, "There is an 80% chance that Dekang will start his own business. If he returns to China, he will 100% come to Singularity." But in Scenny's view, the possibility of Lin joining Singularity is still extremely low, so even though he has already joined Singularity, he has already discussed with Lin about joining Google in the future.

The turning point came in the second half of 2015.

Lin's previous hesitation was about the business direction of Singularity. He told Leifeng.com, "Xiaoyun started out by doing in-depth sharing. In that case, I couldn't really get involved, so I didn't pay attention to it." After Singularity Smart started to develop input methods, Scenny could clearly feel that he saw Lin in the company more and more frequently.

It was not until Christmas Eve 2015, when Wu returned to the United States and met Lin at Google headquarters and said that the company's business would shift to language assistants, that Lin finally decided to leave Google. "Starting a business is difficult. If you don't do it full-time, the success rate is very low." Lin told Leifeng.com that his goal has always been clear - to apply natural language processing technology to real life and make users feel that it is useful.

However, this does not explain why he did not stay at Google.

Today, it has become a trend for academic giants to join the corporate world, but before Lin joined Google in 2004, this was extremely rare.

In October 2003, Lin, who was still a professor in Alberta, began to consider academic leave, so he contacted Peter Norvig, director of research at Google. Peter welcomed the idea very much, so Lin submitted the emails between the two as application materials to the school, which was quickly approved.

But in March of the following year, when the semester was almost over, Lin contacted Microsoft Research again. He had stayed at Microsoft Research Asia during the summer vacation, and he was more familiar with Microsoft's people than Google's. The other party was also quick to act, and sent him an offer the next day.

In addition, Microsoft Research is in Seattle, Google Research is in San Francisco, and the academic leave is nearly a year. Lin considered that his family would not go with him, and Seattle, which has direct flights to Edmonton (where the University of Alberta is located), is obviously more convenient. So he told Peter about his choice, but unexpectedly Peter said to him, "Don't rush to make this decision, come and take a look, and bring your family with you."

Today it may be difficult to imagine the clear distinction between academia and industry, but more than ten years ago, when Lin was teaching at the University of Alberta, "being a professor means being in school for a lifetime, and generally working until retirement." A friend also told Lin that if he was 20 years old, he could see through to 60 years old.

Lin didn’t think it was a bad thing. “At least what I did in school was interesting, and that kind of life was satisfying. I just don’t know what opportunities I’ll encounter in the future. For example, Google,” he said.

Lin has never concealed his excitement when he first came to Google. "I like this place very much. What everyone is doing is great. It seems that we can apply NLP (natural language processing) more directly." Peter took the opportunity to ask him if he would like to come here full-time, not just for academic leave. In this way, the visit turned into an interview, and the two sides quickly reached an agreement.

If Peter hadn't invited Lin, it would have been a foregone conclusion that he would go to Microsoft Research. If he had gone to Microsoft, Lin estimated that he would eventually return to school after his academic leave. So now looking back on his choice at the time, Lin believes that "this was a very correct decision."

"Dekang, you have a PhD, don't you?" Lin's former colleague in Alberta joked after finding out that he did not have an independent office at Google. This sentence, to some extent, points out the change that Lin has experienced from Alberta to Google. Of course, he likes this change.

Even if it is the same publishing of papers, there is a big difference between Alberta and Google. Lin told Leifeng.com that in school, it is necessary to publish papers because it is the annual assessment standard, but at Google, no one encourages you to write, and no one forbids you to write. The motivation is to promote open source projects like TensorFlow, contribute to the industry, and occupy the recognition of practitioners.

On the practical level, "publishing academic papers requires convincing others, so you have to spend a lot of time to go through other possibilities again. This is purely for writing articles, but when writing papers at Google, you only need to convince yourself. Such articles are actually what you want to say, rather than a waste of time."

Although he has published more than 90 papers, which have been cited more than 14,000 times, Lin can no longer remember the last time he published a paper as the first author.

This is related to his original intention of joining Google, and it is also the reason why he transferred from Google Research to the engineering department in 2013. "For something that is truly useful, you naturally don't want to write an article. Products are the best embodiment of research results. If you can make a product, why do you have to write an article to help others copy your product?"

Lin was undoubtedly embarrassed by the fact that Google Research had no product exports. When he was still working on a question-answering system at the research institute, he had to talk to the engineering department about productization. "But people at Google are very smart and think they can make anything, so they are reluctant to use other people's solutions. Even if they do, they are relatively peripheral."

What is even more frustrating is that "the product cycle of the engineering department is relatively short. After a long time, this group of people will do something else, and when the next group of people are found to work together, there is no memory of the past, and they have to solve many of the same problems." When talking about this, the reason why Lin left Google is obvious. He told Leifeng.com:

"Large companies have a lot of inertia. The same is true for Google's voice assistant, which is mainly to put existing services into the dialog box. This is the easiest thing to do and everyone likes to do it. If you make a brand new thing, you will always step on someone's foot or that person's foot. If we are in a startup, we can have full control and do whatever we want."

So in March 2016, after the New Year, Lin Dekang completed the resignation procedures from Google, and in April he officially joined Singularity Intelligence as CTO. In May, on behalf of the company, he told the A-round investors, "We want to be a voice assistant, not a deep sharing company."

On the day when Lin Dekang was interviewed by Leifeng.com, the headline in the technology circle was that AlphaGo Zero completely defeated AlphaGo, and the "unsupervised learning theory" was rampant. Lin denied this argument and said that even if there are many possibilities for a 19×19 chessboard, there are boundaries, and the most difficult part for AI, including natural language interaction, is not knowing where the boundaries are.

Therefore, Singularity's smart voice assistant, Little One, has never been presented in a human-like manner. Compared with Siri and Cortana, its interaction is more targeted. "Little One helps users do things in the application. The application scenario is very clear, and the probability of guessing the user's intention is much higher."

Although he has achieved the commercialization of natural language technology in a completely different way, Lin does not shy away from the cruelty of being in the same race with giants. In his opinion, voice assistants cannot succeed in a very narrow field. They must be either big or non-existent. As for whether the little guy will become cannon fodder or fight his way out, he did not say it directly, but told Leifeng.com,

"From a historical perspective, voice assistants like Siri that connect to backend services have developed slowly and have not covered many functions for many years. We are much faster than them. As long as a voice assistant helps users frequently enough, users will be more willing to use it, even if it is not omnipotent."

In fact, it is impossible for Singularity to build a platform or ecosystem like a big company. Although Lin has a lot of halos, he always maintains the consciousness of an entrepreneur. "If we build a platform like Google and Apple, who will access it?" So no matter from which angle you look at it, Lin believes that the path that Singularity is taking now is "the best."

"Of course, the possibility of failure is also high," Lin said with a smile, "If the probability of failure is small, it wouldn't be our turn to do this." In his opinion, whether he is cannon fodder or not has nothing to do with the correctness of his personal choice. "Because there are too many uncontrollable factors in success, there is no reference value for winning or losing. Anyway, as long as you like to do this, you will not lose. The only question is whether you win more or less."