AIR 023 | From Search Engines to Action Engines, Former Google Scientists How To Build Smart Search

Lin Dekang, who first worked as a senior management scientist at Google, later decided that Google did not make much sense and decided to venture out and set up an artificial intelligence company singularity and intelligence.

Think Google doesn't fly? The reason why we did not know, however, at today's global artificial intelligence and robotics summit, Lindkang said that our current engine is progressing from the search engine to the question-and-answer engine, that is, the progression from language to natural language recognition. However, the ultimate goal that can be foreseen at the moment is -- the Action Engine.

If we want to know Shenzhen's annual rainfall, then we can complete it through search engines, such as those developed by Microsoft. From the search results page we can see very rich search results, the title content looks good, seems to answer your questions. But really, this one is still flawed. Because when we see these results, why do you want to go in?

Lin Dekang introduced that in his last few years, Google was leading a team to develop a question and answer product. This question and answer product can be used in a scene that the user needs, such as when there is a specific problem to be searched, the product can directly find the answer to the user. In general, the search result is to set the key words for search to bold, but if the answer is automatic, it can present the answer directly to the user, so you can see more clearly.

For example, if we need to search for “What is the fastest bird?”, the system can learn keywords, learn something that seems to be useless, and present the user directly with “Falcon (picture)”.

"Sometimes you really want to do something, like buying a plane ticket, Google's question and answer system can not meet the demand, but you can do a lot of advertising." Presumably, this is why he think Google does not fly.

Lin Dekang said that it is actually an era that evolves from speech to natural language understanding, and it is a relatively favorable time for the threshold. “A voice assistant product that we do after we come out is a product that you can help you do when you have a request.”

According to Lin Dekang's introduction, there is such a product. When the user expresses the demand, the product will execute it for you. In the future, when you use the app to acquire certain services, it will be like an invisible finger to help you operate.

Search Engine, Answer Engine, Answer Assistant

Speaking of changes from search engines to answering engines to answering assistants, we can first compare the difference between search engines and answering engines.

Lin Dekang said that when searching, the things that need to be found are actually handed--that is, the key words. "You just have to go to a known address and find a site that has a lot to do with it." But when it comes to answering the engine, the situation is not so simple.

You have to know what you are looking for. When we do a question-answering system, we actually use the back-end documents to find the answers and then present the results of the answers to the stage. This is the best result.

Now Microsoft also does this, but Microsoft's answer system has other uses.

In fact, the operation process of the question-and-answer engine will be preceded by an ordinary search. The special place is the follow-up processing. For example, when the user asks “what enzyme can break down the starch”, the first step is to know what the user is asking, so it will first find the enzyme. In the second step, the system will gather all the results related to the enzyme and then score the results on various indicators. Find out the Focus Words by scoring to find the path that is most likely to achieve the goal.

"For the question that contains these words in the sentence, it is very easy to write a pattern match based on this in the future. Finding out the Focus Words is a simple way to achieve more than 90% accuracy."

So what if there is no WHO in the problem? Lin Dekang explained: "Look at what the results of WHO are similar, even have the same answer, and then you can go there to find out, so you can know what its Focus Words are."

In other words, the quiz engine has more semantic understanding than the search engine. However, search engines and question-and-answer engines give users some information, but they still require users to perform certain tedious actions and needs. The Action Engine is aimed at this "intelligence is not a thorough pain point", hoping to help users perform some things. This is a combination of search engine and voice assistant.

"So if you are young, you don't need to," said Lindkang.