What do you think of when you mention reading and understanding these four words? Fear dominated by four or six IELTS TOEFL? The multiple-choice questions on the language test paper that even the original author can't understand? No matter which answer, there is no way to escape a law: the subject of reading and understanding, which is beneficial to body and mind, must be the same human being as you and me. After all, this thing needs to read a large paragraph of text, and then understand the meaning and then do the topic, which is called the most complicated and most difficult part of the exam, but also the most comprehensive test of a comprehensive ability. So your English teacher must have said this: you have to read the understanding of the world. However, however, if you tell us that today's world is no longer human, but AI, what do you think? Maybe we have imagined that word dictation, sentence translation and even writing can be done by AI, but reading and understanding this thing has been left behind by AI, probably many people have not thought of it. But this is the case. Recently, Stanford University's famous machine reading comprehension event SQuAD has set a global ranking. Alibaba broke the world record with an accuracy rate of 82.440, surpassing the average score of 82.304. Of course, AI is not going to brush you more than TOEFL. Behind this incident, the brain hole is a bit big... What is machine reading comprehension? Machine reading comprehension, although it seems to just let AI come to an exam. However, it is the biggest challenge in natural language processing technology after speech judgment and semantic understanding: let the agent understand the full text context. The SQuAD Challenge sponsored by Stanford University is recognized as the industry's highest level of machine reading comprehension. The basic rule of the SQuAD Challenge is to build a large-scale data set containing about 100,000 questions in a crowd-crowded way, and give an article from Wikipedia about a few hundred words in length. After reading the essay in the data set, the AI ​​model submitted by the entrant answers several questions based on the content of the article, and the answers are compared with the standard answers, and the results are finally obtained. Because of the "smart" adjustment of reading comprehension, it requires a lot of logic, detail and structural analysis capabilities, and directly affects the text materials in reality, so the actual value is very large. For example, the first question we have to face is, if artificial intelligence is better at human average than understanding and answering accurate information in the text, what will it bring us? When AI’s “reading comprehension†score surpasses humans, what does it mean? For example, you may be able to understand this question very simply: on the English test, when the machine can translate words, we are not surprised at all; but when the machine can dictate the whole sentence, we will sigh the progress of the technology; When doing reading comprehension, we will probably think: What else do I have to do this test? The difference here is that when dealing with reading comprehension related issues, AI is not only to calculate and record, but to actively analyze and understand, so the problem of reading comprehension has been asked by people to be the iconic critical point of NLP. But this point is cracked by AI, which means that a lot of work that must be done by humans has been officially taken over by AI. Because reading comprehension problems, not only to deal with speech and simple semantics, but to understand and pay attention to the complex organization network composed of elements such as vocabulary, sentences, text structure, thinking logic, auxiliary sentences and key sentences. Machine reading comprehension "reaching standards", the most direct industrial impact, is the understanding of most of the rules, dialogues, and service information that must be done manually by humans today, all of which can be replaced by artificial intelligence. For example, the work of customer service, information management and recommendation can be replaced by a machine with a lot of speed and high computing speed. How does machine reading understand how it works? Perhaps we have all noticed the problem that today's Internet world is experiencing more and more text content and various kinds of information exploding. Too many things that you know, what you don't know, what you think you know but don't actually know. Even if you double 11 want to grab a hand, there are all kinds of rules of the game waiting for you. Read it yourself, too tired and too painful, no time, ask customer service, it is very likely that the error is still very slow, it is simply Schrödinger. This may be solved by machine reading comprehension. For example, if a customer has doubts about an e-commerce promotion rule, he can directly ask the AI, and the AI ​​can use this problem as a reading comprehension question to provide solution solutions. Machine reading comprehension will play a role in providing non-templated intelligent customer service to customers. And when AI surpasses manpower in these capabilities, the value of machine customer service utilization will likely increase rapidly. In other words, the machine customer service can finally be less than the machine..... It is not difficult to see that this benchmarking of key capabilities has the greatest benefit to a large number of industry lines that emphasize interaction with the average consumer. In a nutshell, machine reading ability is also an effective way to find and recommend content in the entertainment field, such as reading the complex needs of users and making accurate recommendations. Combined with IoT products, giving feedback on users' large language, and even dialogue. Interaction must rely on machine understanding. Understanding, let us not only be a child in the future In addition to knowing that AI can act as a better customer service, why should we pay attention to machine reading comprehension? Perhaps the key is that we should know the importance and expectation index of “understanding†in the current AI world. As a branch of computer science, AI is inherently capable of computing, and hopes to carry out intelligent simulations that mimic human intelligence. The second step is to imitate human perception. Machine vision, speech recognition, and semantic understanding that we see today are all doing this. The third step is to let AI produce understanding. Obviously, recognition has a huge application scenario and will dominate the next long period of time. But AI's ability to understand is the evolution of most recognition capabilities. If it is simply recognized but cannot produce output, then AI is nothing more than a more flexible sensor. From this logic point of view, reading comprehension is not just a test, or a technical blessing of commercial applications, but more importantly, an accelerator that unlocks the understanding of AI. The broader meaning is that we may be one step closer to never having to test machine reading comprehension. When we no longer consider whether the machine can understand human text and language, DeepNLP will be possible, and the scope of human-computer interaction will be expanded. Machine intelligence can begin to capture human logic and correspondence. Maybe it's still far away, it may be very close, but machine comprehension makes us more than just a child in the future, there should be no doubt. Portable Battery ,Portable Power Bank,Portable Battery Pack,Portable Power Pack Zhejiang Casnovo Materials Co., Ltd. , https://www.casnovo-new-energy.com