For the college entrance examination students, the happiest thing is that they have already memorized the answers to the problems they encounter but will not encounter... However, can such a good thing really happen Just yesterday, a Weibo blogger forwarded a video released a week ago, saying that he claimed that "he had predicted the composition questions of the 2022 college entrance examination with AI, and four of them were bet in seven college entrance examination papers across the country."
Wen | Sina Technology zhouwenmeng
Editor: Han Dapeng
Is AI predicting college entrance examination composition a science or a Metaphysics?
Sina Technology contacted the above blogger panghu. As a senior algorithm expert of the AI system, he said that the crawler made by the team captured the comments made by the people's daily, people's.com and other official media within a year and established an AI system, which predicted the most likely seven propositions.
However, some netizens questioned that the AI prediction direction is macro, "reliable Chinese teachers will set the direction, which is macro. In this way, everyone can be called the king of bet questions".
How does AI predict composition questions
If you want to understand the technical principle of AI predicting college entrance examination composition questions, you must first understand the essence of college entrance examination composition questions.
In the video released on June 1, announcing the AI prediction results of the 2022 college entrance examination composition, the blogger introduced the technical principles of the implementation of this AI system - the college entrance examination composition seems to be an exam, but in fact it is the guidance of values, outlook on life and national consciousness. This rule determines the score of a college entrance examination composition.
Therefore, "the essence of the college entrance examination is the ideological and political task, which reflects the consciousness." This is the fundamental principle of predicting the composition of the college entrance examination.
In essence, the idea of composition proposition is the same as the evaluation of Spring Festival Gala sketches, and there is no difference in essence. The proposition group that knows this well will also select a number of topics with high importance and small scope from the international events, national policies, conference documents, guidance and official reports of the past year, and finally select the composition questions of the college entrance examination that have not appeared before and are most suitable for this year's tasks. As an important tool that is best at mining rules from large-scale data, AI can predict the final results more accurately with more data and more reasonable distribution.
Based on the above thinking, panghu and his team specially wrote a set of reptiles. This set of crawlers captured articles, periodicals, newspapers and important documents of high-level meetings published by the people's daily, people's daily, CCTV news and other state-owned media within a year. Then, through the pre training model and topic modeling algorithm in naturallanguageprocessing, an AI system that can understand Chinese News and articles is established. The AI system has learned the idea of writing questions in the college entrance examination in the past few years, and finally predicted the most likely seven proposition directions in the college entrance examination in 2022.
In the forecast announcement released on June 1, panghu and his team announced seven propositions, namely, "difficulties and hope, struggle and self-improvement, exploration and dream, innovation and development, struggle and value, fairness and moral education, beauty and environment." Behind the core words of each proposition, there is a "model representation" equipped with prediction model and supply for expansion and extension.
Take the prediction direction of "difficulty and hope" as an example. In the prediction results, the AI system mainly extracts the prediction topic and core words based on the model representation. In the process of students' understanding of the proposition, the prediction topic can be used as the main writing content understanding direction of the composition, and the introduction provided by the model representation can be embedded in the article as a case reference. The top core word "difficulty and hope" can be used as the main proposition reference direction of the composition title.
How accurate is the forecast
The proposition result of AI system bet has come out. So, is the predicted result accurate?
According to the statistics of sina science and technology, it is found that the seven sets of test papers of the 2022 national college entrance examination (NMET), namely, national a, national B, national new NMET II, national new NMET I, Beijing, Tianjin and Zhejiang, were officially announced on June 7, and the seven sets of NMET composition questions were indeed consistent with the proposition direction previously announced by the fat tiger team on a certain level.
For example, in Volume II of the new national college entrance examination in 2022, the composition requires to carry out composition writing with the theme of "choice · creation · future", and mentions the words "scientists, scientific research, Beidou navigation system" and "new methods, new thinking, new knowledge". Among the proposition directions given by the AI system with the theme of "exploration and dream", scientific and technological terms such as Beidou and chip have become the representation of the prompt model, and the theme direction is also close to the composition proposition of the college entrance examination.
In addition to the new college entrance examination volume I and the national examination volume a, the proposition direction predicted by AI also hit the relatively close theme direction. But can the highly general proposition direction obtained from the statistical perspective of this AI system really help candidates prepare for the exam? At present, it seems that the outside world has different opinions on such results.
Some netizens exclaimed, "cow! Can you predict CET-4 and CET-6?".
And another part of netizens questioned, "Ai fortune tellers who laugh to death, Chinese teachers who are more reliable will do."
As the AI system developer who won the four essay questions this time, panghu also objectively pointed out that "this year, there are many college entrance examination compositions in different places that combine hot topics, which makes the AI system win the four propositions. However, in previous years, the topics generally do not directly come from the hot spots, but are described from the side and see the big from the small. This time, the more hits are also very special."
AI can guide candidates to replace teachers in the future
Although some of the students fail to pass the college entrance examination, AI still does not directly guide the students to prepare for the college entrance examination or replace the teachers in predicting the topics of the college entrance examination.
Regarding the fact that the proposition direction given by AI is relatively macro and not specific, panghu further pointed out that "the AI system predicts the college entrance examination composition more to tell the outside world that AI can already do a lot of things, but over the years, the college entrance examination composition still tests students' knowledge reserve and language flexibility, and the direction of AI prediction can only be used as an auxiliary reference."
In addition to AI predicting college entrance examination compositions, with the reduction of the threshold for the use of AI technology and the continuous popularization of AI technology, the promotion of business process optimization and business prediction based on AI technology is becoming more and more widespread in the fields of medical treatment, agriculture, industry and enterprise business. Are these predictions and judgments based on AI science or Metaphysics?
According to panghu, "in the fields of data classification and prediction tasks, the prediction accuracy of some AI has far exceeded that of ordinary people." At a time when intelligent voice technology and face recognition technology have been widely integrated into public life, AI has already become a frequent visitor in our life, and the prediction of results based on AI and big data statistics is no longer metaphysics.