20 1 1 year, IBM Watson defeated human beings in the quiz and won the championship, which became a milestone in the history of artificial intelligence. Since then, IBM Watson has expanded into medical and legal fields and transformed into an intelligent medical system. If the college entrance examination robot is admitted to Tsinghua Peking University, it may become another milestone.
Thinking college entrance examination robot
With what? Like AlphaGo, the college entrance examination robot has no entity, but an artificial intelligence system that can automatically solve problems. According to reports, unlike the photo search technology based on image recognition and matching, the college entrance examination robot can answer new questions that have never appeared before by learning and training in advance, and give detailed problem-solving steps.
The picture shows the Japanese college entrance examination robot Torobo-kun.
Why did you choose to take the college entrance examination? Japanese college entrance examination robot? Torobo-kun's R&D team explained this. Computers are good at calculating, so they can easily beat professional players in professional chess, chess and other board games. However, the college entrance examination is a difficult test for human society. For computers, answering college entrance examination questions requires not only powerful computing power, but more importantly, understanding human thinking process and information processing process. If you pass the college entrance examination, it also represents a new breakthrough in the field of artificial intelligence.
Zhang, the founder of Xuebajun, said that he hopes to pass this time? PK shows the progress of artificial intelligence in the field of education. He is full of confidence in this special college entrance examination. "After learning tens of millions of questions, the machine has been able to think about knowledge points like people and output the problem-solving process step by step instead of simple violent calculation."
It is understood that AI-MATHS has studied more than 7,000 test sites from primary school to high school, and the calculation amount can reach 2 to the 800th power. Lin Hui, the founder of its R&D team Quasi-Cloud, believes that the research and development of the college entrance examination robot is more difficult than AlphaGo. The reason is that it is relatively easy to describe the rules of Go in computer language, but to develop the college entrance examination robot, we must first make the system understand human language. "For example, when encountering a new word that has not been learned, human beings will guess the meaning of the word in connection with the context, and it is relatively easy to guess correctly; The robot will get stuck. " He explained.
This is the Japanese college entrance examination robot? The reason why Torobo-Kun gave up the college entrance examination. Torobo-kun has been taking the Japanese college entrance examination every year since 20 13. Its goal is to be admitted to the University of Tokyo. In the previous exam, it scored well in physics, but it was not satisfactory in other subjects because of its language processing ability. Professor Noriko Arai, the head of robot research and development in Japan's national college entrance examination, said that under the current technical conditions, it is difficult to be admitted to Dongda University. She explained: "The artificial intelligence system cannot understand the necessary information, and its ability to read and understand the meaning of sentences is limited." Next, Torobo-kun who gave up the college entrance examination will be applied to the field of data analysis.
In fact, before the concept of college entrance examination robot became a hot keyword, the world's top research institutions had made various attempts in various professional fields to develop an automatic question answering system for knowledge processing. Like what? The Cyc project, which started in 1984, aims to build a huge knowledge base of human common sense to answer and solve a series of scientific and technical problems. Halo, a project launched in 2002, aims to develop a scientific knowledge base to answer complex scientific questions raised by students or professionals, while Aristo is dedicated to answering standardized examination questions.
How does AI automatically solve the problem?
Automatic question answering system is a frontier research in the field of artificial intelligence, involving many fields of artificial intelligence technology, including image recognition, speech recognition, natural language processing and so on. According to Chen Ruifeng, the chief scientist of Xuebajun, the process of solving the problem includes three steps:
One is to understand and recognize human language and turn the topic into a language that robots can decode and understand, that is, to convert human language into formal language through natural language processing.
The second is logical reasoning, which uses the knowledge language network of computer to simulate the way and strategy of human processing information and find the best solution.
The third is to answer questions in human language and give detailed steps to solve problems, that is, to transform formal language into natural language.
The biggest difficulty is to make the machine understand human language, which is also one of the core issues recognized by the automatic problem-solving system: semantic analysis in natural language processing. Machines first need to recognize human language and analyze its meaning, including all kinds of common sense, riddles and other hidden clues, such as the classic mathematical problem that chickens and rabbits live in the same cage, suggesting that chickens have two legs and rabbits have four legs, while computers may not know this common sense and are better at accurate calculation under rules, but human natural language is not accurate. For example, in the process of solving physical problems, computers can't understand thinking modes such as ignoring the size of objects and assuming zero friction.
AI-MATHS has also encountered such a situation. When there are new words in the topic that robots have never "learned", such as investment and financial management, they will get stuck because they don't understand.
Another point is the ability of logical reasoning. Robot research teams in different countries found the same problem: in different subjects, robots perform better in solving problems in liberal arts. The reason is that science emphasizes logical understanding and reasoning ability, while machine learning has not made significant progress in this field. At present, the emphasis is on memory and computing ability. So mathematics solves problems automatically.
How does the college entrance examination robot change education?
Like research and development? IBM Watson is not only for taking part in intelligence quiz, but also the college entrance examination robot is not the ultimate goal of artificial intelligence system development. Academically, the college entrance examination robot can test the extent to which artificial intelligence can simulate the process of human thinking and understanding. As far as practical application is concerned, it is necessary to further improve the efficiency and effect of teaching and learning by using technology.
In 20 14, the Ministry of Science and Technology launched the 863 project "Key Technologies and Systems of Humanoid Intelligence Based on Big Data". Wang Shijin, vice president of Iflytek Research Institute, said that in the past three years, a lot of progress has been made, including cognitive reasoning problem solving, automatic Chinese composition writing, subjective question answering based on automatic extraction of geographical knowledge, reasoning problem solving based on historical deep learning, and multi-dimensional intelligent answer reading based on OCR handwritten character recognition.
These technologies have their application scenarios in the field of education.
Intelligent teaching assistant
It can provide students with real-time answering service. IBM Waston Company launched Jill Watson application in the field of education, trying to become a new teaching assistant in the classroom, responsible for providing students with real-time feedback and answering services. 20 14, this application has been put into use in Georgia institute of technology and other schools. After debugging by the research team, Jill can achieve 97% accuracy.
Automatic marking and marking
With the integration of image recognition and semantic analysis technology, the automatic correction and grading of subjective questions can be realized, which can reduce the teaching burden of teachers and accelerate the collection of online teaching data.
Adaptive teaching system
Collect students' learning behavior data, and make diagnosis and analysis based on artificial intelligence and big data technology to help teachers better understand students' learning situation and provide reference for their next teaching activities. At the same time, recommend suitable learning materials for students after class, and provide them with problem-solving ideas when they encounter difficulties, so as to achieve the effect of personalized learning. Article source: Robot
In addition, the development of artificial intelligence has raised a new question for educators: What kind of education do we need in the era of artificial intelligence? Professor Noriko Arai of Japan expressed his concern: "A robot without reading and understanding ability is actually better than most high school students, and most students just cram for the Buddha's feet and don't really understand knowledge. Comparatively speaking, AI does better in memory, so we need new education. "