By challenging the 20 17 college entrance examination Beijing liberal arts mathematics volume and the national liberal arts mathematics volume 2, it took 22 minutes and 10 minutes respectively, and scored 105 and 100 respectively.
Hu, one of the leaders of the artificial intelligence project of the National 863 Program and the president of the research institute, said that AI-MATHS has reached a high level and will be applied to auxiliary mathematics teaching in the future.
AI-MATHS is an artificial intelligence system in Tsinghua University, which relies on big data, artificial intelligence and natural language recognition and takes automatic problem solving technology as its core.
Before the exam on that day, the expert group confirmed through technical verification that the exam was conducted under the condition of completely cutting off the question bank, disconnecting the network, no interference and only 12 servers. 18: 28, the staff took the Beijing liberal arts mathematics volume as the first set of questions input system. Only 22 minutes later, AI-MATHS completed the answer sheet with the score of 105. Then I entered the second volume of the national general volume again. The system was completed in only 10 minutes, and the score was 100.
"The loss of points is mainly in the application problem, and it may be that I don't understand the meaning of the problem." Qi Zu Hai, a math teacher from Chengdu No.7 Middle School, is one of the three judges. He said that AI-MATHS is slightly different from students in solving some big problems. "There is a three-dimensional geometry problem that does not follow the routine and has a clever solution; There is a question about probability statistics, in which there is a sentence' the number of teachers is twice that of boys', which may not be well understood. Overall, it has reached the level of middle school students in the class. "
To Qi's surprise, some proofs of AI-MATHS got scores. "Although its result was miscalculated, the correct problem-solving steps in front got points, probably because the knowledge points behind were used wrongly."
On 20 15, the core team of AI-MATHS participated in the National 863 Project "Key Technologies and Systems of Humanoid Intelligence Based on Big Data" led by Iflytek, and led the sub-project "Key Technologies and Systems for Solving Elementary Mathematics Problems". The results of this public beta will also be an important indicator of the phased evaluation of this project.
"This is the first time that artificial intelligence has publicly challenged the college entrance examination questions. The goal of our challenge is to test artificial intelligence by testing human intelligence, and the result seems to be ideal. " Nonsense, "cognition" is the difficulty of artificial intelligence at present "Let the robot realize that' if you eat an apple, the apple is gone'. This is cognition. The choice of mathematics college entrance examination is also because of the comparison of Chinese, geography and other subjects. The language system of mathematics is relatively closed, which is convenient for artificial intelligence cognition. "
"At this stage, mathematical problem-solving robots are mainly used to assist teachers in teaching and analyze students' learning situation. You can push knowledge points for each student's weak links to help students bid farewell to the' sea of questions'. "Nonsense, before this reference, the learning volume of AI-MATHS was only about 500 sets of test questions, and the practice volume was actually only half that of ordinary high school students.
At present, the AI-MATHS team has developed a series of AI educational products, such as the "Vision Intelligence Assessment System", which can automatically read subjective questions. Among them, the "vision intelligence evaluation system" is a smart pen and artificial intelligence system in the background. When students use this pen to answer questions, the system can collect the data of students' answers in the whole process and "portrait" the students' learning ability.
Due to the flexibility and expansibility of robot models and algorithms, AI-MATHS can be transplanted and developed in many fields such as finance, medical care and transportation in the future.