At the same time, in Beijing, Aidam, another artificial intelligence robot, launched a man-machine war with six previous college entrance examination champions. In the end, Aidam's score is only lower than the average score of the top students in the college entrance examination 1 point.
The "First Show" of the College Entrance Examination for Artificial Intelligence Robots: Disconnecting the Internet and the Library
On June 7th 18: 00, a special college entrance examination simulation was held in Tianfu Xinguyuan 10 building in Chengdu High-tech Zone. The candidate is the artificial intelligence robot "AI-MATHS".
AI-MATHS was born on 20 14. It is an artificial intelligence system with automatic problem solving technology as its core, which is developed by Chengdu Zhunxing Yunxue Technology Co., Ltd. relying on Tsinghua University big data, artificial intelligence, natural language recognition and other technologies.
AI-MATHS is a big black box, which is arranged in a completely disconnected room with a cordon around it. There are 1 1 servers in the server black box to solve the problem. In order to ensure the fairness of the exam, no one can operate these servers at the test center. Outside the room, there is a sign "All devices are disconnected from the network segment library".
The so-called network disconnection and database disconnection is to imitate the closed-book exam in the closed environment of candidates; Test the reasoning ability and association ability of complex logic; Cut off all possible external support, including expert system library. This is also the core of artificial intelligence technology-comprehensive logical reasoning ability in the scene of network disconnection and database disconnection.
Small sample training is the requirement of the national 863 class people problem-solving project. Therefore, it can be verified whether the system can independently form and learn complex logical reasoning and mathematical thinking, exercise thinking ability beyond human ability and human-like reasoning ability, and provide cheap technical support for cross-industry application.
The biggest problem of the college entrance examination robot: I can't read the questions.
"AI-MATHS" challenged the Beijing liberal arts mathematics examination and the second national liberal arts mathematics examination, which took 22 minutes and 10 minutes respectively, and scored 105 and 100. This achievement is close to the goal of 1 10 set by researchers.
The specific scores of the two exams are as follows:
Beijing liberal arts mathematics volume, full marks for multiple-choice questions, 20 points for fill-in-the-blank questions, 45 points for solution questions, full volume 105 points;
The national paper contains 55 multiple-choice questions, 20 fill-in-the-blank questions and 25 answers, with a score of 100.
On multiple-choice questions and fill-in-the-blank questions, "AI-MATHS" is very eye-catching. Losing points is mainly to solve problems, and some problems are even 0 points.
This is equivalent to a high school graduate with average grades.
Because the math scores of the college entrance examination have not yet been released, people can't judge what level AI-MATHS is compared with human candidates. However, the on-site assessment of the marking teacher believes that
"I can't understand this problem." The biggest difficulty in developing the college entrance examination robot is to make the system understand human language accurately.
As a parent, Lin Hui, CEO of Chengdu Zhunxing Yunxue Technology Co., Ltd. and director of the Big Data Center of Tsinghua University Suzhou Research Institute, pointed out a weakness of "own children" when answering questions.
Lin Hui said, "It can easily solve the application problems directly expressed in mathematical language, but its biggest weakness is that it can't understand the scene description language in the examination questions and the questions."
Why can't the math robot in the college entrance examination understand the questions? For this question, Lin Hui made an analogy. If the topic is 10- 1=? The artificial intelligence system can easily answer. However, if the topic is described as "Xiaoming has 10 apples, Xiaoli ate one, and how many apples are there?" Faced with such a situation, the artificial intelligence system is "stupid". "It doesn't understand what Xiao Ming is, what Xiao Li is and what eating means."
In fact, in order for "AI-MATHS" to quickly overcome this problem-solving weakness, the research team of Zhusight Cloud Science has previously conducted 500 sets of test papers on "AI-MATHS" and 12000 times of devilish training.
Researchers say that artificial intelligence mathematics needs to be improved in natural language recognition. In the words of the math teacher who participated in the marking, if AI-MATHS wants to do well in math, it must first learn Chinese.
Researchers believe that with more training and study, AI-MATHS will achieve better results in the future.
Six previous college entrance examination champions narrowly beat intelligent robots with 1 point.
At the same time, in Beijing, a man-machine war for college entrance examination also started.
Aidam, an intelligent education robot independently developed by Xuebajun, an intelligent education company in primary and secondary schools, answered the 20 17 liberal arts mathematics test paper for the first time with six top science students in Beijing.
In the end, Aidam completed the answer in 9 minutes and 47 seconds, and the score was 134. The answer time of the top six candidates in the college entrance examination is 55 minutes, with an average score of 135.
According to Zhang, the founder of "Xuebajun", Aidam's answer speed has slowed down in order to facilitate the statement. Usually, the completion time of each problem should be 7~ 15 seconds.
It is understood that "Xuebajun" has invested several years and gathered more than 200 technicians to develop intelligent educational robots. Senior teachers and engineers have built reasoning engines and complex knowledge elements on 70 million problems, and the machine has explored the simplest path of all problem solving.