News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
Along with the 2016 AlphaGo defeating the weiqi champion Li Shishi, the worldwide artificial intelligence is developing rapidly, and AI threat is also widespread. At the same time, intelligent products are booming, and emerge from time to time. Can artificial intelligence eventually transcend human beings? To what extent are the intelligence of these smart products in the end? Answering these questions requires a quantitative approach to testing the extent of the development of the intelligent system.
Since the Turing test was put forward in 1950, scientists have done a lot of work for the evaluation system of artificial intelligence. In 1950, Turing put forward a famous Turing experiment to determine whether a computer could have equal human intelligence by means of questioning and human refereeing. As the test method of artificial intelligence the most widely used, but not the Turing test test of Ai intelligent development, intelligent system can only distinguish the same with human intelligence, but also affected by man-made factors too much interference, depends heavily on the referee and the subjects of the subjective judgment, so people often get a severe situation in the absence of research under the claim procedure after the Turing test,
In 2014, Professor Reed (Mark O. Riedl) of the Georgia Institute of technology of the United States thought that the essence of intelligence was the power of inventions. He designed a test called Lovelace 2. The test range of Lovelace 2 includes the creation of fictions with fictitious stories, poetic creation, oil painting and music.
There are two problems in the quantitative test of artificial intelligence, including Turing test, including: first, these test methods do not constitute a unified intelligent model. Based on this, we stop analyzing and identify multiple classifications of intelligence. Unable to incur various intelligent systems including human unified testing; second is the test method to the quantitative analysis of artificial intelligence, or only one aspect of the quantitative analysis of intelligence, but the system will reach a few percent of human intelligence and human intelligence, the development speed of how to carry out the speed ratio, these problems did not touch on the study.
Aiming at these problems, the research team put forward: according to the purpose of evaluation, intelligence evaluation of intelligent system has three kinds of intelligence quotient, that is, general intelligence, intelligence, intelligence and value intelligence of intelligent system. The theoretical basis of the three IQ, detailed definitions and evaluation methods will be discussed in the following details.
1. theoretical foundation: the standard intelligent system and the extended Von Neumann architecture
The intelligent system includes human and artificial intelligence systems to stop intelligence facing two major challenges: evaluation first, artificial intelligence system currently does not constitute a unified model; second, the comparison between artificial intelligence system and to human beings as the representative of the life is not a unified model.
To solve this problem, in 2014, Chinese Academy of Sciences, fictitious economy and data science research center is also the future of the intelligent laboratory research team members, Liu Feng, Shi Yong, Liu Ying, von Neumann, David Wechsler of human intelligence structure reference model, domain knowledge management system model DIKW. A standardized intelligent model is put forward, which depicts the characteristics and attributes of AI system and human beings, and regards any intelligent body as a system of knowledge acquisition, control, innovation and response.
This separation model and Von Neumann based architecture, can constitute the expansion of Von Neumann architecture, compared to the Von Neumann architecture, this model can increase the innovation function, which can be based on existing knowledge, knowledge discovery of new elements and new rules, make it into the storage device, and controller for computer use, and after the input / output system and external stop learning interaction. The second increase is the external library or cloud storage that can stop learning sharing, and the external storage of the von Neumann architecture is only the service of a single system. The extended Von Neumann architecture will play an important role in building the IQ of AI.
2. definitions of three different IQ in Intelligent Systems
2.1 AI general intelligence quotient
Based on the standard intelligence model, the research team set up the AI IQ test scale, respectively, with the more than 50 AI systems including Google, Siri, Baidu, Bing and 6, 12 and 18 years old people in 2014 and 2016 to stop AI IQ test. According to the test results, the performance of artificial intelligence systems such as Google and Baidu has been greatly improved than two years ago, but it still has a certain gap with 6 year old children.
It should be said that the AI test is carried out to solve the problem of human intelligence can go beyond AI, this research is every intelligent system including the robot, AI software system, human, animal and other biological agents as equal, and observe its nature, its intelligent degree he agent emerge in the interaction in.
The definition of AI universal IQ as follows: Specification for intelligent model based on, in order to solve "the intelligent evaluation system to carry out the extent" problem, the intelligent agent system as a peer, through a unified AI IQ test scale of intelligence evaluation score, can be called the universal IQ Artificial AI system intelligence General intelligence quotient (AI? G? IQ).
2.2. AI's service IQ
In theory, we found that in addition to the emergence of a few AI systems for scientific experiment purposes and not providing supplementary services to humans, most other AI systems are to serve better people.