Do you know how high the IQ of search engines is?

Intermediary transaction SEO diagnosis Taobao guest Cloud host technology Hall

Author: Liu

1. Differences between machines and the future of human intelligence

On the February 18, 2011, supercomputer "Watson" defeated humanity and stood on the top podium of the race against human intelligence. The famous Futurist Kurzweil believes that information technology is moving towards the singularity of "hyper-human intelligence". When this information singularity arrives in 2045, AI will transcend human intelligence.

But many scientists believe that machine intelligence beyond human intelligence is only a scientific illusion, including MIT's computer science professor Rob Miller Rob Miller, director of the Facebook Artificial Intelligence Laboratory, Yann LeCun, a deep learning specialist, It is believed that machine algorithms do transcend human capabilities in some areas. But in some areas, such as social and cultural cognitive areas, including the constant invention of new words, the dissemination of video of kittens and praise, or to participate in the construction of other cultural phenomena. These are areas where machine algorithms are difficult to match.

In 1950, Turing published the paper entitled "Machine Can think," for the first time, the concept of "machine thinking" was presented. To make an assumption: a person without touching each other in a special way, and the other side of a series of questions and answers, if for a considerable period of time, he can not judge whether the other person or the computer, then it is possible to think that the computer has a similar intelligence, that the computer is capable of thinking. This is the famous "Turing Test" (Turing testing).

Turing tests are often used to test whether a machine has human intelligence, however, Turing test by human factors interference too much, seriously rely on the referee and the subject of the test of subjective judgment, and only to determine whether the machine has human intelligence, but the machine and human intelligence gap and change speed can not be quantitative analysis. As a result, it is often claimed that its procedures are passed through Turing tests, such as 3721.html ">2014 June, a guest professor at the University of Kevin Vauvic in the UK, claiming that a computer software called Yukin Gutman (Eugene goostman) passed the test, But it was soon revealed that their tests were flawed and suspected of fraud.

So is there a way to quantify the level and speed of machine intelligence and compare it with human intelligence, to observe potential risks and to guard against them. Since 2013, we have explored the direction of machine and Internet IQ, and the initial results have been published at the ITQM meeting held on June 6, 2014, and we will briefly explain how to quantify the future battles between machines and human intelligence.

2. Machine and human IQ General test method (M&h IQ)

In Psychological Science, intelligence (FDI) is one of the most difficult concepts to understand. At present there are no more than hundreds of definitions of intelligence, but a basic consensus is that intelligence is the ability of people to understand objective things and use knowledge to solve practical problems.

The measurement of human intelligence, the IQ test, was first proposed by Bunaisimont, and in 1905 a new method for diagnosing abnormal children's intelligence was published, which marked the formal appearance of intelligence tests, compared to the Nasimon Intelligence Scale (Naisimont). L.M Ter Man, a professor at Stanford University in the United States in 1916, will introduce and revise the Nasimon scale, a revised scale called the Stanford scale, which has become one of the most widely circulated standard tests in the world.

Because the current development of machine intelligence is still very imperfect, can not fully apply to the IQ test of people. Almost no machine system can complete the measurement of human IQ, so we need to absorb the advantages of the above IQ test scale, design a new intelligence evaluation system and set up test questions. So that it can be tested at the same time on the machine system and people of different ages. This result will be able to quantitatively see the gap between machine intelligence and human intelligence, through timed testing, can also be found in the machine wisdom and human intelligence gap changes.

Based on the basic cognition that intelligence is the ability of people to understand objective things and use knowledge to solve practical problems, the new machine and human intelligence evaluation system can be established from four aspects, such as knowledge acquisition ability (observation ability), knowledge mastery ability, knowledge innovation ability, knowledge feedback ability (expressive ability), 15 Test points were established from these four aspects, and the Delphi method was used to give them weight.

There are many kinds of programs with machine intelligence, in which the search engine has a strong representative, as one of the most important applications of the Internet, search engine representatives have google,baidu,bing and so on, these search engines are in various ways to continuously improve the intelligence level of the search engine.

For example, from only identifying text to being able to recognize sounds and pictures, introducing "semantic understanding" techniques to try to understand the user's search intent; Therefore, the choice of the search engine IQ test and comparison with humans will have a typical significance.

3. World search engine rankings and the weakness of machine intelligence

According to the machine and the human general Intelligence scale, can establish the machine and the human General IQ test item Bank (because the space question, this article does not carry on the detailed introduction), May 2014 We use this test bank to carry on the test to the world 50 search engines and 3 different ages human groups to obtain the result as follows

Through the test we found that the search engine in the knowledge of the ability to master far more than humans, but in the knowledge of the innovative ability to score 0. Because knowledge innovation is more weighted in general IQ tests (M&h IQ), the overall IQ of a worldwide search engine is much lower than that of human intelligence, and the highest IQ of Google is less than half that of 6-year-old children.

It can also be seen from the above studies that intelligence or intelligence is made up of different factors, such as astronomy, geography, history and other common sense, mathematical computation, and language translation. These aspects of the machine or the Internet has far surpassed the ability of ordinary humans, and even powerful to a frightening point, which is the main reason why Kurzweil's view can deter many people.

But we tend to overlook that, beyond that, intelligence includes guessing predictions, permutations, patterns, discovery, creative creation, disguised deception, and other higher levels of intelligence, in which machines and the Internet perform a basic 0, let alone a comparison with infants.

Looking from the development of science and technology, the low-end intelligence, the machine's ability leaps and bounds, the high-end intelligence, the machine's ability is struggling, has not progressed for several decades, the future wisdom world should be the machine and the human division of labor, the low-end repetitive intelligence is borne by the machine, the high-end creative intelligence is borne by the human. Too much hype over human intelligence, as Facebook's director of artificial intelligence labs Yann LeCun, "Every new wave of artificial intelligence brings a period from blind optimism to irrational end to depression." Futurists are born to be willing to make blind predictions, especially when they are particularly eager for this prediction to come true, perhaps to achieve personal ambitions. ”，

Finally, it is necessary to point out that the machine and the human General IQ test Scale (M&H) still have a lot of room for improvement, but hope this is a starting point, can solve the deficiencies and defects of Turing Test, in order to quantitatively analyze whether machine intelligence can surpass human intelligence to lay a preliminary foundation.