In Saturday, in the Turing test conducted by the Royal Society, a machine designed by a Russian engineer (more accurately software) was successfully tested to become the first machine ever to pass Turing tests. Many tech media are shouting that AI is about to enter a new era, and they are obviously too optimistic.
What is Turing test?
Turing's test came from a pioneer in computer science and cryptography--Alan Turing wrote a 1965 paper, "Computer and Intelligence" (Computingmachinery and Intelligence), Turing thought: The important criterion of whether a machine can think is to imitate a real person. So Turing devised a set of test rules:
"If a person a uses a language that the test object understands, ask two of objects that he cannot see any string of questions." The object is: one is the normal thinking person B, one is machine C. ”
The Turing test will have multiple testers like a to take turns to "bomb" the test object, Turing set a 30% lower limit-that is, if 30% of the testers think it is impossible to tell whether it is a machine or human, the machine through Turing test.
In this test by the Royal Society, there were 5 contestants. The winner, "Eugene", was identified as "human" by 33% of the testers, and most of the testers believed it was a "13-year-old Ukrainian boy". "Our main design idea is that it can claim to know everything, but because it is age-limited, it's not really known, it's perfectly reasonable," says Vladimir Visilov, a developer of the machine. ”
This message again raises the human fear of the machine, but the machine through Turing Test can really announce the era of human power intelligence is coming? (Beijing)
I don't think so, at least for a short period of time.
Natural language restricts the progress of machines
The computer participating in the Turing test involves several techniques, including natural language, natural language including natural language generation system and natural language understanding system, the former is to convert computer data into natural language, the latter is to convert natural language into computer program more easy to deal with the form. In a word, natural language is an important medium for human and machine communication.
In Turing Test, the way of communication between man and machine is the process of natural language understanding and generation, or more accurately, the way people interact with machines is the process of natural language realization. This is the first step of communication between man and machine, only to let the machine understand the people's view, the latter will be involved in big data, cloud computing and other technologies. And just before the test was "successful," Google product manager Dave Orr admits: "Natural language has yet to be developed and is now in a" ready-to-use "phase, with many Internet companies starting to use natural language to develop something that looks cool, such as Siri, Google now, But if you have to say that the machine can "understand" what you mean, it's a bit reluctant.
How stupid is the machine's understanding of language? There is a classic case that allows the machine to distinguish between the following two sentences: "We give bananas to the monkeys because they are hungry" and "we give the bananas to the monkeys because they are ripe." The structure of the two sentences is the same, although the semantics are different, normal people understand that there is not too much problem. But the machine can not understand, two "they" is enough to let the machine into "meditation", also can not give the correct results.
Dave Orr gave another example to prove that the machine could not really understand human language: "Let the machine try to summarize the gist of an article in a few words", at least so far, no machine can do that. And this time through Turing test machine (software) is after more than 10 years of "careful cultivation", also therefore does not have the ability of large-scale replication.
From this perspective, even if a machine (computer) is tested by Turing, without rapid follow-up of natural language processing and other related technologies, humans will not be able to achieve so-called artificial intelligence for a long time to come. ( Shanghai )
The development of artificial intelligence in the new era?