Absrtact: Since the day the computer was born, many science fiction writers have fantasized that artificial intelligence (AI) can replace the fragile human brain. This does not, the United States AI research leading MIT on the newly developed AI system conceptnet 4 IQ test, the results show that the AI department
Since the day the computer was born, many science fiction writers have imagined that artificial intelligence (AI) can replace the fragile human brain. This does not, the United States AI research leading MIT in the latest development of AI system ConceptNet 4 IQ test, it turns out that the AI system IQ is equivalent to a 4-Year-old IQ, the University of Illinois also said that "AI system IQ is far from the level of 8-year-old children, Still a big gap with the human brain. ”
In the past few decades, no matter how many processors are added to the AI, we find that the human brain is much more complex than the AI system. ConceptNet as a semantic network, designed to impart concepts to the system, contains a lot of information, such as it knows Sax is an instrument, and conceptnet also know the relationship between things, so it knows that sax is widely used in jazz music. Depending on the relationship, it can answer questions that other AI can't answer.
Researchers at the University of Illinois in Chicago later turned conceptnet over to see how its IQ was, using the intelligence tests of Webster's preschool and pupil children, finally getting a 4-year-old test result, and the researchers said It is very worrying which real 4-year-old children are exactly the same IQ as ConceptNet. ”
Distinguish between words and two items conceptnet doing well, such as you ask "apple and banana in common" the question is too good for it, but ask a common sense question "nerve knot", such as "Why do we shake hands"?
Such AI may be able to handle a lot of data, but it cannot use these invisible data relationships to make reasonable judgments. It may know that water freezes at 0 degrees Celsius, but it does not feel cold as humans see ice. Human beings have a lot of common sense, these are accumulated from the small.
There is no easy way to build stealth information and common sense into artificial intelligence, even if IBM's Watson supercomputer cannot solve it. Because common sense is more complicated than finding the right answer, it may take years for us to see artificial intelligence systems that can handle large information databases and derive new ideas for these things.
To this end, MIT continues to develop ConceptNet 5, this will open source and put on the GitHub, perhaps one day we will find common sense is only a certain degree of a natural attribute, perhaps one day the code can copy it, but until we found it on that day, Ai has a long way to go.
Via EX
Related:
What? Google is using the GPU to build artificial intelligence