The philosophy of artificial intelligence based on words--Status quo, predicament, thinking and development

Do you remember the future world AI technology depicted in the film "she"? Anyone can communicate with the intelligent operating system through natural language, she is like a real person but no entity. Language-based interactive technology is indeed beautiful and tempting, but it is also a "holy grail" problem-beautiful but difficult to achieve.

Brain vs Computer, re-examine the human brain　　

It is really a difficult problem to make computers understand natural language and text as human beings can. Looking at the current AI technology, the intelligence behind it is artificial. However, people are too optimistic about computer intelligence, the real computer is far from the realization of intelligence. People's computer intelligence always exists such a misunderstanding: computer computing speed faster and more storage, so the computer is bound to become more and more "smart". But is that really the case? Computers faster than the human brain? Not necessarily. The human brain itself is a unique data storage structure that has its natural advantages. While dealing with a simple calculation is far worse than a computer, it has an obvious advantage in dealing with abstract problems, such as the ability of a computer to quickly calculate a mathematical expression to a value, but it does not treat a problem like a human as a mathematical expression, which is enough to suggest that the computer is so vulnerable to the human brain. People's storage is inferior to computer? Not necessarily, the human brain's storage of information is a storage of abstract information. For example, a face, people will remember after seeing, abstract features stored up, when it sees this person again, he can use the abstract information stored to accurately identify it, the brain's storage is a different from the computer storage of abstract fuzzy storage. Although a computer can store all the images and sounds of a person, it may not be able to identify them, because this kind of storage is so simple and rude. The comparison between the two, storage performance is higher than the view. So on the surface of the human brain is far inferior to the computer, and actually the human brain is a real intelligent machine. After the objective understanding of the computer AI, can be responsible for the present, we can let the computer as far as possible to answer the question "correctly", but definitely not really understand. It has not been done in the past and will not be possible for a long time to come.

the dilemma of Chinese room philosophy in artificial intelligence and pan-intelligence mire　　

Now that we can't understand it, let's see what we can do with today's text-based AI technology. First look at how the tall question and answer program is implemented. Now can realize the natural language question and answer program, mostly constructs an input to the output mapping, the program receives the natural language, extracts the keyword, then the database retrieves, returns the correlation degree highest one. I see "Natural language-keywords-search results" This model and search engines are similar, the traditional search engine is "keyword-search results" mode, the optimization is the corresponding process, but in recent years, search giants have entered the field of artificial intelligence, the focus began to do "natural language-keyword" optimization , relying on powerful computing power and storage of massive data, the Giants have made a more intelligent question and answer procedures, for the general question and answer, the program can "correctly" answer, even can handle some tricky flirting. So the question is, if the question-and-answer procedure is considered a black box, it has the ability to respond to natural language, but can it be said that it is intelligent? John Searle, the American philosopher, raised a similar question in the early 1980s--The Chinese room problem, an Englishman with a Magnum bilingual dictionary in a house, and when you talk to him outside the house, you will be surprised to find that the house can understand Chinese. This question was originally proposed to refute the Turing test, but its meaning is not limited to this, the Chinese language room to draw the philosophical dilemma has been the problem of artificial intelligence has to face. The current intelligence, as described in the Chinese room, just stays "looking smart" and the computer just seems to understand natural language.

Yet other text-based AI goes farther than the question-and-answer program-The voice helper, which, in addition to having all the functions of a question-and-answer program, can "understand" the natural language of calling, texting, and scheduling. These features do not need to be retrieved for implementation, but rather a direct response, seemingly smarter than the question-and-answer program looks? For example: You say to Cortana, "9 times 8 equals what?" "He will give an answer immediately." In fact, these such as "to XX call" "Morning xx point do xx" such as natural language are highly standardized, different models directly apply the corresponding process can be, there are some closed questions: There is no, is not it? is also highly standardized. If you ask some open-ended abstract questions, the machine simply can't handle it, because the problem is too abstract and the designer can't abstract out the standardized process, so we have to turn to the question and answer procedure above. Understanding the principles, we can see the writing robot, it will not feel mysterious. The reason why a writing robot can be used to write news is simply that it extracts abstract patterns from the original information to the news, the format of the news, the elements, etc. are standardized. As soon as the initial information is given to the writing robot, the robot writes the news according to this processing mode. From the two examples above, you can see that these intelligent programs are designed to solve a particular problem, so why not design a generic intelligent program (a generic intelligent program) to solve all the problems? In fact, early in the development of artificial intelligence, people have designed a special intelligent program to solve various problems, but when people move toward the general Intelligence program, but found that this goal is like "development of perpetual motive" as absurd. The special intelligent coverage is small, but the production is simple, general intelligence (Universal Intelligence) coverage is large, but the production is very difficult. This shows the philosophical contradiction between specialized intelligence and general intelligence. However, the voice assistant has provided us with a model to solve this contradiction, that is, dedicated intelligence with a certain general intelligence.

philosophy-based approach and language inspiration　　

Above we analyzed two philosophical dilemmas of text-based AI: The Chinese room problem and the ubiquitous intelligence quagmire. Now let's make a bold assumption about what problems should be solved if the computer is to achieve a true understanding of natural language. I think the core is the linguistic problem. Language is not only a technology, but also a profound philosophy. This can be seen from the 20th century philosophical turn of linguistics. For a long time, philosophers have focused on the origin of the world and the understanding of the world, but since entering the 20th century, philosophical studies have turned to linguistics. Now not only computer scientists, philosophers all over the world are trying to decipher the mysteries of language-why language can express all things? The problem is related to AI because computers use 0 and 1 to express the world in essence. For example, 1 means correct 0 for error, which is a description in the real world. Another example is the class in the computer language that describes the real-world classes. All 0 and 1 can be the problem of the computer can be processed, the problem is the language can be table things 0 and 1 can be table? What if you can do that? Perhaps we can get some inspiration from the language, I do not intend to explore this profound philosophical problem, just want to put this issue, for everyone to think about? Sometimes the point of presenting a question is not to solve the problem. 01 The question of language itself is in the midst of a ubiquitous intelligence, and we do not really want to seek a universal intelligence that expresses the whole world, but to some extent achieve a balance. In my opinion, the three abilities necessary to realize a computer's understanding of natural language are:

1, a basic description of the world

2, a basic description of the AC object

3, a basic description of the computer self

Any of the above is a computer difficult to achieve the Arabian Nights, we can not reach, only from the most basic of linguistics. For example, the language can express accurate information, but also has fuzziness and uncertainty, how to deal with the uncertainty of language is a problem, China's Li Deii academician proposed a cloud model to do the uncertainty between qualitative and quantitative conversion, so the computer can express the real world of the concept of uncertainty (presumably, like, about, etc.) , this is a good start.

　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　If 01 can be the world, then why we are not 01, or we ourselves are 01 and we do not know?

AI is still developing　　 ...

While ubiquitous intelligence is in the future, this does not affect our access to AI services. In a limited time to develop a dedicated artificial intelligence services is obviously more than in the unlimited time to develop generic AI to be much smarter, as long as we can in a certain aspect to make the program has a relatively perfect intelligence, for users, and pan-intelligence is no different. So for a long time to come, word-based AI has a high commercial value, and business giants will develop artificial intelligence that answers increasingly "fluid". On the other hand, our exploration of language, in the initial stage, to make a breakthrough in the understanding of natural language, but also need a large number of researchers to do basic research work, after all, it seems that "understanding" does not mean to really understand.

The philosophy of artificial intelligence based on words--Status quo, predicament, thinking and development