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
For different application scenarios, the actual level of application of artificial intelligence technology in the industry is different. In choosing a division level on the standard of Artificial Intelligence, international famous Artificial Intelligence experts Sandeep Rajani, professor in the Artificial Intelligence: the person or Machine (Artificial Intelligence - Man or Machine), the level of Artificial Intelligence and human ability transverse comparison, divided into four different levels:
The peak level -- has achieved the optimal capability that cannot be surpassed
Beyond the human level -- better than all human beings
Strong human - better than most humans
Weak humans - weaker than most humans
Textbooks on the introduction of computer algorithms and data mining technology in the smart Web algorithm (second edition), professor CEO Chen Yunwen resilience data will be arranged the partition methods of translation and introduction, to do something in this here. In the context of today's era, the level and level of artificial intelligence technology in different applied fields are as follows:
Peak level:
In some of the rules is complete, strategies, less space application scenarios, such as in 19 * board renju, today's computer technology can already be exhaustive right as all possible situations, ensure the case with humans in any game can achieve the optimal solution. Also in the Tic - Tac - Toe (a simple tic-tac-toe game), checkers, and other engineering applications such as data statistics, the computer is already at the peak level.
Beyond the human level:
AlphaGo and KeJie go against, and IBM's deep blue (DeepBlue) beat chess champion Garry kasparov, shows in these complex board games on artificial intelligence has reached beyond the human level. In some specific application fields, such as fingerprint identification and iris recognition, the technology has also been very mature, and has achieved the ability to surpass human beings.
Strong human level:
Some intellectual activities require deep field experience, and the computer is far less competent than a small number of professionals, but better than the general public. In Texas poker, bridge, and other fields, for example, the computer is better than most of the normal human, in some special fields, such as face recognition and in good condition (no bad block, illumination, Angle), good condition of voice recognition (no special regional accent, complex environmental noise), and other fields, the identification of plant species in the fields of flowers, the ability of computer also have reached the human level.
Weak human level:
There are a lot of skills most ordinary human isn't hard to master, such as driving a car, but for the computer system, because what you want the signal acquisition and analysis of data is very complex, it is difficult to achieve normal human level, weak human level. Common areas include writing articles, reading comprehension, and human language translation.
In this introduction, the level of artificial intelligence varies from weak human to peak level. The development level of artificial intelligence in different fields is inconsistent, mainly influenced by three factors, including:
The clarity of the rules and evaluation methods
The more simple and clear, and the computer can quantify the problem of evaluation, the smaller the cost of using artificial intelligence, such as board game, the computer can play a big role.
The more uncertain the problem, the more trouble the computer will encounter. For example, driving vehicles, driving methods and road conditions vary a lot, and there is no very strict and clear "win/negative", "good/bad" driving style. For example, mahjong or poker, there are a lot of randomness factor, even have the luck element, the processing condition is not exactly consistent, at this time will also give artificial intelligence better processing to bring the challenge.
Therefore, the more explicit the rules and the more objective application scenarios, the better the practice of artificial intelligence technology will be. In areas where rules are vague, the human mind and solutions that are commonly used in everyday life will have a better effect.
Factor 2 the frequency of the frequency
Many application problems, which are handled in a typical scenario and handled in a variety of special cases, are very difficult to deal with.
Taking the problem of face recognition as an example, the accuracy of the existing machine learning technology to complete the face detection and recognition has been very high in the case of good face shooting conditions and the correct Angle and no occlusion. In practical application, expect to encounter quite a few special cases, such as due to the influence of objective factors such as light, Angle, plus the photographed makeup, wear jewelry, partial occlusions, age, change everything, even a small number of cases there is also the photographed by trying to cosmetic surgery or disguised way, deliberately interfere with the computer identification process, the influence of these factors will be very serious the effect of actual use.
There are also a number of problems with self-driving technology, such as normal fine weather and bad weather such as rain and snow, which are very difficult to solve. All kinds of road conditions also pose a great challenge to the practical use of autonomous driving.
Existing a lot of artificial intelligence application, under laboratory conditions many have achieved very good results, but in industrial applications, due to the use conditions than laboratory environment are much more complex and bad, need to deal with all kinds of anomalies and interference factors, so many applications of actual level, still hovering between strong human level and weak human level.
At present, the technology of artificial intelligence, from the laboratory to practical application, needs to overcome many problems, there is still a long way to go. In reality, the pragmatic approach is to limit the specific scenario first and try to eliminate the uncertainty and simplify the problem. In the development of automatic driving technology, for example, if the limit is fixed lines, or within a closed path between the application of the technology can greatly simplify difficult, at this time often can from weak human level promoted 1-2 level, practical level. In the context of text reading comprehension, if you restrict the industry, type and understanding of the text, you can greatly improve the accuracy of the system, so that it can be practical.
Of course, there are advantages to machines from the perspective of "uncertainty". Because people have limited energy, a long focus on one job can lead to a decline in judgment accuracy, and the quality of work can't be maintained for a long time, but the machine doesn't have that problem. The machine can remain stable in a long and stressful work environment, and will not feel "tired" and will not be affected by emotions. In this case, the artificial quality of work would instead be "uncertainty", while the output of the machine would be relatively "certain".
Factor 3 the amount of training data accumulated
As we all know, "big data + algorithmic model = artificial intelligence", it is not hard to see that the key basis of artificial intelligence is big data. As the saying goes, it is only by accumulating vast amounts of training data that artificial intelligence can be promoted upward. AlphaGo beat top human players by accumulating tens of millions of games of go and playing chess data and training them fully.
In the real world, the accumulation of many training data is just beginning. Especially needed to supervised learning "annotation Data (Labeled Data)" accumulation, often need a lot of manual work, the cost is very high, greatly restricted the artificial intelligence level of ascension in the related fields. In addition, there are some in the field of data due to the limitation of some policy factors, such as medical data, or some data are part of the industry monopoly enterprises, all these lead to data flow, the level of artificial intelligence will be slow.
With the rapid decrease in the cost of computer hardware storage in recent years, the gradual popularization of cloud computing and the rapid improvement of the hardware environment of data accumulation work. The awareness of data collection also gradually awakens, hoping that with the push of application demand, more and more data is digitized and recorded, and trained excellent algorithm model to improve the effect.
From the weak human level, to the strong human level, beyond the human level, the summit, a long way. The steps of scientific development are usually easy and difficult to simplify. With the continuous accumulation and progress of technology, we believe that artificial intelligence can replace human beings in more and more application fields to accomplish more and more valuable work.