Without the breakthrough of fundamental theory, not true can that maths of the mystery of unlock intelligence is equational, so called artificial intelligence can be in only little job gets developing on model type, and drive in the car on Fan Shi of this kind of big job, still very it's hard to say relies on chart.
◎ " car person " reporter Zhang Heng
It is suppose in high speed travel, vehicle Buddhist templeput on the brakes did not live, your ahead is case of one lot paper at the same time, across is a telegraph pole, situation is urgent under, where can you change steering wheel to? Look this is a very idiotic problem, but spoil things by excessive enthusiasm in automatic road-sense today, have discussion value very much however.
Of echo what other says, just do not eat melon masses, in industry and commerce actually also different of it doesn't matter. By the spring tide of epicycle artificial intelligence that alpha dog causes, sweep across the whole world quickly. In automobile industry, no matter be truckload,the enterprise still is Internet technology company, nobody do not talk artificial intelligence and drive automatically, the enterprise must be added in propagandist page " AI " or be " wisdom " word. All sorts of trade plenary meeting, the development that each old man is pair of artificial intelligence more speaks with fervor and assurance, they think the person that dress up oneself into the navigate of intelligent direction. But the dry ration that the fact sees than us however is much.
Before before long " car person " dispatch " Qiandoubai is thrown! Does automatic road-sense go wrong direction? " introduced academic physics, condense big ox of voice physics direction, the Zhang Shoucheng of Stanford university is taught, he thinks current artificial intelligence technology still is in special simple phase, look, "Primary " there still is before " special " . He states present artificial intelligence still is in " copy lays level " , this means people and did not figure out what is basic principle.
The fundamentality that he says is the essence of the thing, namely the primary principle of physics, final expression is formula of a maths normally, be like F=ma. The author thinks, maths is can with demiurgic person communication only language.
Physics is the mother of science, fundamental physics is physical mother, the expression of fundamental physics often is formula of a maths. But the trailblazer that Professor Zhang is not artificial intelligence domain after all, so, the article we might as well the big ox that will see artificial intelligence group, los angeles of American California university separates the point of view that Zhu Songchun of school computer vision, acknowledge, study and director of own robot center teachs. Professor Zhu ever asked vessel horse Er 3 times award (one of highest honor of computer vision domain) , hold the position of domain of American vision, cognitive science, AI twice interdisciplinary controller of cooperative project MURI. In a word, this elder, who don't have not to know Zhu Songchun's name.
Still leave piece example, with present automatic road-sense, if among high speed travel drive automatically car, ahead is case of one lot paper at the same time, across is pole of electrical wiring of a cement, buddhist templeput on the brakes does not live car, must bump go up, how can be artificial intelligence chosen? If the person will drive, brain need not be moved, can bear down on paper box at the same time. And if give present artificial intelligence technology, big probability can be tragic.
Why? Professor Zhu Songchun says, in domain of whole artificial intelligence, deepness study is the branch in branch only, although be in now trade special fire. The characteristic of deepness study is to know its not to know its the reason why like that only, will liken just right with parrot, parrot is met parrotry, but what it does not know it says however is what meaning.
Should be clear about above all, if regard deepness study as,be artificial intelligence, was equivalent to regarding Haidian area as complete China. It is core to study artificial intelligence company of direction with deepness study for the most part at present, also include the magnate of a few big Internet of our country, what its core product depends on is deepness study. Without what deepness study itself is actually marvelous, its essence is the polynomial regression in statistical maths. This, have early statistical made known his position publicly with the expert of mathematical respect. This is a stupid method, but the GPU product that move flower Wei amounts to have the aid of also can be used really. But the problem depends on deepness study, the first, cannot understand a thing between causal; The 2nd, cannot draw inferences about other cases from one instance; The 3rd, not most physical common sense of the foundation. Because do not have physical common sense, so it does not know bump in the meeting on the case to be gotten than bumping in the safety on electrical wiring pole much, the signal that it can send through radar only thinks this is two fraise just, cannot analyse its physics density.
We ever also had said, what deepness study needs most is the study data in dangerous environment, from where do these data search? Because cannot understand its quite causal, it is so in road condition, it can be not judged with the common sense of human driver, if whether others misses overtake other cars on the road, want and the line perhaps hides cart these circumstances. As a result of cannot draw inferences about other cases from one instance, learn information and cannot assure what accident place brings to be in quite the next time not on condition of go the same way, machine can make correct judgement.
Professor Zhu Songchun thinks, a very crucial place, it is to be below the frame of probability statistic likewise namely, the deepness study method of current much belongs to to be called " Fan Shi of big data, little job " , train specific mode namely with much data, if alpha dog plays go. But what Professor Zhu advocates is a contrary train of thought, he thinks the development of artificial intelligence needs to enter " Fan Shi of little data, big job " , it is to use many job, is not much data will model intelligent system and model.
Professor Zhu made a figure of speech, the kitchen of a 18 centuries and contemporary kitchen, as the mankind for, a function that knows them is same. But for the artificial intelligence that serves as deepness study, both news is matched extremely hard. Similar, we know, be in container or water can get over there faucet, but the machine can think here does not have water.
Artificial intelligence is not mysterious, of course it can go up in a very specific issue, namely preamble place says " little job Fan Shi " , if play go, handle a character, choose a picture, the language translates a respect to exceed the mankind. But the car drives is to belong to big job Fan Shi apparently, use same method, be to go wrong direction? Pardonable also drive automatically nowadays the traffic accident below always is incessant Yu Er. Cannot accomplish draw inferences about other cases from one instance, does that drive automatically to you can go out to bring security all right to people really?
Cite the word of an investor: Develop the current situation with current technology for, artificial intelligence turns commercial application from dimensions, too distant still. Development of so called high speed wants immediately by card bogus. Must not say this a few annual meeting have old technique to break through, present technical research has on the theoretical foundation in before last centuries half leaf. Theoretical foundation did not upgrade, wanting to be able to have in patch up of soft hardware level overturn, cheat him mankind or cheat artificial intelligence?
This and the point of view that Zhang Shoucheng teachs are very similar. Without the breakthrough of fundamental theory, not true can that maths of the mystery of unlock intelligence is equational, so called artificial intelligence can be in only little job gets develop and applied on model type, and drive in the car on Fan Shi of this kind of big job, still very it's hard to say relies on chart. (" car person " reporter / Zhang Heng, the picture comes from a network) [copyright states] the article is " car person " achieve a manuscript solely formerly, copyright is " car person " all.
