In August this year, Lei feng's network (search for "Lei feng's network", public interest) will be held in Shenzhen, a global influence of unprecedented artificial intelligence and robot Innovation Conference. When Lei feng's network will publish "artificial intelligence and robot Top25 list of innovative enterprises". At present, we are visiting related companies in the area of artificial intelligence, robots, sift through them eventually elected to the list list of companies. If you want to join our list, please contact: 2020@leiphone.com.
Editor's Note: the writer Zhou Jianming, Liu Yu, man-machine and cognition lab from bupt.
Steam engines considered to mark the beginning of modern society, may also see computers as a symbol of contemporary society. Computers are becoming a blind thing, for example, we will not put the phone as the computer, but it is indeed a computer, perhaps some would say, phone is a communication tool, but someone will explain it to you, "network switch" is an interconnected computer, mobile phone is just a Terminal.
When the phone after upgrading to a Smartphone, also into a layer, mobile phones: artificial intelligence? Or a bit more professional, the computer artificial intelligence? This seemingly simple question always makes people scratching their heads, technical explanations of the hard and soft, between the philosophy of common sense, people often do not have a common language.
Firstly, find out why these seemingly simple problems are so difficult, this can only be understood on the basis of "algorithm", "artificial intelligence" and other concepts of the nature and essence.
| The solution
According to the Church-Turing thesis "computer artificial intelligence do? "The problem is equivalent to example problems:
Algorithms artificial intelligence do?
Turing machine: artificial intelligence?
We can be "Yes" and "no" respectively for consideration:
Computers: artificial intelligence?
If the answer is "Yes", you cannot answer the logical next question: why there is different from the computer of "artificial intelligence" of broad ideas and facts?
If the machine to answer "no", then this is different from the artificial intelligence of the computer, what is it?
This questions and answered are requirements on computer and artificial intelligence for basic nature of compared and discriminant, this not simple classification of problem, in store, and home in the can with habits and set, but in production, and research of frontier technology field and strategy decision of cognitive concept Shang, requirements on problem of for nature sex of judge, these problem of proposed and answered involved to algorithm, and computer, and figure inspiration, and Church-figure spirit topics, and "judge problem", and "figure spirit test", basic theory, actually requirements put can calculation sex theory, Uncertainty theory and artificial intelligence in one more platform to study, research, and that research on the nature of the problem is the scientific theory and philosophy on the most difficult issues.
Like famous of Hilbert tenth problem, figure spirit to "figure inspiration" of structure process full answered has, this problem of proposed and solution is computer times progress of deep power, now, on computer and artificial intelligence of basic nature and relationship of judge also will has deep of effect, in this height Shang understanding, "figure spirit test" on has Hilbert-figure inspiration of "judge problem" similar of meaning and value, paper to intelligent philosophy of angle proposed and analysis these problem.
| Computers and algorithms: the Church-Turing Thesis
Calculation tool is one of the most special tool of all human, one of the oldest and most fashionable, its physical and abstract form of tightly-wound, was only a notion of intuition: procedure for calculating mechanical, and now there is a Church-Turing Thesis: the computer is "algorithm".
When you were a few fingers when you are using algorithms, even though today we have computers in place of fingers, the nature of the algorithm does not change, and we are still difficult to answer "what is the algorithm? "This kind of problem. People think mathematician and scientists are is most smart of people, more alone is great of mathematician and scientists, they among on this problem most smart of answered is that Church-figure spirit topics, this seems to some "Emperor of new clothing" of feel, we are only see computer, we not know what is algorithm, dang you number finger Shi, is in with algorithm, but finger not algorithm, now we can said, software is algorithm, but and machine Church?
"Artificial intelligence", we only ask, the machines in the "smart" algorithm for smart in the head is it? A critical nature of these issues, "algorithms and artificial intelligence is based on grammatical or semantic-based? "Is the philosophy, logic, Linguistics, computer and common basic issues in the fields of artificial intelligence, the most puzzling is that physical form and abstract forms tightly wound each other, layers of interwoven. Church-Turing Thesis from this intuitive nature: we know the Emperor wearing the robes, but we can't see.
| The famous "Hilbert's tenth problem"
Concepts we use to define an object, you can use the "extension" or "content" approach, simply put, the former is for example: what is a "computable" thing? The latter is logic: what is the "computability"? The former easily, which is very difficult. Mathematicians used the former approach define "algorithm", specialized in mathematics there is a classic theory--recursive function theory, a recursive function is the "computable" functions, this extension "precise" definition of "algorithm", that is, the "algorithm" means "computable". Perhaps everyone dissatisfied about this, great mathematician is Hilbert put questions from a different angle: can "invention" (To devise a process) to decide, any Diophantine equation is calculated?
This is known as "Hilbert's tenth problem".
Figure spirit understanding this problem of real requirements, since mathematics in the of handed return function cannot directly answered what is algorithm such of problem, that on have with non-mathematics of method to performance algorithm, figure spirit imitation people of algorithm process, created has "figure inspiration", figure inspiration maximum of meaning is show has algorithm of mechanical process nature, put mathematics meaning Shang of "can calculation of" expression for "can calculation of" process, reveals has algorithm of "can line sex" nature, so "Church-figure spirit topics" said, "can line can calculation of" Is Turing Computable.
With the Turing machine, Turing answered Hilbert's tenth problem, but not "answer" a definitive answer, but unreasonably refuse: this issue is not judge! Hilbert's tenth problem and Turing-refused to answer, now collectively referred to as "decision".
This problem of questions and answered way Zhijian of relationship too turn has, later people simplified has on this problem of understanding, put "judge problem" (Entscheidungsproblem) into has "downtime problem" (the Halting problem), but this explained is turn has another big bent, used logic paradox of way, design a Taiwan computer of self refers to involved, that let a Taiwan computer to calculation himself, from intuition Shang on can understanding, this is impossible of.
But the "decision problem", "stop" interpretation method has led to an even bigger trouble if a computer cannot decide whether "computable", it may be overturned "computability" concept, the "halting problem" means-"computable" machine was not sure his "computability". That is, the "halting problem" This paradoxical interpretation of "problem" only push the problem to the cost of a deeper wound level.
In fact, Turing the "decision problem" refused to answer, is based on a "Turing machine" on the construction process, algorithm for calculating machine can be built--that's "Turing", Turing said, with General "verdict" does not come out of the machine! Hilbert's tenth problem has no solution, it was Turing on Hilbert's tenth problem determination, that is, Turing the "decision problem" denial of human judgment and not get the answer machine.
This is the "halting problem" and "judgment", respectively. "Halting problem" method of interpretation is conceived with a built Turing to judge themselves, although the results you want, but deny the Turing machine itself, although the "decision problem" also denied the algorithm itself.
The "decision problem" and an analysis of the solution to the problem, for us now to answer the question on the epistemology and methodology of a revelation, these are NP theory and philosophy of intelligent base case.
| The "Turing test" (Turing Test TT)
From the "decision problem" in the history of the Enlightenment, we can put "what is artificial intelligence? "Transforming the problem into a problem: how to judge whether a machine intelligence? Actually this issue is also Turing first, was designed by Turing, now called the "Turing test" (Turing Test TT).
But the "Turing test" and "judgment" (Entscheidungsproblem EP) is very different. EP real is requirements to algorithm judge algorithm, TT is is to people to judge algorithm capacity and people of thinking can respectively, actual is artificial intelligence and people of intelligent of discriminant, but judge who is people, is people of standard, so, if for TT developed has a specific of formal of standard, this standard on can by another a Taiwan machine acts as a has, but this also on put TT into has "judge problem". This is the difference between TT and ET.
Because TT is made up of people acting as a judge, so the TT is open, that is, TT is a test mode, people can design specific inspection in this mode, design specific criteria to do different levels of inspection or testing, in this sense, the "Turing test" and "Turing" are similar. Embodiment of the TT does not have a unified standard.
TT mode of greatest significance is judged by their people to show "what is intelligence? ", That is, the ability of people to judge a person's intelligence and machine intelligence are the same and different.
TT is based on a premise that, people can understand was convicted of objects and these objects in between, including judge of symbolic communication, this also means that advance human understanding is one of the basic intelligence, so the TT is based on people's ability to understand human judgments.
Historical understanding, is based on the algorithm ability based on the EP to advance to judgment based on people's ability to understand, only through "what is calculable? "The understanding of the nature of the algorithm, which is through the" what is understanding? "Understanding the essence of intelligent, if the communication between man and machine" understanding ", and this mutual understanding is the standard, then the machine will have the same" smart ". This high degree of comparison of human and artificial intelligence is essential.
Analysis for TT showed that human intelligence is not just "understand" and "understanding ' understands '" therefore, TT did not answer "what is intelligence? "But showed" what is intelligence? "。 In fact, the existing achievements of artificial intelligence "what is artificial intelligence? "But to answer" what is artificial intelligence? "In the sense that these comparisons, the TT is a smart form of philosophy of science, such as Purcell's" Chinese room "is one such famous cases.
So, our understanding of the TT is not limited to how to do the actual testing, a practical machine a machine with performance standards are met, this is in answer to "what is artificial intelligence? "Problems. TT's value lies in providing a general principle of the design of artificial intelligence estimate, as Turing machines are all machine-specific models, can also be said that TT was general models of all the specific testing standard of artificial intelligence. Study of TT actually pointed out: what can we expect AI. So the TT is an open test mode, TT is inherently "uncertainty, NP" in nature, similar to best approximate method of TT (NP-algorithm), through continuous research, design TT specific standards, and advance the understanding of artificial intelligence.
Inspiration from study of TT is: comparison between human intelligence and artificial intelligence cannot be the existence of a general standard. --This can also be understood as "intelligence can determine it? "It refuses to answer the question. If this is the case, that is, the "decision problem" evolved into a premium version in the field of artificial intelligence.
More importantly, we can on the basis of these, further study of "smart" and "understanding" content, this intelligent philosophy may also be one of the biggest contributions.
| Five, "artificial" and "artificial"
The "artificial intelligence" defined part is "artificial" and "artificial" distinction, which is property of, which is the essence of things, which is a concept of the enumeration, which is the "natural" relative nature of issue can only be discussed at the height of the philosophy. MIU MIU iPhone 5 Case
General speaking, non-natural of created real (not involved religious) is "artificial of", machine certainly is "artificial of", so "artificial intelligence" General on is equal to "machine of intelligent", that machine intelligent has artificial sex, this just said, artificial of machine is can imitation people of intelligent of machine; but if put "artificial intelligence" understanding into "artificial of intelligent", is "commitment" has "artificial sex intelligent" and "natural sex of intelligent" of a species with quality sex has, this understanding is led to "strong artificial intelligence" or " Super artificial intelligence "basic causes of the concept.
For example, "robots" is "artificial person" or "people's machines (tool)"? The former seems to be "human machine" inversion is essential; the latter is a tautology: man = manual.
"Artificial intelligence" this term not only in "smart" this concept is uncertain, in the "manual" on the modification of the phrase is uncertain. "Artificial nature" against "natural", not "artificial" integrated, and these relationships are essentially the traditional philosophy of "universal" and "nominalism", "realism" and "skepticism" untied the knot again.
So when we asked and answered "computer artificial intelligence do? "Far from as simple as looking, for example, can be understood as" a computer is a machine intelligence? "This and" algorithm machines ' thinking '? "The same but if understood as" computer artificial intelligence? "The meaning is equivalent to" algorithm is artificial intelligence? "These different levels of understanding of the different, often result in against each other," different planets "unnecessary controversy, made the problem more difficult, that's no clue these seemingly simple question has become the most controversial reasons.
"Artificial intelligence" nature of the concept of uncertainty, so there is a general sense of "artificial intelligence" a precise definition, TT standard can only be an open topic, in this sense, close to the human nature is just a series of artificial intelligence development.
| Six, how to say "computer artificial intelligence do? "
Based on the above, in the case of the competency standards of the machine, the algorithm only mechanical steps, not in excess of their ability "spontaneously" invention, creation or learning, specific computer capacity is ultimately determined by the manufacturers and programmers. In this standard, the computer is not "artificial intelligence", not "artificial intelligence". (According to our interpretation of the Agent, the computer is not the Agent. ）
Note that in a computer model based on ANN's "machine learning" method is the "artificial neural network" function, algorithm of the results of this research method is a mixture of both nature, in this case, not to be confused with answers "computer artificial intelligence do? "This kind of problem to sort out this situation, it is one of the purposes of this article.
Can now be simply answered "computers can think? "Problem: the computer can" like "thought--but only" like "people" think "and nothing more, that is, the computer (algorithm) does not equal the people" think ", it does not mean that" artificial intelligence ", it does not mean that people" think ".
| Seven, of artificial intelligence developed two complementary routes MIU MIU iPhone 5 Case
"Artificial intelligence" the term was first presented at the 1956 Dartmouth Conference and is widely accepted, after Turing's 1950 paper was entitled "computers and intelligent" (Computing Machinery and Intelligence), but this problem is, "can machines think?" (Can machines think?) like he 1936 papers in the by do of as, figure spirit avoid has on "machine" and "thinking" of directly defined, but design has "imitation game", is TT, in people of General of concept in the, "thinking" is intelligent process, so TT also can called "intelligence test", because TT is test machine of "thinking" capacity, "thinking" than "intelligent" a word more to separation has "intelligent" this concept in the of uncertainty of part, Similar had also been using "artificial thinking" (Artificial Thinking) this term, so computers as in the TT "machine", but this does not preclude the TT as the evaluation methods of artificial intelligence, TT is an open mode.
Artificial intelligence research field has been exists "algorithm calculation" and "Agent agent (calculation)" two species route, on this phenomenon of awareness and debate has into has artificial intelligence of philosophy problem [1], they in relative of meaning Shang has various call, as symbol doctrine, and function doctrine, and logic or program doctrine, and joins doctrine and so on, up special Mao Republika Conference Shang proposed of "artificial intelligence" this concept General including has this two aspects of work and was everyone accept, but if distinguish has algorithm calculation and artificial intelligence, " Artificial intelligence "concept more Minsky says Agent (agent) the significance of Agent is the narrow, strict" artificial intelligence ".
But as computer modeling of the mixed nature of the artificial intelligence research based on ANN, chip and Internet based research which algorithm of ANN, always in a mixed development, but theoretically distinguish two principle to guide the development of artificial intelligence is very important.
For those impatient verbose form of people, a crude yet essential definition is acceptable: computers are tools of human minds, artificial intelligence is the intelligence of people of the agent (Agent).
See:
[1] the philosophy of artificial intelligence, and Margaret annboden series, Liu Xirui, translated, Shanghai translation Publishing House, 2001.
[2] Zhou Jianming, smart philosophy: human and artificial intelligence, network
[3] Liu Yu, Zhou Jianming, machines with "learning"--looking for ghosts in artificial intelligence, network
[4] Zhou Jianming Liu Yu, man and machine "war" and "learning", net
[5] Liu Yu, uncertain and confused with NP theory
Lei Feng network Note: reprints please contact us for authorization, and indicate the source and author, not deletion of the article.
No comments:
Post a Comment