Thursday, June 26, 2008

Popular on Artificial Intelligence

Of course, originally planned art-workshop on creating autonomous robots. Your humble servant even made a structural outline of such material, but… after I understood that without basic knowledge in the field of artificial intelligence would be difficult for the reader to understand something - or rather, completely penetrate the meaning. Therefore, in a few rooms, we will consider this topic, which is valid not only for robots, but for most modern software.

Artificial intelligence (AI) as a separate area of science has just recently, almost simultaneously with many other current major disciplines - such as cybernetics, neural networks, other branches of basic: medicine, psychology, economics. Meanwhile, most of those who first confronted with this notion, it might seem that artificial intelligence - is an attempt to fully replace the human brain machine. Maybe because it originally anticipated. But, for example, in one of the mistakes than Leonardo da Vinci (if it can be considered a mistake)? He, like many inventors, his predecessors, it is anticipated that flight is possible on condition the establishment apparatus simulator poultry. While actually not a very successful example, because in the modern aviation, we see a clear link with the biological world - underwater. But in a number of books on this particular AI set an example for illustration as a comparison:). We will take another example that you can see on the streets - cars. No biological form on Earth does not have such a mechanism of movement. In doing so, cars and trains are the kinds of high-speed ground transportation. If not invent them, a mock humanoid or other species that we can see examples of fantastic "steam" of people and horses at the end of the XIX century, the optimal such decisions would be minimal. The third example… you dotronulis to obzhigayusche hot FBR. After reading some scientific works, it can be concluded that in this case, you start to analyze the situation in the brain to compute the most optimal solution, perebiraya it from the many options as possible or not. But what makes people? Just otdergivaet hand, not even thinking about their actions. Refleks.

***

The creation of artificial intelligence - is primarily a difficult task, if possible at best shared some simple. In today's presentation of this science are more to use the notion of agents who transform their own perception of their own actions. To this can be attributed boldly and any computer program, and the robot at work. But the recurrence of Rights no concrete action. Among other definitions AI can be found more comprehensive - for example, "Automation of reasonable conduct". Not long ago, namely in about 2003 - year, was published a table comparing the number of neurons of the human brain to the number of transistors in today's PCs. And Moore's law has been applied, under which the number of transistors in the calculation of per unit area increases exactly twice for the period 1,5-2 years. And it applies even more to statistical studies, because this trend has occurred over the past 25 years, if not more. This table published virtually all media, in one way or another connected with high-tech, because finding was remarkable: the number VMRabolu by 2020. But whether this means? ..

In fact, our brains can not be compared with one computer - a whole structured and interconnected network. And in 80 - or 90 x - x could create something similar - for example, one computer is responsible for vision and image analysis, the second - for the hearing and speech recognition, the third - for tactile sensations, the fourth - for chemical analysis, the fifth - for a decisions, etc. This is possible? Of course! But… no algorithms! Just no! Not invented… I do not really transistors in the case. What we offer mathematics, very difficult, in principle, moreover, is different from the natural perception. For example, our auditory system does not use Fourier transformation for the expansion complex sound waves on some simple. It simply does not know and apply the methods of other more universal - for each frequency has a special receptor. Vision works on similar principles, but with its specifics. By the way, to its biological essence of modern technology and scientists are closest. But, returning to the same ears, it is worth noting optimal modern solutions, because the Fourier transform plus computer is much easier creation of a full model of human hearing. Therefore, science II only in private implies repetition rights, and actually more close to programming agents or the same "automation of reasonable conduct". But, on the one hand, mathematically can be done very much. And since we remembered about Fourier, then Jean-Baptiste and surprisingly suited to further narration. Very vivid example of history, when the French scientist tried to decrypt hieroglyphs of ancient Egypt (with Napoleon he was governor of Nizhny Egypt). It must be said that Fourier, addressing this challenge, spears broke and said that linguistics beyond mathematical analysis. He turned and rights, and no rights at the same time. Do not edit because the Egyptian language alphabet issued his disciple, and rights for the reason which has been widely discussed theorists and practice of artificial intelligence and subsequently with what we face now.

Turing Test

Alan Turing was one of the founders of artificial intelligence as a scientific discipline, or rather, not even discipline, and a scientific direction. In principle, it is their studies and work more reminiscent of fiction-practice. That is a very creative person, plus a good scientific basis. Incidentally, under his leadership in 1940 was the first operational computer Heath Robinson (named after the cartoonist). This was electromechanical device that was intended to decrypt messages transmitted Germans. In 1943 - m the same team of scientists created a more powerful computer Colossus, which already used electronic lamps, a range of appointments has become more broad. Also known Tyuring that wrote one of the first algorithms for playing chess and wanted him to realize at the same Colossus, but he is not authorized the British Government. Therefore, instead of machines for the algorithms worked himself on the progress of computation spent on the half hour, and the technology subsequently became known as "paper Turing machine". On the scientist says a lot, so if you choose to refer to specialist literature and we will continue to talk about AI.

So why was pronounced the word "fiction"? The point is that it was a scientific romanticism. Just as Isaac Ayzimov withdrawn rules for robots, robots themselves when, in fact, has not been as such. Similarly, the Turing assumed, as can behave computer in the future, but as can not. Naturally, he tried to equate it to humans, so originally envisaged the possibility of communicating with computers at the usual language, the availability of computer memory (the ability to store machine), automatic formation of logical conclusions and ability to learn. In doing so, he felt the same machine could not (give the full list): "be good, generous, beautiful, friendly, proactive, have a sense of humour, to distinguish truth from falsehood, to commit mistakes, love, to enjoy strawberries and ice cream, make falls in love with themselves , To learn, to use words correctly, be subject to their own thoughts, to show the same diversity in behavior, like people really create anything new ". Yet Turing assumed that the machine will be able to act intellectually, under the implied rather than no ability to think, as an intellectual presence of conduct. He published an article "Computing Machinery and Intelligence", where the proposed test, which later became widely known as "Turing test". The essence of it is that after the five-person communication and programs within the traffic in the first 30% was unable to determine who was with him in touch: a living person or a computer. Alan Turing even predicted that computers will pass this test in 2000. In principle, as you understand, all very, very conditional. For example, several years ago, your humble servant was a living witness to how people communicated with voice mail on your mobile phone, informing the balance in the account. So he called the amount and character made this key phrase: "It is my understanding, but I am hour ago introduced a new amount! When they are listed? "More seriously, much depends on the issues, the level of perception and the conversation situation. Therefore Turing test himself with many now associated with the same laws for robots Ayzimova. They have, but it is not the basic rules. In other words, a scientific show-business.

The Chinese room

It is known mental experiment. In fact, what then occupied the minds of scientists, now it is obvious to any PC user. But to keep ahead of the future had to use imagination, "the paper machines". So John and Sirl suggested purely mental experiment. Imagine a person in a closed room. He understands only English. On the table lay a pile of papers, some of which are empty and the other documents filled with incomprehensible records in Chinese. There is also a reference book-manual in English, which may help in decoding. And now, through a special opening in the room are reported in Chinese with some guides to action - for example, make a new record, document or find a particular character or move the pile from one place to another. Accordingly, the man himself - this processor, a pile of paper - memory book - the program. This is inside. And apparently it looks like sending queries / commands and receiving
answers / results at the Chinese. And if you ask the system itself, whether it knows the Chinese, you get an affirmative answer. The system could even Turing test. But if you ask the same question Englishman or a book - a negative. Thus, we come to a basic problem II, namely the identity and understanding. Sirl then withdrew several laws, axioms that the programmes are the essence of syntax, but wisdom in its full presentation implies semantics or mental content (understanding the meaning of the word). One syntax for semantics is not enough. Incidentally, for the moment questions of semantics remain unsolved, although many can be seen fairly good intermediate realizations, how would precede, but…

Bridge completion

Let us now think: Why did Fourier has rights in his remark? In fact, he meant primarily semantics. The same deciphered ancient languages in the main largely due to the fact that there were analogues in the other known records. Incidentally, there are an outstanding example of the Second World. Americans to ensure that their messages could not decrypt Japanese specialists, attracted representatives… a small Indian tribe, which has its own language. And all of us, respectively, were on it. The Japanese were unable to do anything in this situation. And let us think: Does a need semantics machines, actually? Your humble servant saw many modern algorithms, there is little that some of them worked and actively participated in projects to create them, but almost all of them, in fact, for machines are "Chinese Room". This is where the abstract. In fact, automated semantic search, the appropriate "understanding" computer or robot realizable. And the principles of such algorithms is understandable to our contemporaries, with a friend obektnoorientirovannym programming, high-level languages, yes… even with the notion of classes.

To be continued ...

No comments:

Post a Comment