Original article was published on Artificial Intelligence on Medium
Artificial Intelligence, which we know as AI for short. The more or less the expression of intelligence that is seen in the behavior of every animal in nature, if any part of it appears in a man-made device, we can consider it as AI. The device could be a robot, a factory automatic control system, or even a mobile app.
The main reason that Ai is heard so much nowadays is its use in many mobile apps. Recognizing someone’s face in a photo, showing them the way to a destination quickly on a map, telling them exactly what word they’re looking for as they type, listening to their face instead of typing, making possible short answers to e-mails, all these are the glory and expression of AI. The field of application of AI, however, is much wider — from medical diagnostics to drug manufacturing techniques, from robotic speed control to automotive cars, and even from the predictions of stock market fluctuations to the massive data analysis of astrophysics, this technology has become essential.
The nature of the intellect:
We have different ideas about the nature of the intellect. The most common of these concepts is the ability to solve puzzles or numbers. It is natural to think that we need intelligence for what is difficult. But the capabilities of the computer are different. It can do a lot of calculations in a moment, find the necessary information out of millions of data. If we do this, the day will pass, it will be wrong. Again, we can easily recognize things around us by looking at them, and we can get a lot of ideas about their shape and location. We can communicate perfectly with each other by speaking and listening. All of this is quite difficult for computers. One of the goals of AI is to make these seemingly simple tasks reliable with computers.
Man’s eternal dream is to make a skilled and intelligent robot like himself. The responsibility of directing and controlling the limbs of that robot is again in the hands of that computer. Although we can walk, jump and run effortlessly, doing this with a robot is not easy at all. Despite many complex kinetics theories, computers are still unable to allow robots to walk comfortably on any terrain. Just as humans mastered these abilities as they grew older, so can computers? This is also a huge challenge for AI.
One of the goals of AI is to make seemingly simple tasks reliable for people with computers.
It cannot do the right thing without giving all the instructions to the computer unequivocally. Thus many long and complex calculations can be done accurately and in moments, but can a computer solve puzzles or numbers? How? In the 1960s, Newell, Shaw, and Simon demonstrated what kind of computer program could prove a geometric theorem . Its main components were a few simple rice depending on the problem and their application method. They designed a computer program to see if the problem could be reached by applying these simple moves over and over again at the beginning of the problem. If not effective, it is possible to think of another way back to the previous situation. In this way it is possible to solve quite complex problems with a few simple steps. This is how we solve puzzles or numbers, but not so much, so sometimes it’s too early, sometimes it’s too late, or the solutions don’t match.
Using this framework to solve problems with a few simple tricks, first a Chinese checker, then a computer game of chess is displayed. The game of chess has a lot of tricks, so in this case the use of knowledge about the problem becomes the most important. The main strategy of this knowledge is to try to evaluate the actual situation of the game according to the position of the dice on the chessboard. Added to that is the rapid speed of modern computers and a huge store of memory. Gary Kasparov was finally defeated in 1997 by IBM’s Deep Blue computer-based chess program. It was a moment of great success for AI. However, much of this success is the result of the brutal power of computer calculations and exploration. The judgment and cunning assumptions that the Grand Master makes are still beyond the reach of the computer.
That was the game. This framework can also be applied roughly to solve many real problems — such as integration into calculus, and in any simplification process in general, the layout of pipes in the chemical industry, the design of integrated circuits, the smooth operation of robots, cutting metal sheets or plywood boards most effectively. These are an integral part of any computer-aided design (CAD) package today.
The role of knowledge in solving:
Inspired by Newell, Shaw, and Simon, they named this structure the ‘General Problem Solver’, meaning the way to solve any problem. It turned out that a lot of knowledge is needed to solve the most real problems. There is a lot of practice on how to solve that problem when you apply that knowledge. This new topic is called ‘knowledge representation’ or method of knowledge presentation. ‘Expert systems’ or expert packages are created to present the knowledge of a particular subject in an appropriate way and seek advice from him in different situations.
In many cases the knowledge of the subject would be contained in a number of rules. The structure of this rule is ‘If X is true, then Y can be true or Y’. Problem-solving rules can be applied in place of simple rice in the same structure as before. In the case of a particular application, such as for medical diagnosis or treatment, a number of rules can be made, and effective treatment can be reached by applying the appropriate rules, starting from the patient’s symptoms. In the 1970s, an expert package called Mycenae was very successful in diagnosing blood infections . However, it was not used so much, because if it is wrong, who will take responsibility?
The essence of written language:
Another aspect of AI is reading a written description, understanding its meaning, answering questions. Although computers easily understand programming language, understanding written language is a completely different matter. The meaning of a piece in the language of the program is the same, no matter where it comes from in the program. In written language, in many cases, the meaning of a word or sentence depends on what is being said, that is, what has been said before. So it is especially difficult for computers to understand the meaning of written language.
To do this, first, separate the words and punctuation marks of the sentence and understand its structure according to the grammar. A suitable internal description is made by understanding the purpose-predicate, noun-adjective-verb, etc. of the sentence. This description captures the meaning of each word and the relationship between them. Needless to say, it is often necessary to move forward with the possibility of multiple meanings until it is clear in the light of new information which of these is correct. Another difficult task is to understand the relationship of one sentence to the next and make appropriate changes and additions to the internal description of knowledge. More needed to answer the question is the ability to compose pure sentences from the description of this inner knowledge. At this stage, sentences can also be composed in another language. That will be the translation of the instrument. There has been a lot of research over the years about this method of answering questions by understanding the written language. Success has come in isolated ways, yet this method has not become suitable for use in daily life. On top of this, when the computer has to answer the spoken word and understand it, the possibility of error increases many times over.
Understanding written language is especially difficult for a computer because the meaning of a word or sentence depends on its context.
Another goal of AI is to recognize surrounding objects from camera images or videos and to have an idea of their mutual position. A robot has this special ability, because only then can it plan how it will work. This issue has attracted the attention of researchers since the sixties, and there has been a lot of work in this regard. However, this problem is so difficult that it is difficult to find a solution that will work in all situations.
The computer is given information about the color and brightness of each point of the image obtained from the camera. From this information the computer tries to understand in different ways what is seen in the picture. For this, first the different parts of the picture are separated — for example, in a picture, a book placed on the wall, table, and table can be separated. The levels of color and luminosity in a part remain almost the same or change very slowly; Initially this principle is applied to separate the parts.
Then the most important thing is to discover the boundary lines of the hidden thing in each part. These lines are made up of the points in the image that have the highest brightness change rate. Each of these lines has an imprint on the shape of the object, with which it can be identified. The layout of these lines varies according to the shape and posture of the object. Possible line format information of familiar objects is listed on the computer. Together with him, he understands what is seen in the picture and what his posture is. Of course the matter is not so simple, and the computer often does not understand, or misunderstands. The main reasons for this are getting lost in the middle of a line in an adverse light, the object is partially hidden from the view of the camera, and above all the huge variety of shapes of the object which is difficult to capture in geometric descriptions. Many more ancillary methods are then added to this initial method, such as guessing which direction the different surfaces are facing from the light shadow. It is also easier to recognize how bright or dull the surface is or how grainy it is. But despite this, many things in daily use were not easy for the computer to recognize accurately. Recently the situation has improved a lot with new technology.
The very brief outline of AI that has been given so far can be called traditional AI. It was based on rationalism, that is, by analyzing the workings of each intellect and trying to understand how computers could be given this ability. Although it solves some simple problems in all cases, many real problems remain out of reach. But if we notice a little, it is understood that our activities are often controlled by our subconscious unknowingly. Understanding or recognizing the surroundings, talking, and listening, we never think about it. We will see in the next episode how the neural network brought our subconscious under control.
Newell, A., Shaw, J. C., & Simon, H. A. (1958). Elements of a theory of human problem-solving. Psychological Review, 65(3), 151–166. https://doi.org/10.1037/h0048495