science_logo_aastik_bhattAUTHOR’S NOTE: Since last two and a half years, I have been writing my second novel, which is primarily a science-fiction story. When I had first started it, there were many different concepts in my mind; but after making several initial drafts, I eventually picked one specific subject to take as the base of the story. However, although I discarded almost all of my initial drafts, some interesting concepts I’d come across never left my attention. Hence, along with working over the second novel, I also kept continue my study over certain important subjects. And after two years of hardworking efforts, I have finally completed my analysis and theories upon one of those study topics─ “Artificial Intelligence”. Now because I have no definite plans of writing a novel extensively based upon the concept of artificial intelligence, I came to an opinion that it is quite a better choice to publish my work as a stand-alone research paper like given below, rather than keep it pending in my mind for some future usage in a novel. I also expect to complete my study over certain other such individual topics in future; and if any of those studies will be found worth enough to be considered as a significant research work, I shall certainly publish it here on this website.

 

 

 

A Short Research Paper On

THE DEFINITION OF A TRUE ARTIFICIAL INTELLIGENCE

AND

A GUIDELINE FOR HOW TO ACHIEVE THE DEFINITION IN PRACTICE

 

 

• INTRODUCTION: It has been over a century since mankind first started to give a sincere and serious concern to the possibility of creating an artificial intelligence. This short research paper aims to evaluate all our efforts so far towards the concept of artificial intelligence, and attempts to set a legit and precise definition of a true artificial intelligence; along with a theoretical guideline for how to achieve such in practice.

 

• OUR PROGRESS SO FAR AND ITS SHORTCOMING: In general, the origin of incorporating a task of some human intelligence with an artificial object could be taken as back as the early development of those devices which helped in mathematics, such as abacus; built for a basic purpose of fastening the calculations a human brain can do. However, with the advancements made in the science of mechanics and various technologies, the benign purpose of helping a human mind gradually evolved towards mimicking and challenging a human brain. Even before the first computer was built, several machines with complex designing in their mechanical functions, such as those used during the second world war to convert a plain text into a cryptic text, were proven quite effective to challenge any human mind. But, after the creation of the first generation of computers, and the progressive improvements in their hardware as well as software components, it became evident that a machine can not only mimic or simply challenge a human brain, but it can perform many things ‘better’ than a human brain. Thus, with the creation of highly advanced super-computers during the recent years, many people are led to believe that we have reached very near to, or have almost built some sort of artificial intelligence in practice. Some well acclaimed examples are often mentioned to support this idea; such as a computer which can defeat any human player in the game of chess, or a weather forecast computer which can make the analysis of weather data better than any individual scientist, or a computer which can win over the human opponents in a quiz, or even a photo-editing software which can create or modify a picture better than any real painter. It is precisely at this point, where some misconceptions and the misinterpretations about the term “Artificial Intelligence” in general need to be clarified and be corrected. In other words, none of the above mentioned examples should be referred as artificial intelligence per se; because any such kind of example merely refers to a computer’s ability to process certain calculations or algorithms faster than a human brain, and does not satisfy the definition of a true artificial intelligence as described later in this research paper. To elaborate the statement, we can take an example of a computer with most advanced software designed to play chess. Now such a computer, due to its well-designed software and a super-fast processor chipset, might be able to conquer most human chess players in practice; but, such a computer faces two major drawbacks compared to a human player. The first one being that we can always check the source codes of that computer, along with its various types of log files, and can know precisely what its next move shall be in a game of chess at any given point of time; but for a human mind, we can never predict the next move with absolute certainty. The second drawback is that, if the same software is installed on two different computers, then their outcome moves will be exactly the same when put in identical conditions; which is again a contradiction, because two different human minds can never be expected with certainty to give the same outcome moves in identical conditions. Thus in a broad sense, it can be said from this example that all of our current generation super computers are yet merely the evolved forms of historical abacus, designed only to process the calculations or algorithms faster than a human brain; and all our efforts so far have been restricted towards adding more and more such ability to the computers for faster processing of the codes given to them, rather than giving them an ability of creating the new codes by their own. In other words, the brilliant capacity of processing and storing the data by current generation computers should not be considered or interpreted as a true artificial intelligence. And hence, at the end of all our progress so far, we still need to find the correct answers for basic questions like what exactly should be the legit definition of term “Artificial Intelligence”, and how can we create a true artificial intelligence like a human brain in reality; the satisfactory attempt for which can be as following.

 

• THE DEFINITION OF A TRUE ARTIFICIAL INTELLIGENCE: In the term “Artificial Intelligence”, the word ‘artificial’ should inarguably be stand for something which is man-made and non-living. The criteria of being non-living is essential to distinguish artificial intelligence from any genetically altered or genetically programmed organism that could be made for any purpose. Now, determining the precise criteria for the word ‘intelligence’ is a difficult task, because intelligence is more of subjective in nature than being firmly objective. But for the purpose of defining the artificial intelligence, we must establish some distinct criteria that can be precise in nature and can be confirmed objectively. To determine such, we can take a simple example of comparison between a two years old average human child and a robot designed with the most recent advancements accomplished by current technologies. Now based upon the common understandings prevalent in our society, a two years old child should be called too young to have any ‘intelligence’. His own body movements as well as responses to any verbal command will be uncoordinated, random, and very immature in a general sense. On the other hand, the robot in our example is built with an advanced computer chipset and memory drive, as well as other highly sophisticated hardware features such as microphones, light sensors, proximity sensors, gyroscope and magnetometer, GPS receiver and so on, along with a well-designed software to manage all its components and processes. Hence, compared to a human child, that robot will certainly be more coordinated in its movements and will follow a verbal or visual command very precisely; and in practice it can actually perform its given function quite better than even any adult human. However, such a robot will still be bound by those two basic limitations mentioned in earlier example of a chess playing computer. Furthermore, if we do a systemic analysis of its various shortcomings compared to a two years old human child, we can reach very near to a conclusion for our purpose of establishing the definition parameter of ‘intelligence’. Such as, if we assume that the software in our robot was not provided any kind of programming codes regarding what to do with a pyramid shaped object, and if a prism is kept in front of that robot, then the robot will simply do nothing with it at all. But if the same prism is kept in front of the child, he may take a hold of it with his immature grip at any point of time, and may even be able to put it down or throw it with his uncoordinated movements. It is due to the fact that in case of the robot, both the processing of input data and its output feedback are “pre-determined” in the first place, whereas a human child is simply “unpredictable” since the beginning. Hence, even if we redesign the software of such a robot with better programming about how to deal with an unspecified object, there will still be countless situations in which that robot will be completely unable to function; while a human mind can always keep continue its functions under any circumstances. In other words, a human mind can ‘create’ new codes for any situation, but our current generation computers can only ‘process’ the codes given to them for each situation. Now if we put under the research this ability of the human brain and take it to a very primitive stage of being just a newborn, it can be observed that even in lack of any highly intelligent functions, a human brain has the distinct feature of being “unpredictable” since the beginning. Thus it can be said that, because of having this distinct ability of being unpredictable in the first place, a human brain can acquire the new functions along the course of time, and it is the basic reason why a human brain is able to function under any circumstance. In other words, because of this characteristic of being unpredictable, a human brain can actually ‘learn’ the new things in a real sense and can function under any situation; while a current generation computer can only ‘obey’ for its given situations, and can never learn by its own like a human mind. But, if we manage to incorporate this feature of being unpredictable into a computer or a machine, it can theoretically function like a human brain, and it can be called of being an artificial intelligence in a real sense. In other words, by determining the existence of this feature of being unpredictable, we can now provide the objective criteria for the word ‘intelligence’ in our purpose of defining the term “artificial intelligence” in a precise manner.

Thus, based upon this research, the legitimate definition of a true artificial intelligence can be established as following─ “Any manmade and non-living object, machine, entity, or system which has the characteristic in its function, or its outcome, or in its processing of being unpredictable at a given point of time.”

 

• A GUIDELINE FOR HOW TO ACHIEVE THE TRUE ARTIFICIAL INTELLIGENCE IN PRACTICE: The above mentioned definition of artificial intelligence clearly states that it must have the characteristic of being unpredictable; but in comparison to a human brain, it is obviously not possible for a computer or a machine to spontaneously create unpredictable functions or processing codes by its own in first place. In other words, a human brain is able to start a thought process in complete absence of any external stimulus, but it is impossible for a computer to generate the program codes out of nowhere by its own. However, this truth can be overcome in practice by the fact that the purpose of an artificial intelligence is not to come in “existence” out of nowhere (unlike a genetically created intelligent species), but it is to “function” or process in an unpredictable way. Thus even though a computer is given the program codes created by a human in first place, if we manage to incorporate an unpredictable variable in the program codes, then the processing of those program codes will become unpredictable as a whole, and we can achieve the definition of a true artificial intelligence.

To elaborate this point in a progressive way, we can take a hypothetical example of a robot computer which is placed at an art museum, and is given a simple duty to greet the every visitor that comes in with a random number. Now, with our current standard computers, it is impossible for that robot to create a random number that is unpredictable; because despite any immense speed of the computer’s processor chip, we can always check the source codes stored in its memory and can predict that when 8000th visitor comes in, he will be greeted with what number. But, if we succeed to incorporate some unpredictable variable in its program codes, then the output number will also become unpredictable. For example, if a simple equation “5+X” is kept as the basic program of that robot, where ‘X’ is derived from an unpredictable and variable source, then the output greeting number by the robot will also become unpredictable. Now how can we establish the source for unpredictable ‘X’ in the equation is explained later on in this research paper, but continuing with our current example we can conclude that a small program made of equation “5+X” completes the establishment of first generation of artificial intelligence theoretically. It is described as first generation, because despite achieving the criteria of being unpredictable in its output, the computer is yet completely ‘dependent’ on the original program “5+X” created by a human and will stay dependent as long as its function is bound by that rigid equation. But, if the programming of that computer is taken to a further complex level, then the limitation of being bound by a rigid code can be overcome. Like in our example, when the first visitor to the art museum is greeted with the number “5+X”, it is stored as ‘Y’; that is, “5+X=Y”. Now the next visitor is greeted with “Y+X” number, which is stored as Y1; that is, “Y1=X+Y”. And then the next visitor is greeted with “Y1+X” number, which is stored as Y2; where “Y2=X+Y1”. We can keep this continue as “X+Y2=Y3” and “X+Y3=Y4” and so on, and thus the robot can greet the each visitor with unpredictable numbers continuing as Y, Y1, Y2, Y3 and so on. Here, of course ‘X’ is derived from the unpredictable source as before, but the difference is that the computer is no longer strictly dependent on the rigid equation “5+X” to keep continue its function. This should be called the second generation of artificial intelligence. Now, we can take this example to a furthermore complex level, that if computer keeps storing the greeting numbers in an ascending order as Y1, Y2, Y3 and so on, and when it reaches Yn where the value of ‘n’ becomes same as original ‘Y’ in initial equation “5+X=Y”, that is when “n=Y”, we can program the computer to remove all the old stored numbers and continue to give the greeting numbers based on a new series starting as “n+X”, which is stored as ‘Z’; that is “n+X=Z”. Of course it must be mentioned that in this example the random value of ‘X’ is never negative or infinite and is always a complete round figure. Now, the computer of our robot can keep continue giving the greeting numbers as ‘Z1’, where “Z1=X+Z”, and then Z2, where “Z2=X+Z1” and so on until it reaches Zm, where “m=Z”, and the whole process is repeated again with new equation “m+X”. The important difference here is that now the computer is not only no longer dependent on a rigid equation “5+X” like a first generation artificial intelligence, but also unlike second generation artificial intelligence, it doesn’t even require to keep the original equation “5+X=Y” as a permanent baseline to continue generating outputs Y1, Y2, Y3 etc. In other words, the third generation artificial intelligence would require the human effort of programming only to “start” its function or process; but later on at a point of time, like when “n=Y” in our example, the third generation artificial intelligence becomes capable of editing and making its own program codes, and the original codes do not remain of a rigid necessity to them. Thus with the third generation of artificial intelligence, after we provide the basic codes to a computer for initiation of its functions, we can expect that computer to create the new unpredictable codes by its own, making both its output and its processing unpredictable and independent on its own. This end example not only states that all the definition criteria of an artificial intelligence can be achieved theoretically, but also signifies that we can create an artificial intelligence functioning much like a human brain in a real sense.

Now to achieve such reality in practice, there are two major difficulties; the first one being how to provide an unpredictable variable, and the second one being how to make advanced programming codes incorporating that unpredictable variable. To obtain an unpredictable variable, we must find something or some system which keeps changing its value with the course of time, and that at any given point of time it is impossible to determine its value. In general, almost all the objects and events in this universe are always predictable, because they are bound by different laws of science, and can be determined by various established mathematical formulas; like in any chemical reaction between two or more substances, where despite the value of each substance can be constantly changing during the course of reaction, we can very easily calculate its rate and can predict its outcome by laws of chemistry. The same could be said for almost all other objects and events which are bound and can be determined by different laws of physics, and chemistry, and mathematics. Thus, it is easy to find anything that keeps changing its value, but it is almost impossible to find something that could not be determined by laws of science. But for our purpose in the artificial intelligence, we can achieve its functions if we manage to find something which is constantly changing and is nearly impossible to be determined at a given point of time even by the established laws of science. An example for this purpose can be described by a hypothetical cube of 1cm×1cm×1cm size, made of an inert material. Now that cube is filled with 10million molecules of argon gas, and all the six sides of the cube are closed and are impermeable. Hence at a temperature of 400kelvin, all the molecules of argon gas will be in constant motion due to their thermodynamic condition. Now, practically it is nearly impossible to determine the position and velocity of each single argon molecule in the cube, because all the 10million molecules are constantly in random collisions with each other. Thus at a given point of time, if we take in consideration the volume of a small and hypothetical cube of 1mm×1mm×1mm size in the center of our main 1cm3 cube, then it is nearly impossible to predict how many exact molecules will be there within the boundaries of that small 1mm3 cube. In other words, we can determine the exact number of argon molecules in the small 1mm3 cube at a given point of time by sophisticated sensors and measurements, but we can not predict it at the same time; thus, we can use that small 1mm3 cube as our source of constantly changing unpredictable value. In future, of course, with proper efforts in the direction inspired from this hypothetical example, we can find out more suitable ways to establish the source of unpredictable variable which can be practically applicable with ease and convenience.

Now, the second and the major difficulty lies in making the complex programming codes which can deal with an unpredictable variable to achieve the true artificial intelligence. All our current computer operating systems and their different program languages appear to be incapable of doing it. The basic problem as mentioned earlier, is that despite using very complex and numerous algorithms in their program codes, they are ultimately designed to follow the specific and the predicted path created by their software developers. This is the reason that if a mistake is left in creating the algorithms of a program or even a small spelling or numerical error in the codes by the developer, then the whole program can go failing to work. In such situation, if we employ the third generation artificial intelligence with ability to remove the original program codes and add the new codes by its own in an unpredictable way, it seems impossible for it to function without failing. Thus, we first need to create a completely new and different type of programming which can work with the unpredictable variable without failing. All our current ‘AND’, ‘OR’, ‘NOT’ kind of logics and different algorithm patterns in programming are required to be reconsidered to incorporate one or more unpredictable variable. And after that, we must take that new kind of programming to a very complex level if we want to achieve the artificial intelligence in a real sense. The above mentioned example of a robot greeting the visitors with random numbers is basically a very simple task, but if we intend to create an artificial intelligence with higher functions such as composing a song by its own, then a very complex programming is quite essential. In other words, because the third generation artificial intelligence will have the ability to create the new codes by its own, the more complex we do its programming, the more complex functions it will be able to perform. We can compare this fact with the biological evolution of species found in nature; that if we do a less complex programming of an artificial intelligence, then its abilities to function by modifying and generating the codes will remain below the human level─ somewhat like any lower animal species; which despite having a brain of its own, is not able to ‘think’ like a human brain in a real sense, due to simply the lack of complex neuronal networking in its brain.

 

• THE FINAL STATE OF TRUE ARTIFICIAL INTELLIGENCE: A third generation artificial intelligence should not be considered the end destination we can achieve. It is due to the fact that even if having an ability to delete and to generate the program codes, a third generation artificial intelligence is still bound by some programming instructions. Keeping continue the earlier mentioned example, a third generation artificial intelligence is of course unpredictable in its outcome of equation “Y+X”, as well as unpredictable in its processing that when will it change its code from “Y+X” into “n+X” at any given point of time; but it is still bound by the instruction programming codes that equation “Y+X” is changed into “n+X” only when the equation “n=Y” is true. Thus a third generation artificial intelligence is unable to determine by its own whether deleting, changing, or creating new codes should be performed when “n=Y” is true or when “4n=Y” is true. In other words, by making complex program codes for a third generation artificial intelligence, we can theoretically ‘train’ it to perform many higher functions; but some basic instruction codes will always be there which it will not be able to change by its own. However, this limitation will practically not hinder the functional capacity of a third generation artificial intelligence in future; because if sufficient complexity in the programming codes is achieved making deleting, altering, and generating new codes complex enough, then a third generation artificial intelligence will certainly be able to function like a human brain to a great extent, and the significance of those basic instruction codes will be only to keep the third generation artificial intelligence remain integrated and fail-safe. But theoretically, the final state of the artificial intelligence can be said to have achieved only when there is a complete absence of any kind of strict instruction parameters; that is, a complete absence of a single rigid code. Thus, theoretically the fourth and final generation of artificial intelligence can determine by its own in an unpredictable way whether it wants to follow the condition “n=Y”, or “4n=Y”, or “n+4=Y” and so on to change the equation “Y+X” into “n+X”. A very important problem with such an artificial intelligence will be that in absence of any single rigid instruction parameter, it can reach towards both the infinity and the zero at any point of time. Some other serious difficulties can also be mentioned; like such an artificial intelligence will require a very highly advanced and complex programming, because once it is started, there will be absolutely no way of predicting the continuation of its functions. Also, because the hardware in which the programming codes will be stored and the processor chip etc. can not have the infinite capability, it will further increase the risk of failing of such an artificial intelligence, even despite its programming processes would be functioning at the time of reaching the hardware limitations.

At this moment, in a complete absence of even the first attempt made towards creating a functional third generation artificial intelligence, it is quite difficult to determine how exactly we will be able to achieve the final state artificial intelligence in practice; because it is something like the creation a fully functional human mind from a living brain tissues into a physical electronic circuits in a true sense. However, it is apparent that instead of having multiple codes with a single unpredictable variable, the final state artificial intelligence will certainly require multiple unpredictable variables into each single code. Like in our previously mentioned example, the code “n=Y” can be converted into “An=Y”, making it less rigid; where ‘A’ is also an unpredictable variable. Then if more unpredictable variables such as ‘B’, ‘C’, ‘D’, ‘E’, and so on are incorporated in a progressive way, we can reach towards more and more dynamic complexity to achieve the final state artificial intelligence. Such as “AnB = Y”, “AnB+C = Y”, “AnB+summantion of C = Y” and so on. However, it should be mentioned here that beside the mathematical possibility of reaching infinity, it will also be very difficult to start the functioning of the final state artificial intelligence in the first place; which, in case of the third generation artificial intelligence, can not be considered a serious issue. Thus, it is quite obvious that creating the final state artificial intelligence will remain the burning challenge for mankind in future. However, during the course of time we finally succeed in achieving it, the creation of a fully functional third generation artificial intelligence will also not be anything less than a huge revolution.

 

• CONCLUSION: Based upon all the facts and hypotheses explained in this research paper, it can be concluded that even if it is unclear at this moment whether we will ever be able to make an artificial intelligence functioning independently and exactly like a human brain, it is certain that if we intend to succeed in making a real artificial intelligence ever in future, then the path of our efforts must be based upon the definition criteria as mentioned.

 

- Dr. Aastik Bhatt.

  February 27, 2012.

  Surat, India.

 

Copyright © 2012.

 

To go back the main website, click : http://www.aastikbhatt.com

 

Back to top
Go to bottom
 

Copyright © 2012-2015 by www.aastikbhatt.com and where mentioned by Aastik Bhatt. All rights reserved.