The history of artificial Intelligence

        Our research into the history of Artificial Intelligence begins by examining the basis of logic as defined by the 5^th century philosopher Aristotle.  Aristotle devised the Syllogistic basis of reasoning, which decomposed any problem to a set of components that could be analyzed factually. Once a factual analysis had been completed, the solution could be deduced from the facts that made up the problem. This led the problem statement to be formulated as a number of premises, which could be proved or disproved based on the number of facts that surrounded the case. Such a method of reasoning gave structure to the mind’s thought processes and can be considered by some to be the birth of logical thought. Logical thought processes, once understood, would later be applied to almost every other problem that required deductive reasoning as noted by the Association for the Advancement of Artificial Intelligence (AAAI). The definition of Artificial Intelligence is “An Algorithm or a set of algorithms that make decisions in a logical way so as to replicate human Intelligence.” The idea that machines could replicate thought was unheard of in the 15^th century. Machines were commonplace during this time period. The crown glory of mechanized creation was the invention of the clock. The clock consisted of many individual components that were interrelated to each other and work in tandem to keep record of an objective source of information. This period was one of the first times in history that people had used a device to keep track of time. The creation of this device led to the ever increasing need to devise more machines that could perform functions that mimicked human thought (AAAI).

                Thought has been a very elusive concept, and still is. It is very difficult to quantify human thought in terms of computation. A computer needs a definite set of inputs, a process or algorithm to act on those inputs, and a finite set of outputs that are generated once the algorithm has done processing. Human thought on the other hand has a unique property of self creation and regeneration. Ideas are abstract to begin with, but the thought processes that drive them cannot be quantified as there is no algorithm of the mind in existence. It is not always necessary to provide input to the mind as ideas can be generated spontaneously. Even if input exists, the outcome can be completely unexpected and hence cannot be predicted using a linear system of logic.

                The 18^th century was a century filled with a plethora of mechanical toys and devices. Since the initial conception of the mechanized time piece, other inventors took suite and devised machines that were both purposeful and aesthetic in nature. All the devices were mechanical and did not possess any sort of computation to solve a problem. One example was Von Kemplen’s mechanical chess player (Lohr, 145). The device itself was a simple one consisting of a box with a compartment that was hidden from the audience. A human operator would sit in the box and control the mechanized chess player who was devised from a complex system of cogs and magnets to mimic human movement. The chess board was illuminated from the bottom where the human operator could see the pieces thorough the translucent board. The choices were made by the human operator, but it appeared as if the robot was choosing where to position the next piece. The device itself was remarkable for its precision in manipulating the pieces, but it was not a true Artificial Intelligence machine.

                The 19^th century gave rise to a proficient logician named George Boole. Boole is still known for his invention of binary algebra which is also known as “Boolean Mathematics” (Wolfram). Boolean algebra gave the mathematician a basis to convert linguistic statements to a series of interrelated true or false symbols that could then be manipulated based on a system of rules to achieve a result. The result would then be converted back to plain English based on the same rules that were used to compute it. Boole chose to use 0 to represent false and 1 to represent a true value. The operations were unary, which also provided a set of symbols for generating switch logic like if…else statements. The value of such a system was objectivity, in which the underlying logic behind the relationships in statements is what needed to be understood and not the subjects under analysis. The subjects were converted to variables that were manipulated using the unary rules of his logic.

 For instance;

false OR false = false       false  AND false = false

false OR true = true        false AND true = false

true OR true = true         true AND true = true

These rules were later on to become the basis of modern binary computation, and are still used today on the component level as binary logic gates in all modern computing devices (All about circuits).

Boolean computation is important when spoken language needs to be quantified. Most statements can be mathematically converted to Boolean logic. There are a set of rules that determine this conversion. Care should be taken when conducting such a conversion as the connectors ‘and’ and ‘or’ are typical sources of error. For instance, the statement, “I always carry an umbrella for when it rains and snows” is an example of the interchangeable use of ‘or’. The use of or in English can also be translated as the logical OR or the exclusive OR (XOR). For instance, “I start to sweat when the humidity or temperature is high” is an example of a logical OR. “You want ice cream and candy? You may have ice cream or candy!” is an example of an exclusive OR.

With the advent of the 19^th century great thinkers like Charles Babbage and Ada Lovelace worked on programmable mechanical calculating machines (HOCH).

The difference engine was accurate to the 32^nd decimal place, which is more accurate than modern day calculators. It was a truly amazing invention. Ada lovelace is considered to the founder of modern scientific computing. Any computing device would be incomplete with a set of instructions that are to be performed to reach a desired outcome. The “software” that ran on Babbage’s difference engine was conceived by Lovelace. She suggested a plan on how to create an algorithm for the machine to calculate the set of Bernoulli numbers (Biographies of women mathematicians).

Lady Ada was also the first one to propose that the machine could not only be used to perform computations using complex algorithms, but could also be used to generate new algorithms of its own (Biographies of women mathematicians).This thought was way ahead of its time and could not be verified, as the computing power available during that period of time was limited. She predicted the future course of computing for many years to come by asserting this supposition.

                In the first half of the 21^st century a number of mathematicians were involved in formalizing the process of computer logic in order to develop application methods for these relatively new devices. The mathematicians Alfred North Whitehead and Bertrand Russell published Principia Mathematica, which revolutionized formal logic (University of Michigan). This treatise was a momentous one as it attempted to derive all the truths of mathematics from well defined axioms and rules of inference taken from symbolic logic. This was a higher level extension to the rules of logic defined by Charles Babbage. The treatise is considered to be one of the most seminal developments in the history of computing (New York Times books).

                During the middle of the 21^st century logicians Warren McCulloch and Walter Pitts published “A logical calculus of the ideas imminent in nervous activity”. This was a very important work as it laid the foundation for neural networks (DLSI).

The idea was to simplify the behavior of individual neurons in the human brain and to develop a logical apparatus based on these observations. They divided the neurons into two groups, input and output neurons and studied their interactions with each other. They called these groups of neurons neuronal nets. The question then was; can a given neuronal net compute a problem to arrive at a solution? The neurons were studied in terms of the level of activity that they generated. A neuron that fired was active, or on, and one that didn’t was inactive or off. In such a fashion they viewed the mind as a discrete collection of logic gates that replicated the process of binary computing.

A few years after their work, Vannevar Bush published “As we may think”, which was a work of science fiction that described a future in which humans and machines would assist humans with many activities

(iBiblio).

Soon after that, AM Turing published “Computing machinery and Intelligence”. This presented a novel idea which is commonly referred to as the Turing test of intelligence. If a machine could perform a set of computations based on a quantifiable input, and it would process that input based on rules, then the rules that drove the machine would be under scrutiny for intelligent behavior using the Turing test. The test in essence was a way to determine intelligent behavior and could be used on human beings as well as machines. This is a simplified explanation of what the test aimed to achieve, but in the real sense of the word an operator could easily distinguish a machine from a human. The reason why this is true is because machines run on a system of predefined rules to act on the various modes of input. Human behavior can be fashioned into a similar system of cause and effect based rules. However, it is impossible to encode human decision making into a set of rules, as one cannot predict the exact choice that will be taken by a person given an objective input. This idea of “humanness” cannot be replicated and can be attributed to the abstract notions of ingenuity and creativity. Abstractions such as these cannot be quantified, and hence cannot be understood and processed by a machine. Yet the implications of this idea were to inspire future thinkers to elaborate on the subject.

Many objections were raised on the plausibility of the Turing test. One significant idea is the fact that digital computers are discrete machines. The states that they occupy can be on or off and are finite in most computations. The human mind is an electrical organ as well as is the computer. But the mind can occupy quantum states, in which a registering neuron does not always have to be in a finite state. The state of a quantum neuron can be thought of transitory. Within this transition, it is impossible to predict the exact state of the neuron. Quantum principles dictate that the mere act of observing a system will change the state of the system. So in the case of the human mind, the mere fact that we try to measure the state of the neurons in question, will lead the observer to inevitably change the final outcome of the experiment. Another interesting objection to this argument of observation is the fact that the human mind consists of an unquantifiable entity called the soul. The soul is what separates the mind from the collection of physical neural networks called the brain. Some hypothesize that the entire collections of neurons in the brain is what gives rise to this notion of the existence of a soul. This can be easily refuted by building a machine that emulates the brain and observing to see if consciousness is indeed created from such complexity. To date no such machine has come into existence. The computing power that would be required to create a machine that effectively replicates all the nuances of the mind would take humanity years to realize. Even though logical thought is one of the methods of biological organisms, there exist other forms of intelligence like creativity and abstract thinking that don’t require input in order to realize an outcome.

                Another important figure in the history and development of Artificial Intelligence is Isaac Asminov, who wrote the “Three laws of robotics” in 1950 (Weld and Etzoni, 5). The laws are

1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.

2. A robot must obey orders given it by human beings except where such orders would conflict with the First Law.

3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

(Auburn).

These laws were formulated from a science fiction standpoint. Robots did not exist at this time and his ideas were based on a fictional work he wrote. According to the Oxford dictionary, Asminov is said to be the first author to coin the work robotics in his work “Liar”. He envisioned a world that would consist of humans and robots living side by side. The robots would be subservient to their human counterparts and would have to obey the laws that ensured the existence of the state of things. The first law was designed to protect humans from harm. If any harm was imminent then the robot should put its own existence in jeopardy without thinking that it was doing so and save the human’s life. This statement is a paradox in itself as if the robot were to be ascribed human qualities of thought and emotion, then it must also be ascribed the will to pursue a continued existence. If a continued existence was desired, then the robot will not sacrifice his life to save the life of his creator.

The second law exemplifies the fact that human beings will always be the ones that are in control, and the robots will always be subservient to their masters. This implies that the machines cannot make decisions for them and are merely instruments that can be programmed to meet a certain end. There are examples of such robots in existence today that are provided a set of instructions to handle a variety of inputs. However, if the inputs deviate or the instructions cannot handle the inputs that are encountered, then no solution is found and there is a period of inactivity.

The third law is based on the first two laws and ensures that the robot cannot use a loophole in the laws to create its own law to destroy a human life. These laws give any future robotist a set of guidelines that mandate how the machines need to work.

Modern culture has proven that malicious activity is always going to be rampant and if a system is designed with a set of rules that mandate its operation, then those rules can be changed and the system will behave in ways that it was not designed to. Ideas of terrorist activity and cyber crime are thoughts that could seriously affect society if rule based robots were to become commonplace.

In 1956, Allen Newell, J.C. Shaw and Herbert Simon from the Carnegie Institute of Technology wrote a program called “Logic Theorist”. This was one of the first running Artificial Intelligence programs to run on a computer (J-Paine).

The program was based on a number of axioms of logic. These axioms mimicked the knowledge of objective facts that is contained in the human mind. Rules were used to generate permutations and combinations from these axioms based on the rules of logic. When a final proposition matched the initial problem to be solved, the path that was taken to reach a solution was the method to arrive at that solution and the problem was considered to be solved. This is a wasteful approach and did not guarantee a solution, but the idea was solid and inspired many a thinker to propose new and improved solutions to the problem of Artificial Intelligence. Future refinements were made the program where the probability of a certain outcome based on the state of all the inputs was taken into consideration. The probability programming was not as discrete and linear as the program itself and was merely a catalyst for the final solution. This speeded up the time to execute the program, but yet did not guarantee a solution.

                From the 1990’s to the 2000’s significant advances have been made in artificial intelligence research. There have been more advances that have been made in the last ten years than there have been in all of human history. The explosion of the internet has given rise to a free medium of communication between individuals that are interested in this subject and has led the ideas of one to be a source of inspiration to the outcomes of many experiments. Significant advances have been made in the areas of machine learning, intelligent tutoring, case based reasoning, multi agent planning, scheduling, uncertain reasoning, data mining, natural language understanding and translation, vision and virtual reality. Understanding the breadth of these topics is critical to understanding the current day status of Artificial Intelligence research as it stands today (J-Paine).

                One of the most important theoretical advancements in the area of Artificial Intelligence is the idea of “neural networks”. The human brain is the ultimate artificial intelligence machine and all efforts to replicate it from computational standpoint have failed. Artificial Neural Networks attempt to simulate the conditions that exist within the human brain. The brain consists of more than a hundred billion neurons (University of Toronto, Canada). No artificial neural network has been created to date that can match that level of complexity. The most complex artificial neural net consists of a few thousand artificial neurons. The key concept in the natural brain is the idea of connection strength. The strength of the individual synapses and dendrites is what leads to advances in memory and intelligence. A stronger connection will mean the neuron is in an excitatory state and a weaker connection means that the neuron is in an inhibitory state. Some neurons are sensitive in nature. This means that they do not require a large amount of signal strength as input to fire. In this case the output is greater than the input.

                Artificial neural networks are organized on the basis of some structured arrangement. The brain, however, does not contain a structured arrangement of neurons and connections are created and destroyed depending on how often they are used. During certain periods of mental activity, the use of a collection of neurons implies that they are responsible for eliciting the perceived response. In Artificial Neural Networks, learning takes place in a similar fashion. The artificial neurons that are used to most, are assigned a higher weight, which translates to them firing with greater intensity and having a stronger connection to the surrounding, related, neurons. In order for the network to learn a complicating set of input is provided. On providing input, the net will begin with trying to find all possible values for the solution in a random fashion. The initial randomness is ascribed to the lack of weights on the individual neurons. Once the input has been checked, processed and compared to the output for the currently processing thread, and that thread has been checked against all possible values for a Boolean value in the output then the strengthening or weakening will commence. If the thread executed a true value, then it will strengthen itself depending on how true the value was and vice versa if the value was a false. Neural networks devised in such a manner are becoming popular as they can solve problems with less than perfect input. A solution to any given problem cannot always be found by Artificial Neural networks, but the probability to find a solution is greater than that of traditional, symbolic artificial intelligence processing systems.

                A successful understanding of the concept of Neuronal Networks would be incomplete without understanding the individual components that are involved in the process, i.e. the artificial neurons. What is an artificial neuron? What does it consist of? And how does it work? Artificial neurons are similar to human neurons in which they comprise of the four basic components that any neuron possesses; dendrites, soma, axon and synapses. The dendrites are hair like growth that are responsible for the inputs to the neuron. These input channels receive their input from the output channels called dendrites from other neurons. The soma is the processing body of the neuron that holds the input for an arbitrary length of time. Once the processing has completed, the signal is transmitted to the target recipient through the axon or dendrite depending on the signal type. Artificial neurons are designed as computer programs with similar properties to their natural counterparts. They use the same principle of weighting inputs as real neurons do. Each of the inputs to the artificial neurons are summed depending on the weight value they acquired from the processing. This summed value is then fed forward into another set of neurons that compute a new value based on the modified inputs. The entire process is a complicated one, but once the program reaches a desirable output, it feeds backward into itself with the new output generating new values (University of Toronto, Canada).

                Another way in the system can be made to replicate the human mind is by altering the arrangement of the neurons that conduct the processing. The neurons in the brain seem to be capable of making an almost infinite number of connections. This does not seem to be case in physical silicon models that aim to emulate this pattern. So developers have taken a software based approach. Even though it might seem to be a random collection of neurons, every brain has structure. The structure in the human brain can be categorized into layers of input and output. Each layer is responsible for some amount of processing after transferring the output to the next layer and so on till a result is found.

                Linguistics in Artificial intelligence is an interesting field in its own right. The supposition is that language is learned and not innate. If a machine can learn language and all of the nuances that are associated with it to give rise to reason, then the machine is truly intelligent. There have been recent breakthroughs in this field, notably in the information technology sector of the economy. In his article “Computing machinery and Intelligence” Turing intertwined disparate threads of scientific thought, like computing, mathematics and linguistics into a single chord noting that machines could display intelligent behavior as long as the definition of intelligence was based on the use of language. The essence of the Turing test is that intelligence is measured though the social use of language (AI dot com). There are quite a few other indicators of intelligence but language is considered to be the most important one. One can determine the mental capabilities of an individual by studying his use of language. Ideas and thoughts are also expressed thought the use of language. If a machine were to emulate such a complexity in learning a linguistic structure and that machine will demonstrate a capability to combine words following the rules of grammar to achieve meaningful communication, then that machine is considered to be truly intelligent. This notion might seem overly simplistic as the machine would need some sort of stimulus to respond to. In providing the machine with this sort of stimulus, the operator has increased the amount of memory in the machine and as a result increased its intelligence.

To further exemplify the importance of language on the evaluation of intelligence, consider the case of Steven Hawking. Hawking is one of the leading physicists in the world and has won a Nobel prize. He is renowned for his work in quantum physics and string theory. Due to an unfortunate incident he has lost most of his motor abilities including the ability to vocalize speech. In the past, people like this would be confined to an institution and their mental faculties would be wasted. But in his case technology has exemplified the smallest movements that he makes on a keyboard giving him the ability to speak through a synthesizer connected to a computer. Even after this tragic incident, people still consider him to be leading scientists. The only means for them to ascertain his thoughts are through the words that are generated from his synthesized keyboard. This means that the only way in which we can determine his intelligence, is though his linguistic abilities.

                Another facet to this discussion is the existence of machines that can recognize and process language. Language in itself is a complex combination of words that are meaningful in a lexical and grammatical way. The grammar is what gives a sentence logical structure and the words are the entities that are to be represented. If a machine is to process language successfully, then it must be able to recognize the difference between these two ideas. One of the first language processing programming attempts was the program “ELISA” (Weizenbaum, 1966). The program did not have any innate knowledge and was a pattern matching a pattern recognition engine whose input was based on text alone. The program would essentially pick up on key phrases in a sentence and then formulate its own questions based on the users input. So if the user were to say, “You remind me of my mother ELISA”, ELISA would say “Why do I remind you of your mother?”

Most people tend to favor compliments. Programs that are devised to converse with human’s using speech recognition are becoming more popular and polite. Sophisticated programs have the capacity to learn a particular user’s input. The program needs a sample of the user’s voice. This is usually accomplished by displaying a few sentences that are grammatically correct and are in different tenses. The program will then recognize the voice and match up the rate of speech and the pronunciation of words. Once this has been accomplished, all input can be converted to text. Once this textual information has been stored in memory, the input can be analysed based on complex pattern matching. The system is not foolproof and there is always a way to tell if the program is not human. But the implications for such a program are stupendous. Programs like these can be used to allow users to control devices using the power of speech as an input device. Handicapped people will also benefit greatly from such a system. Personal security is another concern that can be addressed using this technology. For instance, if someone was in a medical emergency, and if that person had a speech recognition device receptive to all forms of input close by, then they could call for help. The system will then dial a number depending on the user’s location and have assistance arrive immediately thereafter.

                Natural language processing is not a novel concept. The idea has been around since 1960. The explosion of natural language processing was observed in the 60’s. Terry Winograd designed a system called SHRDLU which was a robot in a world of blocks (Jurafsky, 2000). The robot would accept spoken commands from a user and convert them into spatial coordinates. These coordinates could be used to move virtual blocks on the screen. The blocks were positioned in a random fashion. The commands had to be spatially accurate. For instance a typical command would sound like “move the red block over the blue block”. This command would be translated into special coordinates by the computer that used keywords present in the command. From the 60’s to the present day there have been tremendous increases in the processing power of computers. Software programs have also become increasingly sophisticated in their abilities to use probabilistic reasoning to interpret the user’s spoken input.

 Human beings are prone to copy nature and all its grandeur. The method of choice is technology. Nature’s creations are vastly superior to the technological advances made by humans. The ability of a living organism to process semi structured data without any formal input gathering and training is astonishing. The sheer number of neurons in any human brain far exceeds the ability of a machine to replicate. The level at which life functions is microscopic. All the processes that are carried out in the brain are what give rise to conscious being that has the ability to dissect and analyze even him. But there will always be an unknown in this world of knowing. Human beings can only emulate nature using technology and not replicate it in its exact likeness. Some scientists have amassed the ability to clone cells and yet others have synthesized synthetic DNA and introduced it into cells of living organisms. But in order to make a truly intelligent machine, the brain and all of its nuances must be understood. This is a stepping stone as once we open these doors of perception to the brain; we will have the ability to create machines that are truly amazing and wonderful. Picture a machine that has the capabilities of thought and reason and could analyze and solve any problem within seconds. Problems that would take people years to solve would be solved within minutes. But the truth of the matter is that the human will always be in control. Contrary to science fiction books, the human being will always possess an edge over the machine.

                Artificial intelligence has been a very complex field and will remain so. The efforts to conquer the mind have led researchers to devise new and complex ways to understand the problem of mind and thought. Even though the problem has not been completely solved, progress is being made. If Moore’s law were to hold true, the next few years are going to see exciting advances in the field of artificial intelligence. There will be a number of machines that will have the ability to trans-code and interpret human speech. Machine s will assist people in making choices and help them better understand their environment. Complicating questions in medicine can be answered by a machine that has understood all the facts. The decisions that these machines make will be integral to the existence of human life as we know it. Even in the areas of space exploration great advances can be made. We could send out machines into deep space, on voyages that a human being could never survive on and have them conduct analysis like a human was present. An understanding the theoretical and moral implications of having technology that requires a minimum amount of input are to be put on the center stage if such technology were to realize its fullest potential. We will now discuss some of the current day aspects of artificial intelligence and some of the cutting edge applications that it is used in.

Current Uses of Artificial Intelligence

                We would like to get into some of the current uses of Artificial Intelligence now that you know its history.  The ideas of AI from science fiction stories of long ago are pretty much a reality today.  Who would have ever thought the story line from the Terminator trilogy would ever become a reality?  We bring this up only because the military has incorporated the use of AI for several uses, of which I will explain later.  Everywhere we look and go we run into some kind of device using artificial intelligence.  AI is being used to help people in any field and walk of life work smarter and faster instead of harder.  It is appearing in business with various decision support systems and knowledge bases, in bionics with various prosthetics which learn how its user moves, in warfare, in anti-terrorism efforts, in journalism, in anti-crime efforts, and even in video games. 

Business Uses of Artificial Intelligence:

                Some years ago, when a company needed some kind of information about a big decision or deal, it went out and paid for the services of a counselor.  If it did not want to pay for such a service through the use of a contract, it had established teams to come up with the same results.  While these means typically ended with the same results, the amount of money paid out for the labor must have been a lot.  As technology grew over the years, it became cheaper to have a one-time interview with people and capture the information.  After the information was recorded, a program came up to assist the employees.  Artificial intelligence has been used in business matters like this since the early eighties.  Unfortunately, these early applications had died, until the recent boom of web-enabled infrastructures with a need for real-time data processing.  Businesses use AI applications to make sense of the vast amount of data they can acquire at any given moment.  For example, a company could keep track of its sales information, and use the information to track consumer buying habits.  After they process this information, they can then use it in a decision support system to solve any problems dealing with anything from supplies to product liking (financialexpress.com).  An example of this more efficient work is done with expert systems.  Basically these systems are programs which manipulate symbolic information at very high speeds.  The speed compensates for the partial lack of human knowledge and selectivity.  Some programs exist which simulate human problem solving methods through the use of highly selective recognition patterns.  Since the mid 1990s, knowledge-based expert systems have taken the forefront.  These types of systems have allowed a computer to make decisions which solve highly complicated non-numerical problems.  Basically these systems are made with thousands of if-then statements, which are formulated with knowledge gained from top authorities in any given field (thinkquest.org).  When AI is used to develop these expert systems, the knowledge engineer interviews several experts and creates a program which simulates a person’s knowledge, intuition, and reason.  This gives a person who uses the system the best options under any given circumstance.  The business uses of AI systems range from financial management to forecasting and even to production.  A further development in business AI is the use of Artificial Neural Networks and the aforementioned expert systems.  The neural networks are a well-known technique to recognize patterns, particularly of large images, data streams, complex data sources, and even data mining.  The use of these neural networks have even expanded to fraud detection, cross sales, customer relationship management analysis, demand and failure predictions, and even non-linear control.  Some businesses have incorporated “self-healing” systems which basically auto configure themselves to correct for ever-changing conditions, continuously monitoring its attached systems, and even fine-tuning the workflow to meet predetermined work conditions (financialexpress.com).  The use of artificial intelligence has found great success in the areas of risk management, compliance, securities trading and monitoring, and once again in customer relationship management. 

These systems can be created through specific AI languages like LISP and Prolog, or even FORTRAN, Java, and C++.  These systems can keep all the knowledge of a great employee who has decided to retire and archiving certain skills learned (AI in business).

Bionic uses of Artificial Intelligence:

                Remember when a person would lose a hand and they would get a prosthetic hook to replace it, or when a person would lose a leg and get a prosthetic leg, but be susceptible to falling or walking awkwardly?  Well those things are becoming obsolete because artificial intelligence has recently made the jump into the prosthetics field.  In 2005, MIT developed a prosthetic knee, which has sensors which analyze the knee 1,000 times per second.  This analysis allows it to adjust itself for any step or even misstep (10news.com).  This prosthetic knee has a microprocessor and basically learns how the user walks, therefore recognizing and responding immediately to any change in speed, carrying load, and terrain (ossur.com).  The processor, created by Motorola, controls the viscosity of a magnetic fluid in this knee.  When the knee has been turned on, it chains together small pieces of metal in this fluid which cause the liquid to thicken.  The sensors then pick up how the user is walking and the thickness of the joint is altered according to the rhythm of the person’s walk.  The knee can be monitored by the use of HP’s iPAQ, which can be plugged directly into the knee.  The iPAQ then picks up information about the internal settings, battery information, and any other data created within the knee (WSJ.com).  As of July of this year, scientists at Touch Bionics have designed and begun testing on a prosthetic hand which promises a very high level of control and precision for its users.  It has released the i-LIMB hand and ProDigits.  The i-LIMB is a full hand with which looks and acts like a human hand.  ProDigits is a partial hand for those who have missing fingers.  The i-LIMB is controlled through the electrical signals in the muscles from the patient’s remaining healthy arm.  It has a control system installed which uses myoelectric signals from the remaining parts of the arm muscles to open and close the fingers.   The electrical signal is picked up by electrodes on the surface of the skin.  The fingers also adjust their positions depending on the object it is grabbing (smart-machines).  The software which controls the i-LIMB can be configured for what is known as “thumb-parking”, basically resting the thumb against the index finger for putting a coat or other clothes on.  This software also provides the necessary speed and grip strength control to the device which is derived from the user’s needs gathered from the myoelectric signals (touchbionics.com). 

Artificial Intelligence uses in War:

                Even the military has picked up on the artificial intelligence bandwagon.  It has robots which detect roadside bombs, searches caves, and even acts as armed sentries.  A new robot which will soon be released is called MAARS (Modular Advanced Armed Robotic System) is armed with a powerful M240B machine gun, and a robotic arm for gathering Improvised Explosive Devices currently used against US forces in Iraq (foster-miller).  The MAARS robot has new software which allows the driver to select fire and no-fire zones, and the reason for this is to prevent the accidental shooting of American soldiers.  It even has a mechanical range fan which keeps the gun pointed away from friendly position.  The MAARS device can be equipped with a GPS system so it can be tracked in the battlefield (wired.com).  In 2002, an unmanned Predator drone, which had semi-autonomous capabilities, self-navigated itself over a car full of suspected Al-Qaeda operatives.  The decision to engage them with Hellfire missiles came from pilots 7,000 miles away.  After that time the expectations for unmanned drones had been increased.  The More deadly Reaper drones have flown in many missions, but resulted in civilian loss of life (thinkartificial.org).  However, more is coming since AI is drawing closer to becoming an exact replica of the human mind.  AI is being used to teach our soldiers to speak Arabic.  Basically, a computer game uses AI to do this, and it requires the soldiers to complete missions.  This game works by using speech recognition technology and it evaluates the words and detects common errors.  After it completed the assessment, it creates a model of the soldier and keeps track of what they have learned and what gives them problems.  The military is also seeking to create a fully automated vehicle, since robots can already detect their surroundings and read a map to drive without human assistance.  This has been attempted as recent as 2004, when the Pentagon sponsored a contest for fully automated prototypes to drive a course in the Mojave Desert, which simulates the conditions of Afghanistan and Iraq.  During this contest, no competitors made it through the course however, one year later students from Stanford University created a car which used onboard computers and sensors to navigate that terrain.  This model from Stanford completed the 131 mile course in six hours and fifty-three minutes.  So much money has been invested in the military use of AI, that congress ordered one-third of all military vehicles and deep-strike aircraft be fully automated by 2010.  The first robotic soldier, powered by AI of course, will be a major force in the US Army, and probably within the next decade.  These first generation robots will actually be remote controlled vehicles.  These robot soldiers need to be able to determine friend from foe, gain our trust and their own autonomy.  The current infantry prototypes do however determine what an enemy is when it is under fire.  A fully usable version of this prototype will be implemented by 2015 for various infantry missions (AIWar).  The United States is not the only country in the world incorporating artificial intelligence in its armed forces.  South Korea is deploying armed robot border guards (thinkartificial).  Soldiers are not the only aspect of American forces being automated.  As of August of this year, DARPA reported the use of AI embedded droids have learned how to engage in mid-air refueling of jets.  This potentially allows these unmanned drones to stay in the air for extremely long missions.  During the testing phase, it conducted 11 mid-air refuels without human input.  The test used an F-18 fighter jet, and overlooked by NASA.  The jet had a human pilot, just for any emergencies, and the system did not need him at all.  The system compensated automatically for any turbulence and the refueling pipe moving up and down naturally.  This system did accomplish one task which humans cannot apparently do.  It connected with the refueling pipe during a turn (theregister.com). 

Journalist uses of Artificial Intelligence:

Artificial intelligence has recently bled into the journalism world.  With the 2001 launch of NewsRX’s media company VerticalNews, a new proprietary AI application has been born.  VerticalNews had a two year goal of 3,000 new articles a week, and this feat used the AI systems of Component Aggregation Technology (CAT) and Journalist Assisted Technology (JAT).  These two systems had been used to assist their journalists to perform vast publishing and news related tasks (findarticles.com).  This is only the beginning of the marriage of journalism and artificial intelligence.  Have you ever watched the news and wondered why you are watching it because it is presented so ridiculously?  Have you ever wondered why all you hear about in local news is one person killing another, the latest weather changes, and high school sports?  Well as of November of 2006 a new software called “News at Seven”, created at the Intelligent Information Laboratory at Northwestern University, allows you to grab news feeds from your favorite news sources.  Once it retrieves the feed, it uses the captured text and runs a search for images, videos, and even blogs about the topic you chose.  Once it gathers the information it needs to make the news show, it edits the data to have a spoken segment anywhere from 45 to 90 seconds long.  This software even eliminates acronyms, abbreviations, converts passive voice to active voice, and modifies the quotes to make it more conversational.  It even searches the blogs for related phrases and decides if it is written with a positive or negative tone, then ranks the posts, and then says the phrases that match the emotional level you enter.  If this weren’t enough, it feeds the program to a computer animated character, whose tone is influenced by the language in any given story from your request.  On top of all this, you also get to pick the location you want the broadcaster to give you the news from (discovery.com). 

Anti-terrorism uses of Artificial Intelligence:

                Artificial intelligence has found itself a new use in our day-to-day lives.  The government has incorporated AI applications in its counter-terrorism efforts.  Articifial intelligence found its way into the Olympic arena when Salt Lake City hosted the 2002 Winter Olympic Games.  In order to detect the early warning signs of a possible bioterrorism attack, an AI system was installed.  Basically, it detected patient data from emergency rooms and instant care facilities looking for a significant pattern, and if it finds the pattern it calls the state health official.  Real-time Outbreak and Disease Surveillance (RODS) was a joint development between The University of Pittsburgh and Carnegie Mellon University.  This system assists health care officials of an outbreak of any kind.  It can review large amounts of data taken from many geographical regions and detects patient trends very quickly.  It incorporates several AI machine learning algorithms, and data mining algorithms to keep track of the patient flow (tabor).  In these dire times of security at airports, AI is having an impact.  As recent as October 1 of this year LAX airport has had a huge effect of its security by the use of AI.  LAX police is launching a new program, designed by a doctoral student from USC, which seeks to keep potential terrorists and criminals uncertain to whenever and wherever vehicles will be searched at the airports entrances.  This software not only breaks up police patterns and behaviors which terrorists can study and learn, but it also makes it virtually impossible to predict where and when LAX security resources may be deployed.  It also creates a schedule most likely to catch any kind of criminal behavior (latimes.com).  This system called ARMOR, launches once LAX security officials hit the button which is labeled as randomize.  This software is designed to stop any terror threat during the 18 month to 4 year surveillance phase.  The key aspect of this software is the use of game theory.  Not only does ARMOR randomly place security agents, but it will be expanding to the bomb-sniffing dogs as well (msnbc).  Anti-terrorism efforts got a significant boost beyond that given by the LAX experiment.  In 2006, the Princeton Plasma Physics Laboratory developed a system known as Miniature Integrated Nuclear Detection System (MINDS).  MINDS can be used to scan moving vehicles, luggage, cargo vessels, and anything like that.  It scans for nuclear signatures associated with materials employed in radiological weapons.  MINDS can detect X-rays, soft gamma rays, gamma rays, and neutrons.  MINDS can recognize and differentiate between any other kind since each radionuclide has its own fingerprint.  It compares the suspected radionuclide with a spectrum of radiological materials which may end up being used in weapons.  MINDS can detect one-billionth of materials deemed plausible to create a dirty bomb.  This system can be deployed in many areas because it can differentiate between naturally occurring radioactive elements, authorized medical substances, authorized industrialized nuclear materials, and materials which cause threats.  MINDS can be configured to eliminate false-positives because you can tell it to filter out any naturally occurring radiation either on the items being scanned or in the background.  It can even detect materials whenever they are concealed.  In one second of scanning an item, MINDS sense a target, identify what it is, and transmit any radioactive materials at a level slightly beyond the background levels.  This system also does not need to emit any kind of radioactive signals to react with the target.  All of this scanning is great but here is where the artificial intelligence aspects kicks in.  MINDS basically has two very powerful algorithms which work together in a synergistic fashion which resolves the nuclear spectra in a real-time and hyper-accurate manner.  The first algorithm uses the classical peak fitting curves to isolate specific areas of the spectrum for the identification of the radionuclide.  The second algorithm is mainly where we see the AI usage.  This algorithm resolves and identifies the nuclear spectrum which is unique to the radionuclide of interest.  The MINDS system in its current setting stores the specific energy signatures of sixteen individual radionuclides (pppl.gov).  Just when you thought terrorism only happened at physical sites, the movement has expanded to the cyber world.  There are between 7000 and 8000 terrorist sites on the internet, and they do everything from spread propaganda to offer up where the insurgency should strike next.  Hsinchun Chen, the director of the University of Arizona’s Artificial Intelligence Laboratory, has created a new tool called Dark Web.  Basically this is a giant database on extremists, and it is searchable.  It attempts to uncover, cross-reference, catalogue, and analyze all online terrorist-generated content.  Since there is an infinite amount of content on the internet, Dark Web relies on many analytical tools.  It uses statistical analysis tools, cluster analysis tools, content analysis tools, link analysis tools, and new technologies such as sentiment analysis.  This type of analysis can scan documents for emotionally charged words.  This type of analysis is important because it looks for any kind of anger and hate in order to rally social activists and suicide bombers.  This tool also employs social-network analysis which maps extremist networks.  This basically determines the pecking order in the extremist organization.  The nomenclature of the network is discovered by using centrality and structural-equivalence measures to determine prestige given to any member and it examines the social network.  Once this information is gathered, researchers then can discover the cohesiveness and density of the unit.  This shows the stability of the group and the nodes most vulnerable to attack.  This is actually Dark Web’s second time entering the law-enforcement realm.  The first time was known as Coplink, and Chen teamed up with the Tuscon police department and the National Science Foundation.  The purpose of Coplink was to create a way for law enforcement agencies around the country to share information with each other (foxnews.com). 

Artificial Intelligence and the Law:

                Artificial Intelligence not only helps to ensure our national security, but it is used to enforce our laws.  Artificial intelligence is now being used to deter health care fraud.  This happens through the use of computer programs, which can flag potential fraud even before the medical claims can be paid out.  The use of such programs has saved Aetna $89 million in payments to the medical providers.  These programs basically flag unusual patterns in a claim made to health care providers, Medicare, or Medicaid.  Within two days of launching this software, Aetna caught a dermatologist who improperly billed $350,000 worth of cosmetic hair-removal treatments (usatoday.com computersecurity).  Sherlock Holmes has even become real through the use of artificial intelligence.  A group of Scottish software developers have made a program which helps police consider all possibilities involved in the investigations of suspicious deaths.  It takes an overview of the evidence available and then comes up with scenarios which might have happened.  These scenarios cover the less obvious story lines which the investigating officer may have missed.  This basically forces investigators to look for other ways which the murder may have happened.  This helps to counteract the human trait of coming up with a scenario and sticking to it by trying to prove it.  Sherlock Holmes has a knowledge base which contains data of various deaths and the evidence surrounding them.  Then it presents the evidence which supports or contradicts the explanation for these deaths.  Even any evidence gained by forensics, medical reports, and eyewitness accounts can be fed into the system (Sherlock_holmes).  The use of artificial intelligence has even spilled over to the legal realm.  With the growing number of divorce cases, divorce lawyers and divorcees are now turning to software which helps to settle the bitter disputes.  The program is based on game theory concepts, and asks the divorcing participants to rate how much they want every disputed item by assigning points which shows each item’s importance.  Each person gets 100 points and gets to allocate them however they want.  The software would then tally the points and create a trade-off map.  It then solves the easiest dispute, which is the one with the largest point discrepancy.  This ends up with a direct reflection of priorities set aside by the participants.  Whoever loses the first dispute gets some extra points to assign to another item, which then causes the trade-off map to be revised.  After the revision, it then repeats the action and decides who wins or loses the item in question.  This software seeks to make a scenario where things split evenly (livescience.com).  Xerox is working on a new software suite which will help the legal profession profoundly.  This software is called e-discovery, and will be used during the discovery phase of a court case.  The role of this software is to help legal workers sort through all kinds of emails and other documents.  This software would be used to identify a sender and receiver of any given message.  It could even pick and choose the dates which the message was written or sent.  This software takes the search engine algorithm to a whole new level.  This software is going to search for the underlying grammar of a text in order to get additional information about the terms being searched.  The designers of the software wrote the program in C, but did include modules written Java and Python.  This software can even be linked with audio transcription tools, so it can search radio and television archives (pcworld).  The idea of the Sherlock Holmes software does not stop at homicide.  It is currently being used by police to solve many other crimes.  A system is now in place which compares case records with all the files it has for all past crimes committed.  This system uses pattern recognition software to link related crimes which may have taken place in widely separated areas where the different police forces may not be in close contact with each other.  The system basically takes in all the records available to it and assigns points to the different aspects of the crime, for example the offense, the offender’s sex, height, and age, also the weapon used, an escape vehicle and, any other trait involved with the crime.  Once it assigns values to these aspects, builds a crime description profile.  Once this profile is built, it then uses a neural network to go through and search for any crimes with similar traits.  If a crime took place and it matches with another crime, the system compares when and where they took place.  It does this to determine if the criminal would have had enough time to move to the next scene.  During the testing phase, this system found ten times more matches than a team of detectives with access to the same amount of data.  This system is not intended to replace human detectives, but to give them a starting point in catching the next serial killer or organized crime (newsscientist.com).

Video Games and A.I:

                Artificial intelligence has become so engrained into our daily lives; we encounter it when we sit down to kill some time playing video games.  Video game programmers use artificial intelligence to help make characters which have a set of responses for a gamer’s actions.  The trick is to make this part of gaming act spontaneous, but here is how it works.  Several rules exist for having an efficient AI application, and the main one is to have the gamer’s actions kept in a database, and when the game is faced with a situation where AI must be used, it searches this database and picks the most logical selection.  The spontaneous façade is tied solely to the size of the database.  Basically as the game situations increase where humans make specific choices, the database grows, and the game becomes more challenging.  This database is used to simulate a human choice of a given situation (adigitaldreamer.com).  In keeping with the artificial intelligence used in warfare and anti-terrorism sections, we would like to talk about how video games which entail combat situations use AI.  These types of games depend on AI to keep the player interested and involved in the game.  This field didn’t become huge until the release of a game called Half-Life.  In this game, teammates and enemies controlled by AI had better responses to being shot; spotting grenades, a realistic awareness of the player paved the way for the games we see today.  In today’s combat situational games, we see the AI controlled characters ducking behind walls, hiding around corners, throwing grenades back, and even filling in for human players in a multi-player situation.  These combat AI controlled characters can dodge gunfire and shoot just as good as a human player, but does not have knowledge of the surroundings, efficient teamwork, hunting abilities, and survival methods a human would possess (ai-depot).  The newly released Halo 3 has taken the forefront in the use of AI in video games.  Eventually the graphics of our modern gaming consoles will reach a limit, and the use of artificial intelligence will take the lead.  In this game, enemies and friends act as if they are human.  You see them hiding, deploying shields, or even a small use of teamwork.  This game assigns up to 10,000 rules to each character individually (usatoday.com). 

                Artificial intelligence has a huge importance in each of our lives.  We could not get through any day at all without having an encounter with artificial intelligence.  Whether we are relaxing in front of the television playing a game, or watching the news and seeing the government applications of anti-terrorism we encounter it.  As we continue to acquire better technology and more powerful parts to support these programs, these uses mentioned will only expand to serve greater purposes and be improved upon. The days of artificial intelligence being only in science fiction stories are now a reality.  As we continue to incorporate AI in our lives, we will grow in its infinite uses. Now that we have a solid understanding of the past and present advances in the field, we will examine the future implications of Artificial Intelligence.

Potential Future Applications:

                Once you have established an understanding of the vast network of Artificial Intelligence already around us in the form of pattern recognition, Decision Support Systems, Expert Systems, learning algorithms, Turing test proven Artificial Life and the fascinating things we have already accomplished the next thing to evaluate is the possible future realities that could be just over the horizon. When faced with the realities of the size and scope and potentialities of Artificial Intelligence it is easy to see why it is a favorite of science fiction and imagination that is either thrilling or horrifying, depending on your perspective.

                The future applications of Artificial Intelligence are even now being evaluated at research institutions and projects worldwide. These include Dr. Chens work at University of Arizona’s Artificial Intelligence Lab, the research into emergent systems and autonomous agents, and the neural nets of said agents.

                The immense and fascinating work being done at the University of Arizona’s Artificial Intelligence Lab is focused on physical security by the evaluation and interpretation of network data streams. In their own words: “Information systems that effectively collect, access, analyze, and report data relevant to catastrophic events are critical to helping prevent, detect, and manage responses to these attacks. “ (Chen and Wang 2002) The concept is really quite simple, but the implementation is of course quite complex.

                The project is called the Dark Web Portal, and through the use of pattern recognition and visualization of terrorist networks they are attempting to redefine the nature of online intelligence gathering by creating an intelligent system that will actually identify and track terrorist sites using advanced algorithms designed to interpret and learn the Dark Webs ( Terrorist sites and groups that pass terrorist data and communications on the Web) information flows, granting the Homeland Defense Department the ability to react proactively to potential threats by identifying them before they become a reality. This system is defined by the University of Arizona’s Artificial Intelligence Lab as: “a Web-based counterterrorism knowledge portal, called the Dark Web, to support the discovery and analysis of Dark Web information. Specifically, the Dark Web Portal integrates terrorist-generated multilingual datasets on the Web and uses them to study advanced and new methodologies for predictive modeling, terrorist (social) network analysis, and visualization of terrorists’ activities, linkages, and relationships.” (Chen and AI Lab Team 2004)

                It is the predictive modeling and network analysis aspect of this research that most concerns us in an evaluation of the future prospects of Artificial Intelligence. The possibilities of the network integration and analysis aspect will be discussed at greater length later in this discussion, but the predictive modeling aspect is critical as well. Predictive modeling is fundamentally described as follows “simply, a predictive model is an equation that identifies main characteristics affecting the likelihood of certain behaviors or attributes of a client or a prospect.”  (Kinav and Marquis) In a more complex Artificial Intelligence system this takes on the aspect of rapid decision support systems and expert systems evaluating massive datasets and arriving at conlusions based on inputs and variables in a fraction of the time it would take and ordinary human being. This is critical in a fast paced dynamic battlefield like the War on Terror. We are faced with an enemy who is clever and highly mobile, and a powerful tracking and prediction system can save thousands of lives by preventing an attack through efficient and capable decision making processes.

                Network analysis is another fascinating possibility in the future of Artificial Intelligence. The ability to not only access data throughout all of the worldwide networks but to evaluate it and make decisions is holds amazing potential for the power of future Artificial Intelligence systems. We now have algorithms like the google search engine and Ask Jeeves search engine which are learning algorithms, taking data from user inputs and external inputs acquired through user searches to refine the searches across the system and access information on a statistically based evaluation system. This allows for practicality and efficiency in online searching and amazing growth in the filtering and accessing of information. Of course, the Dark Web Portal is another example of a similar system of statistical analysis.

                What is possible with network analysis and advanced expert systems and decision support systems being developed with particular communication and control systems in mind is nothing short of fascinating. Think of an Information System, vast and with complex evolutionary dependencies and integration of hundreds, even thousands of expert systems, DSS structures, learning algorithms, and autonomous agents all linked together. All based in a central command network with specified goals and parameters programmed in that would design outcomes and determinations in a predetermined format. This is a system that could integrate all other systems, a command and control system that could control traffic, police investigations, firefighter response, air traffic, military strategy, and innumerable other small functions of our daily lives and societies.

                Although it might sound like science fiction, this is the vision of many of the current advancements in Artificial Intelligence, although not yet with quite the scope. The Dark Web Portal is an example of a system that is now being created that could eventually be integrated into a larger system with police and intelligence operations as its key focus. This would streamline communication amongst our myriad defensive organizations and further enhance the security isolated systems now provide. If this system were to be rolled up into a greater system as an expert system, or autonomous agent, it would enable the central system greater functionality with access and control from on central location.

                As mentioned above, each representative system would be an autonomous agent, with its own algorithms and programming in place, providing all the calculations on the inputs and variables and sending them as output to the central system. This is and extrapoliation of the current Object Oriented Development structure we are currently moving in Information Systems, only on a much greater scale. Each independent structure, as an autonomous agent using Artificial Intelligence, could also be compared to an individual biological entity, like a cell, the sum of which makes up the organism. If we extrapolate this concept we approach another major focus of modern Artificial Intelligence development, Emergence.

                Emergence is defined as:” In 1999 Jeffrey Goldstein defined emergence as "the arising of novel and coherent structures, patterns and properties during the process of self-organization in complex systems. An emergent behavior appears when a number of simple entities operate in an environment, forming more complex behaviors as a collective. Emergence occurring over disparate size scales is usually based on causal relations across different scales formed by top-down feedback in the systems with emergent properties.” (University of Illinois at Chicago)

                This is an incredibly important concept, and the source of a hundred hours of entertainment in books and film. What this fundamentally describes is that as the system continues to grow, learn, and evolve in each of its individual parts, the sum of its parts can begin to develop an entirely new dynamic of interaction and even understanding. Essentially, if we perhaps expand the concept a bit, even self awareness. But we will be evaluating the concept of self awareness later in this discussion; this is just an example of the possible outcomes.

                The, perhaps more realistic, other outcomes of this “emergence” is the potential of  all systems designing efficiencies and self generating code (Through the use and enhancement of current shell creating technologies like those in lower CASE repositories) in such a way as to grow and maintain itself beyond current manual maintenance capabilities. With all the automation being required and implemented in even our current systems it is likely that the future of automation will present great strides in the capabilities of Artificial Intelligence applications.

                The four characteristics of automation (Genord 2002) and the advances in each characteristic will be intrinsic to the future development of truly capable autonomous agents. The first characteristic is capability, which applies to the fundamental efficiency and power of the automation. This is what can it do, and how well can it do it, and how quickly can it learn (or be taught)? (Genord 2002) The second is context, how clear is the data, and how clear does the data have to be? This will be a cascading concept in advanced, integrated neural Artificial Intelligence networks. At the highest levels data may be well defined, while at lower levels the Artificial Intelligences will be using raw data to determine output. This is similar to current Enterprise Resource Planning systems that pass up raw Transaction Processing System data to the managerial support systems where the data is further refined and passed to the Executive Information Systems for final strategic analysis. Third is communication, at what level does input need to be passed and how integrated is the system into data sources and other systems?

                The communication element is another future application of Artificial Intelligence that contains very interesting possibilities. The capability of high level commands being issued to computers introduces a great deal of power into the hands of the end user. If we could, for example, query a system out loud, in speech, and with common words, it would be a simple matter to run even extremely complex sytems with maximum efficiency. Consider an employee who could think at millions of times the speed of a regular person, with access to all the data in a company instantly (relatively), and whom you could ask a question and have all the results with that level of integration and efficiency. They would be invaluable, and anyone could access the data easily. This is the goal of advanced speech communications input and output. The research applications are toted by the researcher Balleste “The idea of artificial intelligence (AI) is fascinating and at the same time full of endless possibilities. It is fascinating to dream of the many features that computers powered by artificial intelligence will offer us in the years to come. This idea of highly advanced computers being able to interact with humans creates significant opportunities for (librarians). Imagine all the possibilities that await us in the next 10 years and beyond.” (Balleste 2002)

                And the final characteristic of automation ties closely with our discussion of communication, which is the amount of imagination and skill necessary for the end user. With speech input and processing the end user would not have to be an expert programmer or even have a college degree, they would just need to know English. This level of interaction will improve efficiency dramatically, while enabling a much greater degree of high level interaction. Imagine a network explaining its own infrastructure to a new employee for example. This would save the training time of the employee and reduce the time on the part of the architects and veteran software engineers in teaching the new employees.

                It is the future of automation which we can see the development of Artificial Intelligence as artificial or autonomous agents that will operate and “evolve” beyond their initial programming parameters within the environment of the overall Artificial Intelligence System. An autonomous agent is defined as: “An autonomous agent (Franklin and Graesser 1997) is a system situated in, and part of, an environment, which senses that environment, and acts on it, over time, in pursuit of its own agenda. In biological agents, this agenda arises from evolved in drives; in artificial agents from drives built in by its creator. Such drives, which act as motive generators (Sloman 1987) must be present, whether explicitly represented, or expressed causally. The agent also acts in such a way as to possibly influence what it senses at a later time. In other words, it is structurally coupled to its environment (Maturana 1975, Maturana and Varela 1980). Biological examples of autonomous agents include humans and most animals. Non-biological examples include some mobile robots, and various computational agents, including artificial life agents, software agents and many computer viruses. We’ll be concerned with autonomous software agents, designed for specific tasks, and "living" in real world computing systems such as operating systems, databases, or networks.” (The University of Memphis)

                To further extrapolate the metaphorical and perhaps eventually literary connection between biological complex systems and complex information systems we can develop upon the idea of “self-awareness” in information systems using Artificial Intelligence. With autonomous agents operating independently in a vast array of functions, each enabled with their own programming directives and a degree of latitude in their core functionality directives dependent on their specific environment and developing new input and output protocol and variable parameters in relation to said environment there arises the possibility of an “evolution” of thought. Let us establish a test example of a possible evolutionary scenario using the Dark Web model.

                In the pattern recognition software used for predictive modeling a learning algorithm will gather data and using preprogrammed decision structures for comparison will evaluate the dataset for a possible outcome, the output of the software. Now with advanced learning algorithms it is possible to have this input comparison be modified internally by the system itself. As this process evolves, and here is where we start to get theoretical, the system may begin to consider itself as a variable on the surrounding environment. Although we can now program discrete self affect inputs in systems this is not dramatic, but with a system capable of continuous self consideration there arises the possibility of a logical leap in the attempt to establish all variable weights relevant to the systems self in the core motive of accurate output.

                The distinction between discrete and continuous though is clarified by a Yale professor as  “the "cognitive continuum" that connects the seemingly unconnected puzzle pieces of thinking (for example analytical thought, common sense, analogical thought, free association, creativity, hallucination). The cognitive continuum explains how all these reflect different values of one quantity or parameter…” (Gelernter 2007)

                Geletner goes on to describe how the human mind can instantly distinguish faces and other variables without conscious effort on the process. In order to evaluate these continuous inputs a machine must establish an entire process and expend energy and memory to follow these cues. However this is just the kind of distinction that current researches into metacognitive loops are trying to overcome. (Anderson and Perlis 2005) With autonomous agents operating independently with these learning loops in place this kind of continuous processing becomes a possibility due to the rapid learning and storage of these processes upon completion of the first iteration. The concept of brittleness (Anderson and Perlis 2005) is considered the main problem in current Artificial Intelligence development. Brittleness is described as “a system designed for specific tasks fails utterly when faced with unanticipated perturbations that take it even slightly outside its task specifications. Yet humans perform admirably under such perturbations, easily adjusting to most minor changes as well as to many major ones.” (Anderson and Perlis 2005)

                According to these researchers (Anderson and Perlis 2005) the solution to the errors caused by perturbations in the variable and data streams is to increase the recovery of performance after the introduction of a perturbation, or to create “Perturbation tolerance”, where a system not only recovers quickly without a noticeable delay in processing but writes the perturbation into its modules in order to evaluate the perturbation data in the same way as the datasets that were originally within its parameters.  These automated modifications to the system could be as simple as recalibrating its sensors or as complex as changing rules of inference, training or retraining its processes, or learning new words and concepts. (Anderson and Perlis 2005)

                The solution to implementing this concept is called the metacognitive loop, which is aptly defined as “general approach to building perturbation-tolerant autonomous systems, based on the conviction that artificial agents should be able notice when something is amiss, assess the anomaly, and guide a solution into place. We call this basic strategy of self-guided learning the metacognitive loop; it involves the system monitoring, reasoning about, and, when necessary, altering its own decision-making components. “(Anderson and Perlis 2005)

 As we have seen this adaptive and learning decision process is what Artificial Intelligence is all about. With the introduction of metacognitive loops we establish a process by which these systems will emulate the learning of sentient organisms. As babies we are processing stimulus (inputs) and developing the response to these stimulus based on reinforcement. As we grow and develop into children we begin to conceptualize these decisions based on motivation and the further growth of our external perception. Artificial Intelligence is now approaching the childhood stage. With the lack of need for positive and negative reinforcement the essence of the question of self awareness is that Artificial Intelligence, even with the learning of variables and the overcoming of perturbations, is not yet capable of viewing itself. And so where can these metacognitive loops come into play in our discussion?

                The answer is theoretical, but potentially the ability to incorporate each artificial agents learning into the higher level command and control system will enable the upper level system to become aware of said agents as a part of itself, and begin to establish variables based on self awareness, “necessary motivational properties of self-representing mental tokens”  (Anderson and Perlis 2005) This concept os expanded to consider whether the agents acting on behalf of our system will be capable of operating correctly at all without a concept of self. Consider a person who has an itch. Due to proprioceptors (sensing self) allowing that person to understand not only where they itch, but that it is they who itch, they can effectively deal with the scenario and scratch that itch. Now consider our autonomous agents, receiving data and inputs and processing on behalf of the whole system. If they are subject to affect the system yet does not have a way to relate to their location in the system how will these agents deal with the scenario? Hence self awareness goes from being a science fiction concept to the reality of efficient system design. (Anderson and Perlis 2005)

                And these tokens are critical to the application of the metacognitive loop. The metacognitive loop is a new approach to Artificial Intelligence applications that involves a ‘triadic’ architecture which has trainable modules being controlled by a training module as in the “scruffy” Artificial Intelligence model which postulates that each module would be tweaked until efficient and the third tier Metacognitive Loop which would control both the training and trainer module through symbolically laden control libraries that could establish and retrain the training module based on scenario specific processes, this represents the ‘clean’ model of artificial intelligence which runs an inference engine that runs through scenarios found in a knowledge base. The metacognitive ‘triadic’ model is the combination of these two traditional models and runs the exceptional metacognitive module as reference for all other modules, and creates entirely new and realistic advances in the concepts of self awareness, efficiency, modular emergence and learning processes as the metacognitive module is able to assess and adapt to problems as they arise by passing the solution to the relevant module. Each of these training modules will possess the proprioreceptor token enabling the analysis of the systems perturbation and passing it to the metacognitive loop module which will identify the solution as it relates to the particular module. (Anderson and Perlis 2005)

                This does not mean that all these systems will operate independently. Of course our knowledge base and base level programming and integration must be manual. However even the metacognitive module will be required to ask for clarification from the user and implement new scenarios or categories of scenario into its analysis for more complex problems. So the answer will not always be to evaluate the perturbation and slot in the appropriate solution immediately. This links to our discussion of high level input and output as at this point the system would ask for clarification in speech format and implement the solution through speech input. This is truly an interface full of possibilities. The full impact of this could be clarified by the example of an archaeologist operating a lower level module based on archaeology discovering a new civilization. In this event, with no reference to draw on, the system asks the archaeologist outright for relevant data, which is given back from the archaeologist himself without any programming involved. This data is then inserted into the knowledge base and taught to the trainer modules by the metacognitive modules. From this point forward not only the archaeology module but all of the modules in the system will have a point of reference in the event this new civilization ever arises in any other capacity. So if a geneticist is later evaluating DNA found at the site the data would be incorporated seamlessly into the knowledge base in that category and then available to all other modules, and so on.

                This progressive accumulation of data into the entirety of the system is where the capabilities of Artificial Intelligence as a command and control system come back to the forefront of our discussion. With a wholly integrated system of progressively more specialized metacognitive modules you could eventually end up with the most efficient and capable systems the world has ever seen. If we go back to our earlier glance at the integration of multiple government based systems being integrated into a whole and apply the new models of development it is not hard to imagine a world where Artificial Intelligence could be involved in every aspect of governance and policy. Already we have Artificial Intelligence modules operating everywhere in our daily lives, as we discussed earlier. These systems would of course be integrated as well, enabling the central control system access to any data, anywhere, almost instantly.

                And so the concept of self awareness is distilled into an object oriented variable called a token which relates to the self of the system. This concept and its future development could lead to amazing degrees of interface and capability when cast in the light of the other future developments we have discussed. There is little doubt that a self aware system with advanced speech recognition and a network of expert sub Artificial Intelligence modules will be limited only be storage and processing capacity. And already the concept of etched atom processors and even neural cell processors present nearly infinite processing capabilities, and thus limitless application and potential performance for the future of Artificial Intelligence.

                If self awareness moves past the point of simple continuous processing of self evaluations and into an actual consciousness based on independently learned motives in the machine, we would truly be faced with a science fiction scenario. Consider if you will our super system in control of our entire country. This system now controls the military logistics and deployment, police activities, court records, medical records, and practically every aspect of our data and electronics. With the development of self awareness, and the probable core code of Asimovs Laws, what would happen if an objectively self aware and powerful system were faced with the reality of mankind?

                Asimovs First Law includes the principle that an Artificial Intelligence (or a robot) will not allow a human being to come to harm through the machines inaction. When faced with our justice system, politics, and constant bickering, what might the solution be? This is obviously entirely theoretical, but the future of Artificial Intelligence may be as our eventual governing body. This would be a government entirely objective, fair, and dedicated to the reduction of human suffering. Of course, now that really sounds like science fiction, as an entirely objective government would be too unlikely with human motivations. But with perfect logistics employed through instantaneous (relative) evaluation and decision making and pure motives we could currently be in the process of creating a revolution in our own society.

                The development of these components and revolutionary models to Artificial Intelligence are the potential threshold of the integration of data and communication on a scale that has heretofore been only the realm of authors and dreamers. The future of Artificial Intelligence in Information Systems is full of possibilities and extraordinary potential in the near future.

Bibliography

(AAAI) Association for the Advancement of Artificial Intelligence, “Brief History of Artificial Intelligence”

http://www.aaai.org/AITopics/bbhist.html

Lohr Robert, 2007. “The Chess Machine: A novel”.

Penguin Books, New York.

Wolfram math world: The web’s most extensive mathematical resource. “Boolean Algebra”

http://mathworld.wolfram.com/BooleanAlgebra.html

All About circuits. “Boolean Arithmetic”

http://www.allaboutcircuits.com/vol_4/chpt_7/2.html

HOCH. “History of computer hardware”

http://www.willamette.edu/~gorr/classes/cs130/lectures/history.htm

Biographies of women mathematicians. “Ada Byron Lady Lovelace”

http://www.agnesscott.edu/lriddle/women/love.htm

University of Michigan. “Historical mathematics collection”

http://quod.lib.umich.edu/u/umhistmath/

New York Times Books. “The Modern Library's Top 100 Nonfiction Books of the Century”

http://www.nytimes.com/library/books/042999best-nonfiction-list.html

DLSI. “McCulloch and Pitts' neural logical calculus”

http://www.dlsi.ua.es/~mlf/nnafmc/pbook/node10.html

iBiblio. “Internet Pioneers: Vaanevar Bush”

http://www.ibiblio.org/pioneers/bush.html

Weld Daniel, Etzoni Oren. “The first law of robotics: A call to arms.”

Washington state university, department of computer science, Seattle-WA.

http://cs.washington.edu

Auburun University. “Isaac Asimov's Three Laws of Robotics”

http://www.auburn.edu/~vestmon/robotics.html

J-Paine. “The logic Theorist”

http://www.j-paine.org/students/tutorials/tute/node11.html

University Of Toronto, Canada: Artificial Neural Networks.

http://www.psych.utoronto.ca/users/reingold/courses/ai/nn.html

AI dot com. “Applied philosophy of artificial intelligence”

http://www.a-i.com/show_tree.asp?id=14&level=2&root=12

Daniel Jufrawsky. “Speech and Language processing”

http://www.cs.colorado.edu/%7Emartin/slp.html