Disciplines

Artificial intelligence can be broken into several different disciplines. Each is unique but often they intermix to accomplish a programming task and the differences become fuzzy. The different disciplines are expert systems, natural languages, simulation of human sensory capabilities, robotics, and neural networks.

Expert systems: Expert systems are also known as knowledge-based systems. They are computer systems that rely on knowledge base of rules that pertain to a specific application area. Expert systems are in wide use. Bank loan officers use expert systems for guidance in approving and rejecting loan applications. The military uses expert systems to analyse battlefield conditions and make tactical suggestions. Expert systems are used in a broad range of disciplines, including automobile repair, medical diagnosis and oil exploration.

Financial planning, chemical analysis, surgery, locomotive repair, weather prediction, computer repair, trouble-shooting satellites, computer system configuration, operation of nuclear plants, newspaper layout, interpreting government regulations, tax preparation, and many others. The intelligence of expert systems arises from the fact that they are able to solve problems. They are able to break a large problem into smaller parts in order to come to a solution. They understand the information given to them by the user. If the information is contradictory or ambiguous, the expert system asks for clarification or more information. If an expert system is faced with a new problem, it will save the information and the solution the user chooses for future use. Expert systems are not self aware in the human sense but are intelligent agents with a purpose.

Natural language systems: Natural language systems are systems that enable computers to accept, interpret and execute instructions in the natural language of the user. The intent of natural language systems is to create a more natural interaction between human users and computers. Database query is the area that has benefited the most from natural language systems. Other areas appropriate for natural language include machine translation, summarising and searching bibliographic texts, and analysing style and sentence structure.

Human sensory simulation: Simulation of human sensory capabilities focus on seeing, hearing, speaking and touching. Let us focus on two systems — voice recognition and vision input systems. Voice recognition systems allow the user to speak to the computer. Vision input systems are used to enable computers to see and understand it’s environment. Vision input systems are still in their infancy. The storage space and processing speed needed for a fully functional vision input systems are too restrictive. The human vision process is extremely complex. One commercial use of visual input systems is in assembly line inspection robots. This type of system will have a camera placed over the assembly line to inspect parts for defects. The digitised camera image is compared to an image in a database. If a defect is found, the system alerts the main assembly computer to bring it to the human operators attention and appropriate action can be taken.

Robotics: Robotics is the integration of computers and robots. Robots are taught to perform repetitive tasks. Intelligent robots incorporate the other disciplines of AI. They use human sensory simulation for touch, sight, and hearing. A useful robot in use is a security robot for warehouses. They are given an internal map of the area they are to patrol. They listen for abnormal sounds and look for intruders or fire. If an abnormality is found they contact the authorities and employees. Another useful robot collects aluminium cans and trash in office building after everyone has left for the day. The basis of a robot’s intelligence is in its programming. A knowledge base may be used for the robot to figure out what to do if it finds a fire. They must be aware of their surroundings and where they are in their surroundings. They will faithfully serve their particular purpose.

Neural networks: Neural networks are a fairly new addition to the field of artificial intelligence. Neural networks are a simulation of the processes of the human brain. Neural networks are not very good at solving simple problems. They are best at solving complex problems that cannot be solved by a simple algorithm. They are also good at solving problems whose input is noisy. Neural networks can be built to solve specific problems, but they are most useful for their ability to learn. A string of inputs is introduced to the network along with the desired result. The network will form the solution. For this reason, neural networks are the masters of pattern recognition. Incorporate fuzzy logic and this ability is increased tenfold. The major fields of interest for neural networks make use of the neural networks ability to filter noise and recognise patterns. These fields are handwriting and speech recognition, and predicting stock patterns. Neural networks are also found in fields such as finance, reading IRS tax forms, defence systems and vehicle control.

Scientific progress

The fields of brain building and robotics are closely related. A brain requires a body through which to experience and interact with the world. A body requires a brain before it can do anything useful. Many artificial intelligence scientists now believe that true artificial intelligence will only emerge through the evolutionary development of autonomous mobile robots. And in order to create human-like intelligence, the robotic body too needs to be humanoid. This is why all of today’s major brain building projects also involve the building of a humanoid or some smaller mobile robot. The line separating brain building from robotics is naturally blurred.

Utah-brain project: This research group at Utah State University is currently building an artificial brain for a robotic kitten. The aim of this project is to build an artificial brain that will control the behaviour of a life-sized robotic cat. The brain will contain an artificial neural network of 75 million neurons that live in a specially built highly parallel computer. The kitten will be able to see, hear and feel. The artificial brain via a radio link will remotely control its behaviours such as walking and playing. All of these behaviours will be evolved rather than pre-programmed by a human operator. December 1999 saw the completion and delivery of the brain hardware. Encouraging results were found from the experiments conducted to understand the machines capabilities. The main task of 2002 is to design and evolve 64,000 or so modules that will make up the artificial brain. It is hoped that the intelligent act shall be completed in a few years time.

COG and Kismet, MIT: COG is a humanoid robot developed by the Artificial Intelligence Laboratory at MIT The robot is used as a platform on which to bring together and explore the many sub field of human and AI. It consists of a head, torso, arms and hands, has no legs or a flexible spine. It is equipped with sensory systems. The aim of this project is to artificially reproduce the behaviour of a two-year-old child. COG’s artificial brain is made up from a mixture of different processing components.

ASCI White, supercomputer: This computer, located at the Lawrence Livermore Laboratory in the USA, is currently the world’s most powerful operational supercomputer. It contains 8,192 IBM processors and can perform 12.3 trillion operations per second. This is about 1/1000th of the estimated computational power of the human brain

Living neural networks: Scientists at the California Institute of Technology are attempting to build an artificial brain from real living neurons. The neurons are dissected from the brain of an embryonic rat and grown in a petridish. The petridish contains an array of electrodes that enables the scientists to stimulate and communicate with the individual neurons.

The Honda robot: Honda of Japan have built a couple of humanoid robots. The latest, the P3, is about 6’ tall and looks like a human wearing a spacesuit. The robot is able to walk and climb stairs. It’s brain, however, is still very primitive. Scientists at Carnegie Mellon University are currently improving this by developing its vision and navigation system.
All of the above mentioned systems are intelligent. They are in their infancy but the potential for growth is evident in the advances made over the last 10 years. At present, AI computer systems meet the attributes of intelligent agents. They learn, reason, adapt, solve problems, understand, use language, perceive and model the external world.