There are two main schools of thought in regards to the technological processes in creating artificial intelligence: the symbolic approach and the sub-symbolic approach. These two camps are very similar in belief to those who differ on the requirements of determining a human-level intellect. The symbolic adherents attempt to recreate the processes of the mind using a psychological replication – by reproducing the capabilities of the human mind – and those of the sub-symbolic school seek to follow a process of biological replication, creating the base of the human mind and letting the program act on those systems. This argument is further complicated by the debate over whether an AI needs a body and sensory input to complete its conception.
The Symbolic Approach to AI
Symbolic AI more closely resembles what Turing had in mind when he proposed his test of machine intelligence. All intelligent reasoning is nothing more than a manipulation of symbols in the mind, and human beings are essentially nothing more than computers made of organic materials. Other aspects of what we would call humanity, such as emotions or self-awareness, are merely the results of combinations formed at the root symbol manipulation process. Therefore, if a programmer could find all the pieces to the puzzle of the human mind, putting them together would only be a technicality.
To the symbolic programmer, constructing an AI consists of achieving one basic goal: human problem-solving techniques must be simulated so that the AI is able to learn through logical means. Everything else will follow naturally. This technique has already shown some success in the increasing complexity of knowledge-based systems (such as expert systems) and, theoretically, if one could cram a program with enough information the program would achieve human intelligence.
The Sub-Symbolic Approach
But can human intelligence really be reduced to pure symbol manipulation? Those who adhere to the sub-symbolic school believe that is not so. The theory that drives them lies in the belief that computers must be able to develop their own intelligence by using a base that closely resembles the workings of the human mind. If the system itself can be replicated, intelligence and self-awareness will eventually follow. Sub-symbolic adherents are big supporters of using neural networks and genetic algorithms to achieve their goals, and these types of systems are gaining quite a bit of popularity in the field of artificial intelligence.
Closely related to the sub-symbolic approach is that which preaches the necessity of direct perception and interaction with the environment. In this theory, the aspects of life that humans may take for granted, such as movement and perception, lend themselves to internal visualization and thus self-reflection. Of course, the symbolic approach to AI would suggest that none of these things is needed other than possibly a reproduction of their effects in program format.
Evolving AI Technology
The desire seen in scientists and the commercial public (the target market for AI products) is becoming more focused on the realm of socially interactive AI. Pushing programs to do more work is still the foundation of most research, but the potential for an AI to be human-like and, thus, more accessible in an emotional and intuitive way is making its way to the forefront of design. This desire has brought together experts from many different fields, such as social behaviorists, biologists, computer scientists and mathematicians, to create advances in the realm known as biologically-inspired computing.
The Growing Trends of Biologically-Inspired Computing
Biologically-inspired computing is a “ground-up” approach to developing AI similar in thought to the sub-symbolic design. In fact, most of the developments in this sub-field lend themselves to the efforts of sub-symbolic programmers. The theory purports that, by using computers to model the processes of nature, it can be better understood, and nature, in turn, can be studied to create models on which to base new computer technologies. If nature knows what it is doing with regards to intelligent life, copying her is the best way to go about creating intelligent life.
There are many specialties within the realm of bio-inspired computing, each with their analogical equivalent in biological sciences such evolution, biodegradation, neuroscience and studies of the immune system and sensory organs. If there is a part of nature (human or otherwise) that is being studied, there is most likely a digital equivalent.
Biologically-Inspired Computing at Work
The most obvious strength of bio-inspired computing models is that the computer has the power to teach itself. This is in contrast to the “creationist” methods of traditional AI where the computer only knows what the programmer allows it to know. In addition to creating less work for scientists, bio-inspired experiments often end in results which are quite unexpected and very enlightening.
One need only specify a simple set of rules and create a method for applying those rules and then let a simple set of digital organisms which adhere to those rules loose in the system. This method lays the basic groundwork for progress while not giving the computer strict instructions and code to follow. There is a goal in place and, led by a reward system, the program must decide how to achieve that goal. After several generations of rule application, complex behavior begins to emerge within the organisms – the program begins to figure out how to play the game and the best way to solve the problems involved. The end result is a population of digital organisms which are smarter for all their hard work.
The main biologically-inspired computing techniques consist of neural networks, genetic algorithms and the growing trend of robotics. Each has its own distinctive strengths and weaknesses as well as its own set of complications which must be overcome before it can be perfected. Combined, it may be that these three systems are the answer to formulating a working model of strong AI.