Language Mapping and Artificial Intelligence from MindMeld: CEO & AI iBook
By Thomas B. Cross @techtionary
The following is an excerpt to Chapter 4: Expert Systems – Language Mapping of MindMeld: CEO & AI Merging of Mental & Metal book available now from iBooks –
Book Review – “As the CEO of a energy industrial company and actively involved in CEO Leadership Forums I have been following AI for more than a decade. Indeed the promises for improving many technical tasks are interesting yet in reality often prove more complex to manage than proposed. MindMeld was very profound in proposing that AI starts not at the bottom of the organization but with CXO decision-making and worth reading by anyone in or rising to the boardroom.” George B.
Language mapping (charting) can yield both an effective methodology for language processing and a flexible approach to machine understanding and AI. Mapping takes language out of the linear mold as well. Using language, like reading these words, is a relatively slow process that forces the reader through a step-by-step progression with minimum allowances for deviation. It is difficult to choose the following:
– A convenient means for starting and stopping
– Time indifference, the reader’s choice of when to read
– Random searching, the reader’s option to jump around in the text, front to back, or in any other way
– Convenient packaging, large amounts of information that can be moved over great distances with or without human assistance
– Extremely explicit dogma or propaganda, the exactness and completeness of the communication process
– Other issues that range from ambiguity to fantasy
The point is that although language formed by words has many key and important advantages, it does not reflect the full communications process. It does not convey how humans or other creatures understand, process, or consume information. Humans process concepts, not words. Concepts can include everything from hieroglyphics to integrated information processing. Computer analysis of language (computational linguistics) is the application of machine intelligence to problems, issues (thinking), and communications, rather than to the machine translation of foreign languages. It is often found that communicating fails, except by chance. Applying machine technology offers little hope of creating systems that can cope with the technical aspects of communication, much less with the more complex issues of understanding and the resulting rational action to be taken. However, the aim of language mapping is to develop models that explain both technical and human aspects of communication in terms a machine can accept as well as use to produce output as a human does. The first challenge is to analyze the existing language; an alternative approach is to abandon old languages for new ones. Without exploring language at this point, the process might simply be to design a metalanguage that adds valued meaning to phrases. More precise, however, is a metalanguage that a machine can process. This is not to say the machine understands as a human does. It assumes only that the language is processed, like a payroll system for example. Theorists have sought to create systems that rise above language, an approach that seeks to cope with context, which is difficult to do. The purpose of this metalanguage approach is to define what words actually mean; for example, the words “offensive reaction” have a dual meaning: striving forward or obnoxious. The context of the sentence determines which definition is appropriate. Another approach is to define language and understanding as computational entities, and to process these entities through an architecture and its associated building blocks. Expert architectures are combinations of single cells that together construct programs which start to “think.” Organizing these programs in other than a random manner allows the computer to consider the elements, or cells, contained in the system. As more and more elements are organized into systems, languages are developed with inherent models that have meaning associated with them. The arrangement of these cells clarifies the context of individual cells: “Mad” with “dog” = “rabid”; “mad” with “at him” = “anger.” When numerous elements are networked, the complexity vastly increases, and the system develops its own behavior. In the case of computer modeling of psychological processes, networks of behavioral elements are organized, just as they are in most games. Some early systems using this approach were reactive in the sense that they only responded to changing conditions and required no intelligence. Their response could be compared to that of a thermostat for a home heater. The next level is found in current computer programming technology. If-then- else statements allow and-or-not approaches to problem-solving. This process expands beyond the on-off stage to allow for branching, that is, choosing among options.
The next stages might approach thinking, where complex formulas, models, or computational heuristics evolve into seemingly intuitive thinking systems. At this level, there will be few, if any, rules; there will only be appropriate, if not humanistic, approaches to the issues at hand. As theorists have discussed, one person’s meat is another person’s poison.