Another significant requirement is the need to find an efficient method for reducing the amount of computational searching for a match or a solution. Considerable important work has been done on the problem of pruning a search space without affecting the result of the search. One technique is to compare the value of completing a particular branch versus another. Of course, the measurement of value is a problem. As real-time applications become more important, search methods must become even more efficient in order for an Al system to run in real-time.
Natural Language Processing
There has been an increasing amount of work on the problem of language understanding. Early work was focused on direct pattern matching in speech recognition and parsing sentences in processing written language. More recently, there has been more use of knowledge about the structure of language and speech to reduce the computational requirements and improve the accuracy of the results. There are several systems that can recognize as many as several thousand words, enough for a fairly extensive command set in a “hands busy” application, but not enough for business text entry (dictation to text).
A number of production natural language command systems are capable of understanding structured English commands. These systems are context-sensitive and require that a situation-specific grammar and vocabulary be created for each application.
An obvious application for Al technology is in the development of software without some of the more tedious aspects of coding. There has been some research on various aspects of software program development. Arthur D. Little, Inc. has developed a structured English to LISP compilation system for a client and an equivalent commercial system has recently been announced.
It should soon be possible to build a Programmer’s Assistant that will assist in the more routine aspects of code development although no development has apparently been completed beyond a system that assists in the training of ADA programmers and a prototype system that converts a logical diagram to LISP (Reasoning Systems). True automatic programming that will relieve a programmer of the responsibility for logical design seems to be some time in the future.
But ‘Automation’ in AI from self-learning systems to systems that are designed to assist or automate certain tasks have come under fire for their potential to be flawed. It is very difficult for the flawed or biased human mind to create and automate a system, process or task without the possibility of including some of our inherent flaws and biases. Thus, it has been challenged recently that these AI solutions being developed for automation may be including flawed logic elements.
This has been proved, to a degree, by some of the Microsoft attempts to automate chatbot learning by exposing it to content on the internet or at least content within Twitter. The system very quickly picks up on biased views and can quickly become very politically incorrect in a short period of time.