Issue No. 01 - February (1996 vol. 11)
It is now more than 10 years since artificial intelligence techniques were first applied in telecommunications. That is a long time, given the current speed of technological change. It is reasonable, therefore, to consider some ways in which the application of AI to telecommunications has evolved over this period.
A decade ago, the AI approaches applied in telecommunications were primarily rule-based. The article "Tuning Numeric Parameters to Troubleshoot a Telephone-Network Loop," by Christopher J. Merz, Michael J. Pazzani, and Andrea Pohoreckyj Danyluk, describes a typical system in this category: Max, the underlying system that its approach helps tune. Apart from the AI techniques themselves, however, AI use 10 years ago was more of an evolution than a revolution. For instance, the typical approach was to create a trouble ticket in much the same way as a technician would have done. This ticket went through stages of processing eerily similar to the stages undertaken by manual methods. In other words, the initial approach was to make computers mimic humans.
However, computers are not human beings. Although initial AI applications in telecommunications have paved the way for what came after, they have also given us the legacy of piles of inconsistent trouble tickets that move from database to database. In other words, we have torrents of data and a trickle of knowledge. It is easy to criticize the past, but that is not my intent. Rather, my main purpose is to urge paradigm shifts concurrent with technology shifts.
"Data Mining and Forecasting in Large-Scale Telecommunication Networks," by Raguram Sasisekharan, V. Seshadri, and Sholom M. Weiss, addresses this by describing a proactive approach to network maintenance. Rather than focusing on hard failures that have already occurred, as is the case in typical systems, this approach monitors the performance of networks continuously, searches the resulting time-varying diagnostic data using machine-learning and correlation techniques, and generates a set of rules to predict potential failures before they occur.
Whereas that article explains how to take the warehouse of data that has been accumulated and extract information from it, "Tuning Numeric Parameters to Troubleshoot a Telephone-Network Loop" describes hill-climbing techniques for tuning parameters in the knowledge base of an existing diagnostic rule-based system. It is interesting that more recent techniques such as simulated annealing and genetic algorithms have influenced hill climbing so that a modified hill-climbing technique is used in this instance.
"Planning Cordless Business Communication Systems," by Thom Frühwirth, Pascal Brisset, and Jörg-Rainer Molwitz, describes constraint-based programming to optimally place base stations for wireless PABX (private automated branch exchange) using a branch-and-bound method. In this instance, although the technique itself has existed for some time, the application domain is relatively new. While the focus of the article is, specifically, the placement of base stations in a building to implement a wireless PABX system, this approach has relevance in the broader context of wireless systems in general—systems that provide, to areas of the world without a significant wired infrastructure, the capability to leapfrog using the telecommunications techniques of the next century.
On the flip side, it is disturbing that this approach potentially faces the same problems that have plagued such approaches in the past—namely that the optimization technique theoretically has exponential complexity. Additionally, the context, in terms of the correct grid size for the problem at hand, must be determined by trial and error. But perhaps we should not be too hard on the technique. Other techniques have faced combinatorial explosion and exponential complexity in various situations with regard to the application of AI. Yet those techniques have helped solve practical problems with reasonable success; no doubt this will continue.
The potential for AI application is significant, and vast, yet untapped areas remain. Typically, we consider AI techniques during the development phase of building telecommunication solutions. If, however, we explore such techniques more often during the specification or requirements phase, perhaps we will find many additional, and equally important, appropriate uses of AI in telecommunication systems in the future.
V. Seshadri is a distinguished member of the technical staff at AT&T Bell Laboratories. His research interests include expert systems, case-based reasoning, and machine learning. Most recently, he has been focusing on data mining in various domains, such as communications and retailing. He received a BTech in electrical engineering and computer science from the Indian Institute of Technology, a PhD in electrical engineering and computer science from the University of Notre Dame, and an MBA from the University of Pennsylvania's Wharton School of Business. Readers can contact Seshadri at AT&T Bell Labs, Middletown, NJ 07748; email@example.com.