TELECOM and AI: The Emperors New Clothes?

1 Introduction

In R&D the telecommunications industry employs world­ wide more than 500 000 engineers. Whereas in the past most of their efforts were directed at hardware, current trends show the increasing importance of software technologies. Most of the well-known research laboratories support AI divisions working at the cutting edge of this technology. The reasons for conducting research in AI are manifold.

Public switching systems are among the most complex ma­ chines ever built. Despite their deterministic local behavior, we need powerful and proper AI techniques in order to fulfill demands for the relevant services in an optimal way.

Most telecom companies are actively involved in develop­ ing telecommunication management platforms as a step in ad­ dressing global fault, performance, security and configura­ tion problems. Work is under way to incorporate AI techniques into these platforms. However, the panelists cur­ rently identify a gap between promising AI research results and their application in telecommunications.

2 Current State of the Art in Telecom Greg Vesonder

It is fitting that the state of AI and Telecom be reviewed since it has been ten years since the first AI and telecom paper was presented at IJCAI-83 in Karlsruhe, Germany. At that time we introduced ACE, a diagnostic expert system for pinpoint­ ing problems in the local loop. The last ten years has shown a flurry of activity in AI and in telecom. There have been con-

• Fuzzy systems for network management
• Very large knowledge bases

It will be shown that all these techniques are necessary and partially effective in certain areas. Finally the following ques­ tion will be addressed: .. Are these techniques the most impor­ tant ones for next generation telecommunication network management ? ...

3 Difficulties and Obstacles in Applying Ex isting AI Technologies Werner Remmele

Siemens AG is a major supplier within the telecommunica­ tion market for both public and private communications sys­ tems. With DM 18.6 Bill. in 91/92 almost 25% of Siemens sales were within the two communication groups of the com­ pany. Obviously, there is a strong necessity for ongoing in­ novation of their technologies.

Some of this R&D work is carried out in the company's corporate department for Research and Development (ZFE), which is responsible for innovative technologies, and their transfer into the different groups. There are strong links be­ tween the communication groups and the department for Knowledge Based Systems within ZFE, and some projects have been completed jointly.

We describe briefly two expert systems that Siemens is us­ ing in the telecommunication area, one for support of human interface design and one for configuration of switches.

Hicom, the Siemens private communication system, is a highly flexible system which can be configured customer­ specifically to meet a wide range of different needs, depend­ ing on the customers language and country, number of exten­ sions and many add--on features. Consisting of approximately 300 different hardware boards and vast amounts of software, its configuration requires substantial knowledge on the inter­ action and dependencies of the· various components.

Siemens is using an expert system, XPS-configurator, to support engineers in selecting the customer specific compo­ nents and check the resulting layout for consistency. The sys­ tem is well integrated into the engineering process from ten­ der formulation to hardware assembly and has been in routine use for several years.

The Siemens digital telecommunication system, EWSD is in world-wide use at PTT administrations. Its operator inter­ face is the Man Machine Language (MML) according to a CCITI recommendation. Siemens MML currently comprises about 2000 commands and 4000 output masks. The language is continually adapted to technical innovations in EWSD. To ensure user-friendliness, new or changed commands need to conform to well defined standards of consistency and quality.

We developed an expert system, CODEX, to support the design of MML commands and enforce the relevant regula­ tions. Consisting of an MML-checker and an MML-advisor, which contain knowledge of the relevant design regulations, it interactively helps language designers to meet all require­ ments. Ultimately, the system contributes to the high MML

quality which in turn makes the operators work smoothly and efficiently.

While in the 80s AI techniques were expected to penetrate all areas of computer applications, their actual deployment has happened more slowly.

Despite the examples cited, the tremendous potential in telecommunication for such engineering tasks as design, diagnosis, monitoring, but also image processing has not been matched by a corresponding record of installed AI solutions. Indeed, we are far from a routine use of knowledge based sys­ tems as a standard technique of software engineering. From our perspective, we found the following issues as most signif­ icant for the wide usage of AI:

• Integration of AI-techniques into traditional systems
• Request for high availability of telecom systems
• Usage of common platforms
• Heuristic approaches

One difficulty in integrating Al-techniques lies in the very high engineering investment represented by a modern tele­ communication system. The hardware and software develop­ ment effort invested makes it impossible to completely rede­ sign a whole system using the latest technology; instead one always has to live with large portions of existing code, and any AI solution has to fit in with it. Thus, although substantial knowledge is incorporated in any telecom system design and implementation, it is seldom in a form useable by a knowl­ edge-based system.

For this very reason, most real projects have chosen the ob­ vious approach of starting with a small prototype, e.g. diagno­ sis of one particular version of one component of a switch. This did work, but for a practically useful system a more com­ plete solution is needed, and when it comes to scaling up the effort in knowledge acquisition gets so high that it cannot be justified economically.

A different obstacle derives from the extremely high stan­ dards for system availability; these have always made fail­ safeness a requirement in telecom systems, and solutions like redundant hardware have been provided that leave little space for additional improvement by, e.g. knowledge based diagno­ sis.

Another obstacle that was relevant in the early days has been largely overcome recently -it is the isolation of AI sys­ tems in terms of the hardware and software platforms avail­ able. Indeed, in the old days of the Lisp machine it was as easy to create an interesting prototype as it was hard to transfer it to the real world computing environment of the telecom switch. Meanwhile, the problem has been addressed from both sides, i.e. AI has learned to run on standard hardware and integrate with standard languages, and telecom has adopted modern software concepts, like object oriented languages that make it easy to create advanced applications.

Another difficulty has been described by the term ,.first generation expert systems ... The fact that simple prototypes can be built with just about any tool obstructed our view for the difficulties of scaling up to large systems. Telecom sys­ tems are inevitably so complex and so rich in customer specific variations that only a model-based approach can make the effort in knowledge acquisition manageable. What we need is a principled method for using design knowledge consistently as the basis for different applications. Instead of developing a knowledge base a posteriori it has to evolve from the general engineering process.

Several ongoing developments in the AI area promise solutions to the problems stated above. In particular, the field of model-based diagnosis has produced encouraging results in building expert systems in a more systematic way from a library of component models for a technical domain. Tech­ niques of non-monotonic reasoning and qualitative models are applied here successfully.

Progress in knowledge acquisition has been made, espe­ cially in the KADS projects. Using a standardized approach to knowledge based system development such as Common­ KADS enhances both the predictability of the results and the integration with conventional and existing software.

Finally, the effort to standardize knowledge representation formats and allow the re-use and exchange of knowledge bases between different applications, is expected to yield sig­ nificant progress toward the economics and large-scale ap­ plication of AI techniques.

4 State, Difficulties and Future Directions Concerning AI in TMN

Franz Barachini, Georg Pauthner

Alcatel concentrated its AI efforts at the beginning towards the development of expert system shells such as SPOKE and PAMELA. The latter is especially oriented towards easy in­ tegration of expert systems into existing process control sys­ tems. Real-time features as interruptability and the imple­ mentation of a high performance inference engine have been realized within PAMELA.

Subsequently, expert systems for fault management (SAMPAN, ALAMOS) and automated generation of routing tables (ARX) of public switching systems have been imple­ mented. We made the experience that e.g. ALAMOS was per­ fectly shaped towards the problem. Nevertheless reusability, generality, knowledge maintenance and acquisition was still a problem.

Concerning the current state of network management sys­ tems we identify the following situation :

• The introduction of central network management facilities based on proprietary, vendor dependant approaches is finished
• The upgrade from proprietary solutions to TMN compliant ones is ongoing
• The most relevant features of network management sys­ tems today deal with the immediate operating and main­ tenance of a network, like alarm collection or remote op­ eration

For the features mentioned above there is no need for the usage of AI technologies. This is justified by the fact that from the experimental prototypes, exploring the possible benefits of AI techniques, very few of them have really been fielded. In addition some of them were used as marketing arguments only.

In contrast to the past, the future of AI technologies in tele­ com systems can be drawn in rather bright colors. Since the size and complexity of networks is increasing, provisioning of powerful support systems for planning and maintenance of networks will require the handling of large amounts of in­ formation being partly incomplete and inconsistent. Accord­ ing to the application areas ofTMN the use of AI can be for­ casted in the following areas :

• Fault management : diagnosis support by e.g. correlation of alarms or reasoning of the cause based on a model of the network
• Configuration management : population of routing tables based on rule sets or e.g. planning support for new net­ works
• Performance management : evaluation of test results and simulation of the network behavior
• Security management : detection of security bottlenecks or diagnosis support for security alarms
• Accounting management : no need for AI

Currently the Alcatel-Austria Research Centre is developing FAMAP, a generic alarm and fault management package. During our investigations we recognized that there is a need to modify a knowledge base not only by telecom suppliers but also by operators.For this reason mechanisms are provided, so that existing core knowledge can only be modified to a certain extend.

The main purpose of FAMAP is to allow the implementa­ tion of expert systems which are improving network avail­ ability by providing mechanisms for fast fault identification and by reducing the load on the operators. Moreover FAMAP components can be reused for different equipment and net­ work structures. A first expert system based on FAMAP is un­ der test in an operation and maintenance center for mobile ra­ dio.