Artificial intelligence - The prospects for AI in the TPF world
By Phil Adams

As application of Artificial Intelligence moves out of the research laboratories of Carnegie-Mellon and Stanford and into the business world, a rapidly growing number of companies are realizing the benefits of AI technology in their

data processing environments. The demystification of Artificial Intelligence for the business world has brought Al into focus as a practical tool for system development. The most visible portion of AI technology that companies are benefiting from now is knowledge-based or Expert Systems.

AI/TPF: The Prospects
Expert Systems enable a company to capture and automate in software an expert's knowledge. The benefits to the company include the fact that knowledge captured in the expert system never resigns, can be dispersed throughout the company, and the encoded expertise implements business policies and procedures in a more consistent manner. There are quite a few Expert System development 'shells' on the market which provide the tools and interfaces necessary to develop these systems. In the IBM world, these systems typically run on the mainframe under MVS, VM, IMS, and CICS. They also have counterparts on the PC running under DOS and 05/2. Most of the tools provide components to link an expert system to external interfaces such as DBMS's, users, and existing software. A few examples of these tools are AION's ADS (Aion Development System), Inference's ART-IM (Automated Reasoning Tool for Information Management), IBM's TIRS (The Integrated Reasoning Shell), and AICorp's KBMS (Knowledge Base Management System).

Another way to develop expert systems is to use an AI-oriented language such as LISP, 0PS5, or PROLOG. While the languages do not provide all the features the development shells provide, they do facilitate the development of Expert Systems. Compilers are available for 0PS5, OPS83, Borland's PROLOG, NASA's CLIPS, and LISP on a variety of hardware/operating system platforms excluding TPF. The question that arises is, why hasn't AI technology been applied to a TPF development tool? It would appear that there is a market for such tools since many of the world's largest companies currently use TPF systems. Many of these companies already use Expert System's tools in non-TPF environments. Would it not be advantageous to apply AI technology to the TPF environment?

The software products and educational seminars provided by AMDAHL, TPF Software Inc., and other TPF tools and services vendors suggests that TPF is a viable operating system for now and the future. With the recent release of third generation language compilers for TPF, perhaps the market is ripening for artificial intelligence to enter the TPF arena.

AI/TPF: The Dilemma
Why has an Expert systems tool or language for TPF not been developed? One reason could be the functional nature of TPF applications. Traditionally focusing on creating, sorting, and summarizing of data, TPE applications have essentially been skeletal in nature. TPF file maintenance and other 1/0 intensive utilities demanding a larger percentage of system resources are usually run at off-peak hours.

A typical TPF application provides create, display, modify, delete and utility operations and includes a measure of logical validation of data. This type of application is not computationally intensive. Expert system applications are computationally intensive and demand more system resources than TPF applications normally require. TPF applications do not usually venture into intensive data analysis of the type found, for example, in an insurance company's expert underwriting system. TPF applications typically are not used for 'judgmental' types of processing.

While a rule-based application of this type may require inordinate amounts of CPU time (not to mention I/O), there is merit in applying expert system technology in TPF systems. For instance, if a TPF airline yield management system could be directly linked with the fares and scheduling packages, real-time response to market fluctuation could be realized. The strategic power of such a system is yet to be fully implemented in the TPF world where such applications are shifted to other operating systems in the shop.

The use of expert systems outside the TPF environment does not exclude the opportunity to realize real-time market response in on-line TPF applications. Expert systems outside of TPF with a 'soft link' to on-line TPF applications are an alternate to a real-time TPF tool. Using carefully implemented interfaces and selectively chosen applications, this link can provide a business with the benefits of expert systems in the TPF environment.

From the above, it may not appear that TPF is a candidate for expert system development running under the TPF operating system. However, TPF is in the running for application of Al within the environment. Consider the building blocks of expert systems for a moment.

Probably the most powerful mechanism for representing the objects within a domain is a concept called object-oriented programming. The uniqueness of this paradigm is that both the data describing an object and the operations performed on that data are defined at the object level. In other words, the object 'knows' how to manipulate itself. Objects not only store data about an entity, but also how the data is used and how the object interacts with other objects.

The object concept also supports encapsulation whereby the process manipulating an object does not need to 'know' about object-specific implementations. Construction of hierarchies representing complex structures difficult to represent in the procedural data processing world is possible with objects.

Representation of knowledge in an expert system typically takes the form of IFTHEN rules, sometimes called 'productions'. Expert systems are also referred to as 'rule-based' systems for this very reason. The rules form the basis of manipulating objects in the expert system.

Determining which object to 'activate' or which rule to 'fire' at a given time during the execution of an expert system is the responsibility of the 'inference engine'. The inference engine is composed of problem solving algorithms and the processes to execute the algorithms against the rules and objects in the system. The processes apply the problem solving algorithms based on the state of the system and the construction of the rules.

An example of TPE system packages that can benefit from the object concept is the network control program. TPF was originally designed to provide high-speed communications facilities in its network control program. Bjarne Stroustrup used the object concept to extend the C language into the C++ language standard at AT&T's Bell Labs in order to model a large telephone switching system. The type of conceptual designs required by a telephone switching network are not unlike the communications network management required in TPE. As TPF is applied to larger networks such as that of AMADEUS, and as distributed processing continues becoming more effective, real-time 'smart' or rule-based network management may become a method of choice.

Design and implementation of expert systems, like a network manager, requires tools that support the conceptual level of design as well as facilitating the transition from conceptual design to efficient implementation. Unfortunately, the TPE market is devoid of the kind of tools necessary to implement such systems. The situation appears to be improving now as more sophisticated development tools are introduced for TPF and as the move to the C language gathers momentum.

A more strongly-typed language such as C will facilitate the development of an expert system tool for TPF. While a tool in TPF may require the sacrifice of some of the developer's interfaces of traditional expert system shells, support for the elegant and robust functionality is still feasible.

AI/TPF: The Current Status
The culture of TPF shops has not traditionally nurtured the move towards expert systems development. The bias towards traditional development only recently opened up to application of modern methods and CASE technology. As TPF shops gradually open themselves to new development cycle methods, this situation will probably change as well.

The development of AI tools for the TPE environment and linkage of expert systems to TPF appears to be a future prospect for the field. As companies recognize the need to use their data processing investment as more of a strategic workhorse and not just an office automation tool, the need for realtime 'smart' systems increases. The continued increase in hardware capability will remove some of the obstacles to implementation of these systems.