Artificial Intelligence Development Hardware


An Al development system is primarily intended for use in discovering and characterizing knowledge, rather than for delivering it. In some applications, it may be feasible to also use the development of hardware and software in the fielded system. As noted below, this feasibility is influenced by various factors such as cost and physical environment. An analogy to development and delivery systems is the distinction between start-up and production tooling.

Dedicated Systems
There has been a preference for utilizing dedicated hardware and software for developing Al systems. The hardware is variously referred to as LISP machines, symbolic processors, or dedicated machines. The general attitude has been that greater performance, as well as increased efficiency in the development process, can be achieved by using hardware specifically designed for that purpose. This attitude has been reinforced by the continuing drop in prices of specialized machines. It had been necessary to invest from $75,000 to $150,000 in symbolic processors and software tools in order to do serious AI development work. For many organizations, this was difficult to justify, particularly since such systems could typically be used by only one person at a time. However, the use of dedicated machines is becoming more cost-effective, because the prices have been dropping rapidly. In addition to the benefits of processing power, graphics, and other architectural features, these dedicated systems frequently feature an extensive and powerful programming environment. They include software utilities and analysis tools which greatly expedite the development of complex knowledge systems. As compared to what is available on personal computers, the larger specialized machines can accommodate much larger knowledge bases and more sophisticated manipulation of knowledge.

Mainframes
Until about a decade or so ago, a mainframe computer, unless used in a dedicated mode, was not considered suitable for developing AI software. A general-purpose, time-sharing system would be overloaded by the large overhead requirements of an AT language development environment and the operating system so the economics favored a special-purpose machine.

One company was formed to develop and produce an Al-based system for training applications. The target market for this product were owners of large mainframe computers. Development of the product was carried out on a high-end LISP machine and successfully completed. Unfortunately, the developers were unable to obtain adequate processing speeds in the mainframe and the company was forced to restructure under rather difficult conditions for both the investors and employees of the company.

It is certainly possible to develop and deliver many Al systems on conventional computer mainframes or minicomputers. In fact, many of the original and successful applications of AI were developed on conventional computers using conventional software. Considerable controversy has been generated on the subject of using conventional machines. Proponents insist that conventional machines with good LISP compilers can run AI programs as well as specialized symbolic processors. This is disputed by many AI developers who question the quality of the software and development environment provided by conventional machine vendors. They also claim that there is too much reliance on third-party developers. The attitude of many AI systems developers is that symbolic processors are superior to conventional processors in development features and operating capability.

READ  Artificial Intelligence Patent? EPO Gives The Green Light! - Intellectual Property

The advantage of conventional machines, of course, is their widespread availability, familiarity and cost per user. They also have a better track record than the LISP machines for reliability, maintenance and other customer service features. Primarily for these reasons, LISP machines have had difficulty in penetrating the office environment. In addition to surmounting the status quo hurdles, users of LISP machines have had technical problems relating to connectivity and these machines have not been able to do some simple office functions.

Conventional mainframes will have an important role as delivery vehicles. Much of the data needed by knowledge systems is available only on mainframes so that cost and connectivity issues will drive the selection of these delivery methods.

Other users may insist on development on mainframes for security reasons. Many installations have elaborate security procedures already in use. Knowledge systems may, indeed, utilize sensitive information and it could be expensive to duplicate the same security procedures that have already been developed for the mainframe computers.

Hybrid Systems
As noted in my previous posts, most AI systems are now being developed for use in specific applications rather than for demonstrations or proof of principle. Also, the AI portion of the system may be embedded in a larger system built with conventional software methods. This requirement will normally force the developer of the system to use conventional types of hardware and software. However, there are some types of applications where the requirements of the application will force the user to use some form of a hybrid system where the symbolic processing is done on a specialized machine and the more conventional database and I/O operations on conventional-architecture machines. The problems of interfacing associated with the use of such a mixed system are being considered by some manufacturers of Al systems but much work remains to be done. Some of the particular applications where the capability of specialized machines will likely be required include:

1. Large knowledge bases
2. Stand-alone, single-purpose systems
3. Systems with real-time, complex reasoning features

Workstations
There is a pronounced trend, particularly for vendors of Al software tools, to port their LISP-oriented systems to widely available workstations. Although these workstations may not have all of the features of the larger symbolic processors, the necessary compromises are not too severe for many applications. Top-of-the-line engineering workstations are challenging market segments dominated by symbolic processors (Verity, 1986). These workstations have good graphics capability and considerable computer processing power. Importantly, they are oriented towards user requirements. They have a large 1/0 bandwidth and include such features as memory management and data manipulation. Most workstation manufacturers are working with third-party suppliers of Al systems and languages to provide a capability equivalent to that of symbolic processors on their machines. Providing a development environment equivalent to that of symbolic processors has been the most difficult problem for these third-party suppliers.

READ  Artificial Intelligence could help monitor jobsite social distancing | News

Personal Computers
With the continuing proliferation and increasing cost effectiveness of personal computers (PC), there is much interest in developing knowledge systems on these machines. Many useful applications can be developed on small machines. Implementation on a PC can also be helpful in defining the scope and establishing the feasibility of a project which will eventually be designed and built on a larger system.

Another application of a PC-based knowledge system is its use as a learning tool. These learning tools can be quite effective but, typically, they are capable of representing only a segment of what can be achieved with Al technology.

Migration
If the project plan does call for the more usual migration of the development system from a symbolic processor to a PC, a number of design factors should be considered. Subsets of development languages are available on the PC, which greatly assist in porting the system from the large machine to the PC. However, troublesome bugs in the software are possible when this porting takes place. Also, some of the features which would be useful on the large system are not available on the PC, perhaps resulting in a less-than-optimum product. Well-qualified knowledge engineers and software specialists are needed to ensure an effective transition from the symbolic processor to the small machine.

A particular problem is the user interface. It is difficult to obtain the same extensive graphics capability on most PCs as are readily available on symbolic processors.



READ SOURCE

LEAVE A REPLY

Please enter your comment!
Please enter your name here