Over the past few years the field of intelligent systems has seen the re-emergence of knowledge-based tools and techniques. It’s now generally recognized that artificial intelligence (AI) can provide knowledge-based support to well-bounded problems where deductive inference is required. We know that AI performs less impressively in situations with characteristics (expressed in software as stimuli) that are unpredictable. Unpredictable stimuli prevent designers from identifying sets of responses, and therefore limit the applicability of "if - then" solutions. We know, for example, that so-called expert systems can solve low-level diagnostic problems, but cannot predict the technology industry’s structure in 2020. While there were many who felt from the outset that such problems were beyond the capabilities of AI, many were confident about the possibility of complex inductive problem-solving.

The "intelligence" in conventional expert systems is pre-programmed from human expertise, while neural networks receive their "intelligence" via training. Expert systems can respond to finite sets of event stimuli (with finite sets of responses), while neural networks are expected to adapt to infinite sets of stimuli (with infinite sets of responses). It’s alleged that conventional expert systems can never learn, while neural networks "learn" via processing. Proponents of neural network research and development have identified the kinds of problems to which their technology is best suited: computationally intensive, non-deterministic, non-linear, abductive, intuitive, real-time, unstructured/imprecise and non-numeric.

Expert systems will routinize many decision-making processes. Rules about investment, management, resource allocation, and office administration will be embedded in expert systems. It’s unlikely that individuals will go onto the Web and execute trivial transactions. Smart support systems will automatically execute hundreds of pre-defined “authorized” transactions.

Intelligent systems differ from conventional ones in a number of important ways. First, conventional systems store and manipulate data within some very specific processing boundaries. AI systems store and apply knowledge to a variety of unspecified problems within selected problem domains. Conventional systems are passive, where AI systems actively interact with - and adapt to - their users. Conventional systems cannot infer beyond certain pre-programmed limits, but AI systems can make inferences, implement rules of thumb, and solve problems in much the same way we routinely decide whether or not to buy a Ford or a Chevy, or accept a new professional challenge.

Intelligent systems will simultaneously serve as the ultimate force multipliers and smart surrogates. They will extend the power of existing unintelligent computer and mechanical systems, augment and eventually even replace human problem-solving, and assume larger and larger operational responsibility. Unlike human expertise, which cannot be easily duplicated, expert systems can be copied as often as the need arises. It will thus be possible to distribute expertise where human experts might never venture.

Intelligent systems are morphing into “intelligent agents.” To most observers, a software agent is intelligent if has most of the following traits:

• Autonomous: self-initiated
• Social: able to communicate with users (and other agents)
• Reactive: able to answer questions and initiate action
• Dynamic: are time and space sensitive
• Asynchronous: action independent of linear/linked events
• Event-Driven: can pro-act and react to events
• “Inactive” user interaction: users can ignore
• Self-Executing: can run themselves
• Self-Contained: have what they need to run themselves

Intelligent systems can deal with a variety of tasks: