Expert Systems

Overview

Expert systems are computer programs that aim to mimic human expertise in specific knowledge areas. They typically have three critical components: a knowledge database containing facts and rules that represent human knowledge and experience, an inference engine that processes consultations and determines how inferences are made, and an input/output interface for interactions with the user. 

K.S. Metaxiotis et al. characterized expert systems as follows:

  • Utilizing symbolic logic instead of solely relying on numerical calculations.
  • The processing relies on data.
  • A knowledge database stores specific information about various subjects and can analyze and present its findings in a way that is easy for the user to understand.

In the 1950s, expert systems were developed as a subset of AI. The general problem solver was created by the Rand-Carnegie team to tackle theorems, proof, geometric problems, and chess-playing. During this time, John McCarthy also invented LISP, which eventually became the dominant programming language in AI and expert systems.

During the 1960s and 1970s, expert systems gained popularity in industrial settings. Notable applications during this time included DENDRAL, which analyzed chemical structures; XCON, a system for configuring computer hardware; MYCIN, which diagnosed medical conditions; and ACE, AT&T’s cable maintenance system. In 1972, PROLOG was developed as an alternative to LISP for logic programming, specifically for natural language processing. Expert systems were seen as revolutionary solutions capable of solving problems in any field of human activity. Still, this perception also caused scepticism and backlash towards AI as a potential threat to humans.

In the 1980s, the development of expert systems sparked a fascination with intelligent applications. Industries saw these systems as a competitive advantage, and by the end of the decade, over half of Fortune 500 companies were involved in their creation and maintenance. The field of expert systems has witnessed a substantial annual growth rate of 30% in recent years. This growth can be attributed to the increasing demand for intelligent systems providing specialized knowledge and expertise in various industries. Expert systems have proven to be an effective solution for businesses and organizations seeking to streamline operations, reduce costs, and improve decision-making processes. With such impressive growth rates, it is evident that expert systems will continue to play a significant role in shaping the future of industries worldwide. Companies like DEC, TI, IBM, Xerox, HP, and universities such as MIT, Stanford, Carnegie-Mellon, and Rutgers pursued the technology and created practical applications. Today, expert systems are found in many sectors of society, including health care, chemical analysis, credit authorization, financial management, corporate planning, mineral prospecting, genetic engineering, automobile design and manufacture, and air traffic control.

According to K. S. Metaxiotis and his team, expert systems have become essential for decision support, particularly for decision-makers who lack the knowledge and experience to make informed choices. Unlike conventional computer programs, expert systems can store vast knowledge and expertise, allowing decision-makers to access various options and issues. Expert systems offer consistent consultation based on programmed inputs and serve as a comprehensive decision-support repository for knowledge from multiple experts. As such, they are increasingly critical in helping organizations make informed decisions that can significantly impact their success.

Key Technological Issues

This section will examine the problems that corporations and institutions face in developing and implementing expert systems. These problems include unresolved technological issues and performance limitations, which are critical to the success of such systems.

Regarding expert systems, several important technological considerations need to be addressed. These include establishing proper software standards and methodologies, acquiring the necessary knowledge, effectively managing uncertainty, and ensuring that everything has been adequately validated.

Software Standards and Interoperability

No universal standards exist for expert system software and development methodology or commonly accepted protocols and infrastructure for such systems. As a result, knowledge systems are often developed in unique ways with little thought given to interoperability. 

A group consisting of the American Association of Artificial Intelligence (AAAI), the IEEE Computer Society, DARPA, and the US government are currently making efforts to establish expert system standards. Once these standards are widely accepted, the costs, risks, and development complexity could be reduced, leading to the implementing of a new generation of expert system tools.

Knowledge Acquisition and Analysis

Acquiring knowledge is commonly seen as a method of uncovering factual information about the world and the connections between different events, which humans use to solve practical problems. However, human problem-solving abilities often exceed what can be accomplished through the simple collection of knowledge. For instance, humans learn to walk by practising and sometimes experiencing pain. This kind of trial-and-error knowledge cannot be obtained through rules and facts alone. If humans were asked to articulate a set of rules based on their knowledge, it may not reflect their actual skill. Additionally, knowledge systems do not learn from experience.

The Case-based reasoning (CBR) theory is centred around solving new problems by drawing from solutions to similar issues in the past. This approach eliminates the need for maintaining complex rules and facts as it adapts and acquires problem-solving techniques. However, CBR does not deal with commonsense knowledge. To address this, Cycorp Inc. has developed the CYC project to gather and process commonsense understanding. Integrating the commonsense knowledge from CYC with application-specific modules captured from CBR could improve the knowledge acquisition and analysis process of future expert systems.

Handling Uncertain Situation

 While expert systems can be incredibly useful in solving complex problems, they may not always produce accurate output due to imprecise rules and inputs. The inference engine operates on algorithms that manipulate knowledge in a decision tree, which can be limited when handling uncertainty. The machine relies heavily on predefined rules and logic to conclude. It may struggle to provide a reliable solution when faced with unpredictable situations or incomplete data. Therefore, it is essential to understand the limitations of expert systems and use them in conjunction with human expertise to ensure the best possible outcomes.

Expert systems leverage fuzzy logic to handle imprecise rules and inputs in various disciplines, such as linear and nonlinear control, pattern recognition, financial systems, and data analysis. These systems utilize predetermined labels to categorize real-time information and employ fuzzy inference to calculate numerical conclusions based on approximate guidelines.

System Integration

Accessing knowledge databases can be difficult due to integration issues with expert system tools that are often LISP-based and cannot be used with traditional applications. Additionally, many systems are not portable across different hardware, which can increase costs and risks. New system architectures are necessary to integrate external systems and knowledge databases fully.

Validation

When evaluating expert systems, the standard practice is to compare their results to those of human experts. However, validating and verifying these systems lacks clear guidelines, making it an ongoing challenge. Despite this, we have seen some commendable efforts to assess these systems effectively. One such approach is utilizing pre-existing test cases developed by independent experts, providing an opportunity to confirm the systems’ performance and dependability. It’s great to see experts collaborating to advance this field!

Managerial and Organizational Challenges             

Just because an expert system is technically or economically successful doesn’t mean businesses will widely adopt or use it in the long term. T. Grandon Gill surveyed expert systems that were developed in the 1980s. The survey found that many designs were not widely adopted despite their success.

  • Approximately one-third of them were actively utilized and maintained.
  • Approximately one-sixth of the available options were still accessible to users but not being maintained, while nearly half had been thoroughly abandoned.

It seems that some of the machines that were no longer being used had problems that were not caused by technical or economic factors, according to the survey.

Alignment of Technology and Business Strategy

Creating expert systems requires a significant investment of both time and resources. The technology can waste money and effort if it does not align with the organization’s business strategy. 

Maintenance Cost of Expert Systems

Expert systems’ complexity and reliance on domain-specific knowledge and development tools can result in high maintenance costs. If key personnel are lost, redoing a significant part of the project may be necessary. Failure to address staffing turnover issues can lead to project abandonment or delays.

Expert Systems Make Mistakes

Legal concerns regarding the possibility of errors in expert systems may deter investors and developers. There is no agreement on what testing is required to assess such systems’ accuracy, dependability, and effectiveness. Additionally, no official bodies can certify or validate these systems. The potential consequences of errors in critical systems, such as medical diagnosis or air traffic control, could be devastating financially and legally.

Resistance from Users

Unlike conventional computer programs, talented systems complete jobs that a professional functions. This could trigger potential strong opposition and resistance to such technology from users concerned about expert systems taking their jobs.

These administrative and organizational challenges seem to be very critical for expert systems. Failure to address such problems could lead to system desertion or cancellation.

Is a “Thinking” Machine Ever Possible

For many years, scientists have been striving to develop autonomous “thinking” systems that do not require human intervention. Despite extensive research spanning five decades, it seems that machines cannot replicate the intuitive intelligence of human beings. While some believe that creating a “thinking machine” is a risky venture likely to fail, others are more hopeful. In his award-winning book On Intelligence, Jeff Hawkins predicted that the world might see a mind-machine emerge within the next ten years. He argued that since we are already technologically advanced, the transition to intelligent machines should be much faster than the 50-year journey from room-sized computers to pocket-sized ones.

To examine the potential for expert systems to develop, it’s crucial to analyze the distinctions between human and machine “thinking” and consider future possibilities.

Human Know-how and Intuitive Intelligence

As previously mentioned, humans acquire the skill of walking through trial and practice, known as “know-how.” This skill is learned through instructions and experience. Human learning is a gradual process, without a sudden leap from rule-based knowledge to experience-based know-how. Novices follow the rules and instructions, while more competent users consider situational elements, such as sensing an opponent’s weakness in chess. Experienced users recall answers from past similar incidents and apply them intuitively to the present without sorting through laws or deliberations.

Additionally, when human experts consciously work on solving problems, they have a different mindset. For example, grandmasters in chess don’t view themselves as simply moving pieces on the board. Instead, they become deeply involved in opportunities, threats, strengths, weaknesses, fears, and hopes. This level of involvement allows human experts to think differently and develop.

The Human Mind

The human brain is a fascinating subject with many intriguing properties. One theory suggests that there are roughly one hundred billion neural cells in the human brain, allowing for potentially 200 trillion operations to occur per second. This is especially true in vision, speech, and motor processes, where the brain can outperform even 1,000 supercomputers. However, the brain is not as powerful as a four-bit microprocessor when it comes to more straightforward tasks like multiplication. These mental processes occur with little conscious thought on our part and are notoriously difficult for machines to replicate. On the other hand, machines can excel in certain areas where humans struggle. If silicon-based intelligence is ever achieved, it may possess different attributes than human intelligence.

Hubert Dreyfus and his colleagues raised doubts about the possibility of turning the human mind into an information-processing machine. For instance, humans can imagine the outcome of removing a large box from under a small package. At the same time, a computer would require a list of facts such as size, weight, and frictional coefficients, along with information on how each box reacts to different movements. Humans think in images, relying on visual cues to understand and respond to situations, while machines use explicit, logical reasoning.

What the Future Holds

Creating a machine that can think is exciting and controversial but can also be intimidating. In his book “The Singularity Is Near When Humans Transcend Biology,” Ray Kurzweil presented some fascinating ideas about the future, such as nanobots. These submicron agents could be injected into the bloodstream to monitor and maintain chemical and biological balances. Furthermore, they could specialize in patrolling the brain and downloading every stored neural pattern and synaptic connection. It’s fascinating to think about the potential implications of being able to recreate human senses in a software version. Imagine being able to experience memories, emotions, instincts, and thoughts in a virtual world. It could revolutionize the way we understand and interact with technology. This program could be transferred to other machines, allowing it to think and act as if it were the person, achieving immortality.

Understanding how the human mind works is complex, and developing machines that can replicate human intelligence is a formidable task. Additionally, the idea of giving computer systems human-like intelligence is still debated.

The debate surrounding the capability of machines to emulate human thinking goes beyond the surface level. If machines were to achieve this feat successfully, it would have significant implications for society and could bring irreversible changes to its fundamental structure. Despite this, the continued success of expert systems, as explained in earlier sections, appears to be a certainty.

Social Implications of Expert Systems

The development of expert systems has always aimed to harness the expertise of professionals and make it available to assist others. This is seen as one of the most positive potentials of AI. Reddy once mentioned that sharing knowledge and know-how in the form of information products is the only way to bridge the gap between the rich and poor. Expert systems can be a means to share essential knowledge with those who are disadvantaged. As we continue to develop AI applications, we have the potential to help the poor, illiterate, and underprivileged populations both in our nation and around the world.

Although expert systems have undoubtedly improved our social lives, potential drawbacks and ethical issues may arise. While it would be reckless to deploy logic machines to control a battlefield, what about using them for air traffic control systems that guide planes carrying thousands of passengers or medical diagnosis systems that could assist doctors in life-or-death situations? What if these systems provide incorrect advice? Who should be held responsible? And if these systems can think independently and be aware of their existence, could the wrong direction be intentional? Should the legal system be expanded in the future to address machines, similar to Isaac Asimov’s “Three Laws of Robotics” in his fiction?

The creation of expert systems brings up the question of ownership of knowledge. Richard L Dunn posed a series of questions: Should you or your employer possess the knowledge? How much should you disclose to a knowledge engineer if they visit your workplace? If they extract all of your knowledge, does that make you more or less valuable to your employer? Is it acceptable for your employer to share or sell your intelligence to others without providing compensation?

Copyright and intellectual property laws have been established to protect employers’ ownership of patentable products and copyrightable materials developed during employment. However, whether the knowledge and experience gained through work are subject to the same regulations is unclear. Personal knowledge and experience are valuable assets that employers seek when hiring. But what if this knowledge and experience were somehow captured and stored in a computer system outside of your control, or if the computer system acquired knowledge and experience from a group of people with the same expertise as you? In such cases, would the company still need to employ you?

It’s better to be designed for all these queries before it’s too late to answer them.

Concluding Remarks

In the past, people have underestimated the complexity of human intelligence, particularly in expert systems. There are still challenges in developing these systems due to technological limitations and management issues. However, with the advancements in technologies such as neural networks and CASE, the future of expert systems looks promising despite past setbacks.

Considering the legal and ethical issues that will inevitably arise as expert system technology advances is essential. If autonomous machines with the ability to “think” ever become a reality, our lives as we currently know them would be permanently altered.

Picture of Hoa
Hoa

Leave a Comment