A Quick Guide to Expert Systems

Defining expert systems?

Seth Adler

Sponsored Content

The term ‘Artificial Intelligence’ or ‘AI’ is used, when algorithms for a given program are made to simulate the behavior, intuition or judgement of a human being. When this judgement and behavior is simulated from a human or an organization that has expertise in a particular domain, a new branch of AI is introduced, and it’s called ‘Expert Systems’. This special branch of AI uses data stored in a knowledge base to find expert solutions to problems that are usually addressed by a human specialist, e.g. a medicine doctor. This knowledge base is developed with the help of human experts (from particular domains) who add the necessary data into it.

Expert systems is a great example of the intersection between computer science and business. In the last few years, there has been a sharp rise of interest in AI and its applicability by corporations in the business world, and expert systems is what seems to grab their attention the most. This is mainly because corporations have the incentive to use expert systems as a means to increase efficiency, reduce human errors and thereby generating more profit.


Listen to Manik Patil discuss the theory behind Technology Business Management

Source: The AI Network Podcast


In order for knowledge-based expert systems to work properly for a certain organization, the first and most crucial step is to select an appropriate domain. The specialists in an organization and the knowledge engineering project team must collaborate before the start of a project (see Figure 1 below).



Figure 1 Knowledge Acquisition Flowchart


This collaboration is necessary in order to identify the different type of tasks in the organization that require human expertise, and the feasibility of setting up a knowledge-based expert system with its current limitations. According to this scientific research paper called “Selection of an Appropriate Domain for an Expert System” by David S. Prerau, in some cases there is a very specific application, chosen by management, for which an expert system is to be developed. In this situation, it is likely that, those who selected the application area had little technical knowledge of artificial intelligence or expert systems. Thus, the project team must decide whether the selected application is one that is best suited to solution by present expert system technology, or if there might be a better way (or, possibly, no way) to attack the problem.

The factors that affect the selection of a particular domain varies in each different organization and instance. However, the following list outlines some common requirements that need to be fulfilled to develop any expert system.       

  • The domain should contain characteristics of expert knowledge, judgment & experience that can be artificially ‘mimicked’ & packaged in a program.
  • The functions performed by the domain cannot be ‘automated’ or ‘computerized’ by conventional programming approaches. The addition of expert systems introduces complexities and requires taking technical risks, which is why it should be avoided if the problem can be solved simply by using standard and regular (non-AI) computer programs.
  • There should be credible experts that already deal with solving the problems. An artificial expert system should not be developed in the absence of real human experts.
  • The completed system should be expected to have a significant payoff for the corporation.
  • Among possible application domains, the domain selected should be one that best meets overall project goals regarding project payoff versus risk of failure. For example, a conservative approach would be to attempt to develop a system that would meet some criterion for minimum payoff if successful and that seems to offer the best chance of success.


Difficulties of Expert Systems

As it can be seen, the process of building expert systems is not a very simple one and involves many factors. One of the biggest challenges of making expert systems work is the involvement of individuals (human experts), who have very little understanding of AI, and have difficulties to translate their specialized knowledge to AI developers. On the other hand, AI developers could have problems to truly understand the thought process of a human expert in a certain domain, e.g. engineering. For an organization to implement expert systems, it would have to convince a real human expert to cooperate with the AI developing team. This could sometimes prove to be a tricky situation. The article titled, ‘Why Expert Systems Fail’ published by The Journal of the Operational Research Society states the following problems:

  • The human expert may not be available. Gaining knowledge for an expert can take a considerable amount of time and diverting him from his regular work may be too costly for the organization.
  • The human expert in not willing to communicate his ideas. It is possible for the expert to see the AI development as his substitute and feel threatened.

The same article outlines the fact that expert systems lack the ‘common sense’ of a human expert. For example, an expert system in the medical industry may not have a certain disease in its knowledge-base, and it could potentially diagnose a patient who is suffering and clearly looks unhealthy, as healthy.   

According to an article titled, ‘Diagnostic Expert Systems: From Expert’s Knowledge to Real-Time Systems’, another problem with expert systems is its strict domain dependency. This is tricky in situations where the problem that the expert system is trying to solve may need some knowledge that lies outside of its domain.

Furthermore, in domains such as medicine and engineering, the solution of a single problem may require expertise from numerous branches. As stated in the research paper, ‘Expert Systems: Perils and Promise’, expert knowledge is sometimes distributed. Experience with systems that have survived the feasibility demonstration stage suggests that reliance on any single expert can either create blind spots in the knowledge base or result in a system that will not have users. Problem solving often takes place in a community where many different experts pool their expertise.

Even though expert systems have been around since the early 1980’s, it simply was not robust enough until recently. This is mainly because expert systems require a lot of data to work with, which thanks to the recent big data revolution by giant internet corporations such as Google, Amazon, Yahoo, Facebook etc. is now readily available. This solves the data issue and helps companies such as Google and Amazon to operate. However, much of the deeper concerns of expert systems mentioned in this section remain unsolved and are still being researched.   


Applications of Expert Systems

Next, let’s take a look at some of the existing applications of expert systems in the market. One of the earliest applications of expert system was adopted in the field of medicine. The expert system developed was named MYCIN and it was made to identify various bacteria that can cause severe infections and to recommend drugs based on the patient’s age, sex and weight.

MYCIN primarily uses backward chaining, or a goal-directed control strategy. The deductive validity of the argument is established in the same

way, but the system’s behavior is quite different. In goal-directed reasoning a system starts with a statement of the goal to achieve and works "back- ward" through inference rules, i.e., from right to left, to find the data that establish that goal, for example: (see Figure 2 below)



Figure 2 Simplified Logical 'Hierchy' of MYCIN


This is an example of how the algorithm (simplest form) in MYCIN represents knowledge mostly as conditional statements or rules. In the given example, the goal is to find out about ‘C’, that may be a certain disease. Rule 1 states that if ‘B’ is true then ‘C’ is also true. Here ‘B’ can be considered as a symptom. So MYCIN is looking for a certain symptom ‘B’ in a patient that can be attributed to her sickness and thereby diagnosing the disease as ‘C’. It also contains an implicit rule that calls out that ‘A’ causes ‘B’ i.e. if the patient was involved in a certain activity ‘A’ then it is possible to expect the symptom ‘B’. For example, let’s consider a hypothetical scenario; MYCIN is dealing with a patient who has the HIV virus that causes AIDS. It will look for a common symptom of AIDS like fatigue or loss of appetite. Next, it will justify its reasoning by asking the patient if he was engaged in unprotected sex or any other similar known causes of AIDS. If the answer is ‘yes’ then MYCIN will use its rule-based system to infer that the patient has AIDS. Keep in mind that this is a very simplified example for the functionalities of an expert system such as MYCIN. In reality there might be hundreds of variables to consider and numerous subsequent rules to be created in the form of algorithms, and finally enormous amount of data to analyze in order to identify symptoms and diseases.


Learn more about how AI is transforming the healthcare system with Dr. Gadir

Source: AIIA.net: AI and Intelligent Automation


Another recent development of an expert system was made in the talent recruitment/human resource industry. Wade&Wendy is an AI based expert system that is programmed to have a conversation with a client (employer), and use its heuristics driven by knowledge-based algorithms to figure out the exact information that it needs to recruit a suitable candidate for the organization’s hiring needs. Just how a human recruiter tries to have meaningful conversations with CEO’s & managers, in order to fully understand and adapt to their changing needs; the same way Wade&Wendy will keep on growing and learning as it engages in a greater number of conversations. This particular expert system is an example of a certain type of AI called a ‘Chatbot’; a computer program or an artificial intelligence which conducts a conversation via auditory or textual methods. Such programs are often designed to convincingly simulate how a human would behave as a conversational partner, thereby passing the Turing test. Chatbots are typically used in dialog systems for various practical purposes including customer service or information acquisition. This is where 2 branches of AI namely, ‘Natural Language Processing’ and ‘expert systems’ coincide.

We are heading towards a time when AI and humans will work together to solving problems. However, there are still many challenging issues that need to be solved in order to make expert systems acceptable by the masses of people. The branch of AI called expert systems is mainly responsible to help human experts and to learn from them. A lot of hard work and research is still required to find a sweet spot between the work of human experts in various fields and capabilities of artificial intelligence. The use of expert systems in the industry is not yet widespread but in the last few years with the advent of big data, there has been a considerable rise in the adoption of this technology, as more and more corporations try out the power of AI.



Smith R. (May 8, 1985). Knowledge-Based Systems – Concepts, Techniques & Examples

Prerau D (Number 2 1985). Selection of an Appropriate Domain for an Expert System
            AI Magazine Volume 6 © AAAI)

Panwar S. (October 3, 2018). Expert Systems and Applied Artificial Intelligence [AI]


Buchanan B. (NA) The MYCIN Experiments of the Stanford Heuristics Programming Project


Rouse M. (September 5, 2005). IM bot

Makelburg D. (August 1, 2017). Using Intuition & Data to Ask Better Interview Questions

Bell M. (July 1, 1985). Why Expert Systems Fail - Journal of the Operational Research Society


Angeli C. (NA). Diagnostic Expert Systems: From Expert’s Knowledge to Real-Time Systems


Bobrow D, Mittal S. (November 9, 1986). Expert Systems: Perils and Promise