How Expert Systems Are Implemented: Part Six

Dr. Ai: Although expert systems have been applied in various degrees, they still have some limitations that affect their development and usage.

Firstly, the bottleneck problem of knowledge acquisition has not been well solved; it mainly relies on manually summarizing expert experiences to acquire knowledge. However, since experts are very rare, acquiring expert knowledge is quite challenging. Furthermore, even if experts are willing to help in knowledge acquisition, the diversity of actual situations makes it difficult for them to summarize effective knowledge, even though they can perform their work and solve problems effectively. For example, many students can ride a bicycle; if someone cannot ride a bike and falls off immediately, they may wonder how you can ride so skillfully. You might find it hard to articulate the knowledge you use to ride well, even though you can do it effortlessly. This exemplifies the bottleneck problem encountered in constructing expert systems, which is one of the main obstacles to the utilization of expert systems.

Secondly, knowledge bases are always limited; they cannot contain all information. Human intelligence is reflected in the ability to learn patterns and features from limited knowledge. Rules are static, but humans are dynamic and can flexibly apply knowledge to solve new problems. Knowledge-driven expert system models can only reason using the existing knowledge base and cannot learn new knowledge. Within the scope covered by the knowledge base, expert systems may solve problems well, but even a slight deviation can lead to a drastic decline in performance or even an inability to solve the problem, reflecting the system’s fragility.

Moreover, knowledge-driven expert systems can only describe specific domains and lack generality, making it difficult to handle common-sense problems. However, knowledge is dynamically changing, especially in today’s era of big data, where the efficiency of manually or semi-automatically establishing rule systems is too low to adapt to changes and updates in knowledge.

The most important aspect of expert systems is knowledge. Many types of knowledge are very difficult to organize and update, resulting in the so-called knowledge acquisition bottleneck problem, which poses significant challenges for the promotion and application of expert systems.