An expert system in AI is a type of computer or application that can provide answers, solve complex problems, and make decisions just like a human expert in a specialized domain. Users rely on expert systems to perform complex calculations, analyses, and predictions. Expert systems are considered the first successful models of AI.
When were Expert Systems Invented?
Expert systems were first introduced by Stanford University researchers in the 1970s, although it has been on computer scientists’ minds since the early 1940s and 1950s. Edward Feigenbaum and Joshua Lederberg, who were key members of the Stanford Heuristic Programming Project, developed the first expert system in 1965. The researchers wanted to create a specialized system as opposed to a general-purpose one.
One of the device’s early applications included chemical analysis (DENDRAL) and medical diagnostics (MYCIN). Mycin, an infectious disease diagnostics tool, makes deductions through backward chaining.
Expert systems possess explanation facilities that let users ask them how they reached a particular conclusion or why they weren’t able to. That said, it is capable of justifying its reasoning and output.
What are the Parts of an Expert System and How do They Work?
An expert system in AI has three main components.
User interface (UI)
The user interface (UI) is the space that facilitates communication between the system and its users. It’s synonymous to your computer desktop or smartphone home screen. Users enter their queries through the UI, which then go through the inference engine.
The central component is the inference engine, which looks for facts from the knowledge base. It applies rules for interpreting them to arrive at new facts.
How an inference engine works is similar to how you make decisions in your everyday life. Before you commit to a choice, you’d probably think it through by looking back at all accumulated experiences and knowledge from past years. The inference engine works the same way: It follows a one-way line of reasoning (if-then rules) to get to the bottom of an issue (i.e., establish a new fact).
Finally, there’s the knowledge base. It is a database where information contributed by experts from a specialized domain is stored. Think of a knowledge base as a book or an article. To make passages from a book credible, you have to cite information from experts to back them up.
Do Expert Systems Make Mistakes?
They do. Expert systems do not possess human capabilities and are, therefore, limited in terms of decision-making capacity. At times, they are not able to detect clerical errors, and thus, make wrong suggestions as a result. Human intervention is still needed to ensure that expert systems operate correctly.
How have Expert Systems in AI Changed through the Years?
Expert systems nowadays are a far cry from their prototypes, which are bulky, run a little slower, and can only handle one command or output at a time. For instance, the Symbolics Lisp Machine, whose photograph can be seen on the Wikipedia page for expert systems, is huge and uses an outdated interface. It also needs a real human expert for knowledge acquisition.
On the other hand, current iterations of expert systems possess machine learning (ML) capabilities and can self-learn just like a human being. As such, an expert system can learn from datasets entered into its knowledge base and improve from experience. It also means that expert systems have far more sophisticated debugging abilities.
What is a Recent Example of an Expert System in AI?
The ROSS platform, designed based on IBM’s Watson cognitive computer, is an example of a modern expert system. Dubbed as “the world’s first artificially intelligent attorney,” ROSS is a legal research AI that helps law firms speed up research for court cases.
ROSS employs natural language processing (NLP), which means you can ask it a question, and it can interpret that question to find answers, and accomplish other tasks through text, content, and sentiment analysis. As such, ROSS can transform unstructured data into structured data that humans can easily understand.