Expert Systems vs Machine Learning

Expert Systems

Expert Systems determines the binary outcome of a new situation of conditions by a set of rules created by learning certain situations of conditions from knowledge or past experiences.

Expert Systems is a complex computer system that uses a set of rules given. These rules are programmed based on past experiences of conditions or from previously given cases to determine a specific outcome from a new input of those conditions. For example, one of the uses of these emerging fields is in a biomedical situation. For example, we may take the case of COVID-19. Some of the symptoms of coronavirus are sore throat, headache, fever, skin rash, and difficulty in breathing. In a specific situation of these symptoms, the patient may or may not be diagnosed with coronavirus. Using past experiences with patients, the doctor will be able to tell whether or not the patient is diagnosed with the disease.

Similarly, just as the doctors use past experiences, one could feed this information or rules of these symptoms and binary outcomes to the computer system. Using these rules, the system would be able to recognize the binary outcome from the previous scenarios given. In these scenarios, a set of rules is given to the computer program. However, there are many drawbacks or problems when it comes to writing these rules. The rules may be very complex, inexpressible, unknown, or there may be too many factors to consider.

Problem 1 — Inexpressible Rules

For example, if there were some people stalling outside your house, you would probably be able to find out whether they’re involved in some suspicious activities by their body language. However, a security camera wouldn’t be able to sense this kind of “body language” but might assume that the person is just taking a walk by your street. This kind of data cannot be fed to the system in terms of rules. Additionally, a person’s “sixth sense” wouldn’t be something that can be shared with the system to determine.

Problem 2 — Lots of Factors to Consider

When given a series of factors, there may be thousands to consider. For example, a group of doctors may have come up with a vaccine for a disease. They would like to give the vaccine for a specific amount of days. Additionally, there are several other factors that may affect the number of days the vaccine takes to bring the disease to negative. Several of these factors may include blood pressure, weight, height, age, etc. Using a set of rules with all these parameters will take an extremely long time as there are so many factors to consider.

Problem 3 – Complex Rules

Even if one manages to write all the rules with several combinations, the rules may be extremely complex, and therefore hard to follow.

Problem 4–Unknown Rules

In some cases, the rules may be simply unknown. For example, for new diseases, the symptoms may not be that easy to consider and aren’t known, in which case, writing the rules is an impossible task.

Note:

Expert systems use binary outcomes, Machine Learning (or ML for short)is a similar computer system. However, in addition to binary classification, it can also return the probability of the outcome occurring. Machine Learning is another such computer system that uses a specific function, f(x₁, x₂, . . ., xₙ), with inputs of x₁, x₂, . . ., xₙ for all the relevant parameters and relates them with the outcome, or the mentioned probability.

Machine Learning

Machine Learning uses a function to relate each of the parameters and the output, by plotting several points and trying to find this function. Machine Learning mainly revolves around the computer system “learning” how to identify this function without being explicitly programmed.

There are 6 extremely crucial parts to achieve success in Machine Learning. (more coming soon in the next article). Thanks for reading!