Am I right? Wrong? AI Approaches
While To be right or wrong, zero or one (boolean approach) couldn’t help us to find the answer to many questions, there are many other theories that can help us. One is the probability approach, another one is the Possibility approach that is also used in Fuzzy systems. Let’s see what these are and how we’re using them.
The boolean approach
The boolean approach is used in many systems and it could work perfectly in specific situations. In the old, the rule-based system was fitted as AI systems and an example of it, is the rule below
“If the house is on fire then call 911”
There were some problems in rule-based systems, such as it couldn’t learn new situations and be adopted automatically, So that was a reason that another approach is applied to AI systems.
The possibility approach
Using the possibility approach to give something like weight to each rule could help us to have better learning systems. Fuzzy Systems introduced by Dr. Zade was used the most in the late ’90s and early ‘2000s for AI applications. In this type of system, the rules are the same as before but another approach was given to produce rules. By producing rules we mean the ability to learn new situations.
To have an idea of how the rules are produced, it’s important to know what is fuzzy logic. In fuzzy logic, there is no 0 or 1, and a wide variety of values can be given in the system. Think of the boolean logic that for a person it can be said that he/she has a sickness or doesn’t. In fuzzy logic multiple values can represent he/she has a sickness, for example, a person is sick with a possibility of 0.5. Finding out how the fuzzy logic works can be a little tricky because it is using the possibility theory but for now just have in mind that any fact can be accepted, rejected, or more values in between(slightly accepted, slightly rejected, and more …).
So in this type of system, the rules will be generated via fuzzy logic, with the conversion of each input value as a fuzzy number (fuzzification), and in output, each value can be represented as a fuzzy value or a crisp value as in boolean logic (with defuzzification process).
The probability approach
While the possibility approach for rule bases in fuzzy systems gave a reasonable performance of learning, in more new AI methods statistical modeling is applied to systems with the base of probability. In this type of system probability as a backbone is creating an abstraction of how many times a state will be seen in the timeline of having different states. As mentioned above the new AI systems are statistical models, they are used to manipulate and find the patterns of data in our space. Whatever the distribution of data is, this method will try to find a new space to project them onto it and create a pattern for it.
A famous example of this type of AI system is neural networks which is a subtype of the soft-computing topic. In this type of model, artificial neurons are used to find the best feature representation and predict the newly came data’s output.
Conclusion
Until now We’ve talked about different kinds of approaches to creating an artificial intelligence system, from the old AI systems such as rule-based systems that mostly used the boolean approach and to the newer ones fuzzy systems using the possibility theory and the most famous ones such as neural networks that uses the probability approach.
The goal behind this article was to introduce the reader about different approaches used in AI systems and for each approach, there may be many overlaps (Also some old approaches can work perfectly in today's problems, and they are being used now).
If you want to go further and see how rule-based systems, fuzzy systems and other new statistical methods are working right now, I’ve left some links to some useful books at the end.
At last Thanks for you reading this article and as always if you have any questions feel free to ask.
Useful resources:
