There is a lot of excitement and enthusiasm to adopt new technologies such as Artificial Intelligence, Machine Learning, and Deep Learning today. But is the adoption based on actual needs or embracing state-of-the-art technology just for the sake of it?
What happens to technologies such as rule-based systems which were in vogue before Artificial Intelligence systems such as Machine Learning came into existence? Should all organizations shift to Machine Learning for better results, or can they continue to use rule-based systems?
In this article, let us discuss rule-based systems, machine learning, or self-learning systems and discuss the advantages, limitations, and the business needs to apply them.
What is a rule-based system?
When you program the system to make decisions based on a certain set of rules, they are called rule-based systems. Rule-based systems are built by human experts with in-depth domain knowledge to guarantee the best possible outputs. Hence, they are expert-driven systems.
Rule-based systems rely on a set of facts and apply “if-then” rules on the facts to make decisions. For example, when a bank receives an application for a loan from someone, the bank can use a simple rule like “If the applicant’s age is less <= 50 and his income is >= $60000 annually” the bank can approve the loan. Of course, this is a simple rule, and rule-based systems can be built much more complex rules to qualify the applicant and decide whether to approve or reject the request.
As mentioned above, these rules are built by humans who are domain experts, and the business knowledge they bring into building these rules is extremely critical for the system to accept inputs and provide outputs accordingly.
What is a machine-learning system?
Machine learning systems examine large amounts of past data and make decisions based on their learning from the data. For instance, in the loan application example above, a machine learning system can see that a loan applicant whose age is <=50 and income >=$60000 can be approved based on the vast number of applicant data from the past.
The important point to note here is no one needs to tell the information above to the Machine Learning based system as the software can make this logical deduction on its own by simply analyzing the data and looking for correlations.
In short, a machine learning system learns by itself from patterns within the data fed to it, while rule-based systems use rigid “If-else” rules that have to be hand-crafted.
Rule-based systems: Should you ignore them
Based on what we have seen so far, it looks like machine learning is the way to go because it doesn’t need any human interaction and learns itself from “past data” to make decisions.
But from where can Machine learning systems get this “past data,” which is critical to the model’s behavior? An existing rule-based system can provide the same. The data from the rule-based system can come in handy in increasing the accuracy of the machine learning algorithm. A 50% accuracy (which is the same as a coin toss) is something you can expect from a machine learning model that uses the rule engine’s data.
However, rule-based systems are prone to human error, and the integration of rules can be time-consuming and expensive. Complex and too many rules also contribute to performance degradation. As rules become stricter, there is a high possibility of losing out on good customers.
Having said that, rule-based systems can execute decisions much faster with proper training. They are reliable. Rule-based systems can be valuable when exact answers are required, and the number of rules and options is relatively simple. The output of a rule-based system is easy for a human to debug.
Is Machine Learning the way forward?
The expectation that comes with using machine learning is that they are artificial intelligence systems that provide high levels of accuracy compared to humans. This notion leads to “machines replacing humans,” “reducing human effort,” “time savings,” and so on. But building a machine learning system is no joke, and a wrongly built machine learning system can cost an organization dearly on all fronts, including cost, effort, and usage.
While machine learning systems provide significant advantages over the capabilities of a rule-based system, it would be a mistake to consider machine learning as the silver bullet to all of your problems.
Machine learning models take time to understand and learn from data. A machine learning model is only as good as the data it absorbs (Read our blog on AI and Data Quality), and it can take months for the system to be ready to replace rules. There is no point in expecting “super accuracy” from a machine learning model overnight or complain that it is underperforming when compared to your current rule-based engine.
A good approach when transitioning from rule-based systems to machine learning is run rules in parallel with machine learning. This helps to compare results over time and decide when to replace machine learning in the place of your rule-based system.
You can also consider operating rules and machine learning systems in tandem, which would be more beneficial to the organization than replacing rules entirely. Machine learning is not meant to substitute humans, but instead to augment what humans are capable of. Results from the rightly built machine learning model can imitate human capabilities, completely complementing human efforts and helping in increasing their productivity.
While machine learning’s strength is in the amount of data it can analyze and track in real-time, a human’s strength is in providing context and intuition in analyzing outliers and other edge-case scenarios.
It is important to consider the domain expertise that a human brings to the building rules that makes rule-based systems successful.