How relevant is statistical machine learning in the age of Gen-AI? Should you learn statistical machine learning or directly jump to learning Gen-AI, Lang chain and so on? This question is similar to how relevant are motorbikes in the age of cars? Well, obviously there are situations where motorbikes are better compared to cars. In this blog, I am going to mention few points or few scenarios where using statistical machine learning is better compared to Gen-AI.
Point number one is, motorbikes provides a direct connection between the rider and the road. You have better control. Let’s say if something goes wrong or if you want to kind of interpret the behavior of it, then motorbikes are easier compared to cars. Similarly, statistical machine learning models such as linear regression, logistic regressions, they are easier to interpret. All you have is coefficient and when you give the input, you can even write mathematical equation and interpret the results. Compared to that, Gen-AI models are not that easy to interpret. They are like black box and your AI explainability is lower.
In the fields like finance and healthcare, where there are regulations and the requirements are such that you need high interpretability. In these situations, statistical machine learning models are better. I have a friend who works in a finance company. They are building credit risk model and in that case, they sometimes prefer logistic regression over some complex model, let’s say some neural network model. Even though it is giving let’s say three or four percent more accuracy, they will go for logistic regression because it has high interpretability. It has high explainability.
The second point is motorcycles consume less fuel compared to cars. Similarly, statistical machine learning models, they consume less computer resources when it comes to training as well as inference. For LLMs, you need GPUs, your compute cost, electricity cost will be higher for training as well as inference.
The next point is motorcycles are agile in congested traffic. Let’s say you are in a Dhaka traffic and if you’re going through motorbike, you can maneuver easily compared to car. If you have a big car, then moving that car is harder.
Statistical machine learning models are simple. Even if you have a smaller data set, they will work okay. Avoiding the need of massive training data and complex architecture.
The next point is motorcycles are best when you are navigating narrow roads. Let’s say if you’re navigating narrow roads, let’s say if you’re going to a village and the road is really smaller, people prefer bikes compared to cars. So similarly, statistical machine learning models are better for narrow tasks where you have a small data set where precision and structure outcome is the key.
The next point is motorcycles are easier to maintain compared to cars. When you change engine oil in a motorcycle, it will be easier compared to cars, both in terms of convenience as well as the cost. Let’s say if you have a Mercedes and something is broken, or let’s say even if you want to do a regular service, the cost will be higher in cars compared to motorcycles.
Similarly, statistical machine learning models are easier to maintain. Let’s say you deploy it in production, there is some data drift, you want to train it on a new data set, you can quickly train it. Compared to that, genuine models, when you want to fine-tune train it, they are a little bit harder and costlier.
The next point is motorcycles are cost effective for some type of trips. Let’s say you are going in that village trip, then the overall cost for motorcycle will be less compared to cars. So there are some tasks in the world of AI for which if you use statistical machine learning or Gen AI, both will work.
But when you use statistical machine learning, you will end up spending less money, you will get quicker results. It is like you want to cut something and for that you are using sword instead of knife. Knife here is statistical machine learning and sword is Gen AI.
Alright, so when you are working on your next machine learning or AI project, please keep these points in mind. If you have any questions, there is a comment box below.