Differentiating AI, ML and DL

 AI (Artificial Intelligence), ML (Machine Learning) and DL (Deep Learning) are the buzzwords right now. Almost everyone is talking about it right now. And every day there’s a new technique or tech-related AI is developed by people. In this blog, I’ll try to explain what the field is about and what the general difference is. First of all these three are not separate fields, there’s a hierarchy. On the top its AI then one subset of it is ML and one subset of ML is DL.

Artificial Intelligence

Let’s talk about AI-first. Artificial Intelligence is an idea to develop a fully functional Intelligence with the machine. Machines are capable of doing lots of complex calculations very quickly compared to humans. The theory of AI goes way back to 1950s. The idea is to develop a thinking machine, a brain build by copper wires. Now to develop intelligence, we should define intelligence. There are two main approaches to build AI. The first one is the Symbolic AI. The idea behind symbolic AI is that every problem can be boiled down to math and logical manipulation of equations. So be it a task to play chess, to create complex geometry, prove theorems, etc. after the following decades people started building AI. But there is a flaw if you look closely it is just like any other machine you feed instructions, it calculates and gives output. The difference is the size of calculation is increased. For example, to build the chess-playing robot we add all the rules of the game and through calculations, it gives the best strategy. And this AI is not flexible, for every problem we have to write the clear commands for example to win chess we have rules. But what If the problem is too general say to distinguish between cat and dogs, what set of instructions can you give to the machine to differentiate the animals. That’s where ML comes in.      

Machine Learning

Now to solve the cat vs. dog task, we need to understand how we do it. We differentiate animals because we had seen them many times and we understand the features of any animals. So Machine learning is about taking data, images of cats and dogs, create a program that understands the features of cats and dogs like the type of fur, body structure, the shape of the mouth, etc. In ML we use mathematical formulas such as y = mx + b and many more to find these features in Images. 

After creating a program we can give the image of cat and dog and it will correctly identify it as cat and dog. But there is still a problem, for a program to correctly distinguish between cat and dog, we manually have to give all the features like how the cat looks like, shape of its body, type of the fur, body colour, etc. and also thousands of example images. And now, DL comes to save the day.       

Deep Learning

To solve the above-said problem, we take inspiration from how our brain works. We have billions of neurons a type of cells that constantly communicate with each other and process the information. So in Deep learning, we define the small program as neuron which process information and we build hundreds of these create a whole network of interconnected neurons and thus the name neural network. The main advantage is that it extracts the features of data on its own and computes the desired output. So for our cat and dog problem we provide hundreds of images of cat and dog, it extracts features from those images on its own and predicts whether the new image is of cat or dog. 
The above explanation of ML and DL is little bit AI-centric, very theoretical. But in the real world, the ML and DL are the tools used to extract patterns and gain insights from huge and complex data and to provide solutions to problems.

Hope you have now a clear idea about the field. If there are any questions, ask in comments.