AI vs ML vs DL: "Well These Are Not Really Different Things"

Following is a little article I wrote up back in 2018 for my college magazine. I was 16/17 and stupid, today kids are busy raising millions for startup but I was just figuring out coding, ego-fighting with friends and trying to get a girl out of pure FOMO and status back in college.
So yeah this blog has a lot of grammar mistakes, but I intend to keep it as is because it would be fun to read this later and hey in todays world where everything is heavily moderated using AI tools, its good to have some raw unedited content too. hehe.

How would you differentiate them AI, ML, DL?

I think you shouldn’t write those terms like this (AI, ML, DL) or (AI vs ML vs DL). It should be like AI<ML<DL. To make things simple ML is just a sub topic of AI and DL is just a sub topic of ML.

AI (Artificial Intelligence)

In simple words the intelligence acquired artificially is called Artificial Intelligence. Like for example the bots in the game PUBG or the older version Google Assistant and all the voice assistants made by us simple Computer Engineers. A Machine-made Intelligence. Yeah! In some cases, they are truly intelligent than us. But don’t forget human intelligence is more important. Well AI is emerging into many fields. Like in the Agricultural and Medical sectors.

Now believe it or not somewhere in Solapur, Maharashtra a Farmer hired four Computer Engineers to work in a field. In fact, the whole field was automated. That’s the magic of AI. Self-parking cars are already there for public. Self-driving cars are there but cannot fully replace the current driving style of crazy Indian masses. Now there are mobile processors coming with AI as one of their superb features. Gigabyte just released a new Aero 15 laptop what they call the world first AI enabled laptop. But after some tests the AI laptop proved to useless. Still needs a lot of work. Lots of industries are already investing a lot on this technology.

Let’s understand it in a different way, 1 + 2 = 3 == a + b = x teaches this equation to an AI program and then if you give a, b it will return you x.

ML (Machine Learning)

Machine Learning is a sub topic of AI. Here a machine learns or thinks from itself using different learning methods. Some of the learning methods are supervised learning, Semi-supervised learning, unsupervised learning, Reinforced learning. You’ll find semi-supervised learning in current Voice Assistant Services like Google Assistant, Amazon Alexa and Apple Siri. They learn a lot from the users. Some voice assistants have got a security feature that is it would only respond to your voice and no one else’s. Now that’s really intelligent. Many phones now got cameras having AI, while Google started the AI camera trend. Here the AI camera intelligently uses some patterns to see what is there in the frame, if it’s a human then make his face intelligently smooth and lighter.

Machine learning type of algorithms helped the scientist in many ways solving different problems. It works like this [ 1 2 = 3, 2 4 = 6] here the machine is feed with the numbers and the answer but without a pattern on how the answer came. What things can come between those numbers +, -, *, /. The machine will then try to put these different signs to get the required answer and try to do the same with all numbers and then create a pattern for itself. Well this is how it gets trained.

DL (Deep Learning)

Deep learning is something related to the way our brain works the relation method. Well if you don’t know our brains work in a relational pattern. They are way faster than a computer because we can find relations in very odd thinks. It’s sort of really difficult for computers to do the same.

Deep Learning involves in creating Neural Networks the same way our brain has. Many geeks out there have already started building game playing Neural Network. Some playing DOOM, PUBG, Mario Super Bros and just many others. If you love a game so much why don’t you teach your best friend Computer to play it, some or the other day it will just become the world’s best player. Because of computers computational power the computer will also be able to understand every possible move that would be taken by the opponent players during the game.

For example in a PUBG like game where the opponent player sometimes jumps and try’s to shoots others so that the other player won’t be able to follow up and shoot him, but when the computer knows that this guy is just going to start jumping like a monkey during a close encounter the computer will also jump and while jumping understand the time intervals between each jumps and also the movements the opponent player and give him an absolute headshot. That’s awesome right because saying is such easy than actually doing it. There is a lot of things to look after while playing a PUBG like sophisticated game where there is a need of total immersion for a success.

It’s not over yet people have done a lot of crazy stuff with the whole AI segment. Now Google worked on an epic project AlphaGo which started some years back. The team behind it was the Google’s Deep mind ANN professionals. This AlphaGo was an AI created to learn and play the famous Chinese board game GO. Over a period of time AlphaGo learned to play by competing between different professional GO players. The AlphaGo also learned different techniques used by different types of players during matches. Finally, AlphaGo became the world’s best player after defeating every GO professional in straight matches. This is quite an impressive thing.

The hardware requirements for running these neural networks at full potential are just tremendous and really expensive. You’ll will have to get thousands of powerful GPU’s all running together in a server like environment to make those AlphaGo like sophisticated ANN (Artificial Neural Network). Well again Google came to rescue by bringing a new thing called TPU (Tensor Processing Unit) a hardware unit totally dedicated to make neural networking easier. These TPU’s are programmed using different languages but also adding an extension/package called Tensor Flow for making neural networks easier. TPU’s are said to be smaller, lighter and faster that GPGPU’s (General Propose Graphic Processing Unit).

Technologies like Tensor Flow, Tensor Flow Lite (a lite edition for edge devices), TPU’s were invented by Google. Love you Google! Our simple Google search engine also learns and extracts patterns from our searches.

Do you know it would take millions of such TPU 3.0 to just create out brain! That’s how complicated our neural networks in the brain are just too brainy right!!!

[Alphons Jaimon - 2018]