Difference Between AI, Machine Learning and Deep Learning


What is AI, machine learning and deep learning

Differences between the three

Real world examples

Applications and classifications of the three


There is a lot of buzz about artificial intelligence in the world of late and the word seems to appear a lot in the media. What is the buzz all about? You may not realise it but you are interacting with AI on a more regular basis than you might think. So AI is not just a blood-thirsty Arnold Schwarzeneggar that can never die!

What Is Artificial Intelligence?

The word artificial intelligence was first coined by John Mccarthy in the year 1965 at a Dartmouth conference. He gave his definition of AI as follows, “Every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. An attempt will be made to find how to make machines use language, from abstractions and concepts, solve kinds of problems now reserved for humans and improve themselves”

Now what this basically means is we can train computers to perform tasks or solve problems that are supposed to be done by humans, and the machine might not only be able to do it, but also learn from it and get better at it. Maybe even better than a human could! This is also what machine learning revolves around. But we will get to that later.

A great example of how far AI has come today could be how Facebook’s two artificial intelligent programs started interacting with each other. They were told to carry out negotiations about trade, attempt to swap hats, balls and books in a language incomprehensible to humans, and pretty soon, the bots were talking to each other, chanting in a language nobody could make sense of, they were also asked to improve their bartering, which they did! A snippet of how their negotiations went were as follows

Bob: i can i i everything else . . . . . . . . . . . . . .

Alice: balls have zero to me to me to me to me to me to me to me to me to

Bob: you i everything else . . . . . . . . . . . . . .

Alice: balls have a ball to me to me to me to me to me to me to me

Bob: i i can i i i everything else . . . . . . . . . . . . . .

Alice: balls have a ball to me to me to me to me to me to me to me

Bob: i . . . . . . . . . . . . . . . . . . .

Alice: balls have zero to me to me to me to me to me to me to me to me to

Bob: you i i i i i everything else . . . . . . . . . . . . . .

Alice: balls have 0 to me to me to me to me to me to me to me to me to

Bob: you i i i everything else . . . . . . . . . . . . . .

Alice: balls have zero to me to me to me to me to me to me to me to me to

Yep, let that sink in . . . . . . . . . . . . .

Machine learning and deep learning are both concepts that come under AI. For an easier understanding, consider a venn diagram with 3 concentric circles. Let AI be the outermost circle and machine learning be the circle inside and deep learning the innermost circle. Hence, you can say machine learning and deep learning are subsets of AI.

Some real world applications or examples of AI could be:

– Machine learning

– Computer vision

– Natural long processing

– Robotics

– Pattern algorithm

– Knowledge management

There are two types of AI, strong AI and weak AI.

Strong AI works like the human brain, connected by neural networks, like how the human brain is connected via neurons and this forms the basis of this artificial brain. This system thinks like a human and gives an insight on the how the brain works

Whereas, weak AI also behaves like a human but does not give insight on how the brain works.

R2D2 and C3PO could be classic examples of strong AI, but they exist only in the realm of science fiction: we are yet to get there.

An example of weak AI could be IBMs speech recognition bot known as Watson. Watson can learn and improve from various sources like how a human brain would.

What is Machine Learning?

Coming to machine learning, as I mentioned earlier, machine learning is a subset of AI. It involves providing to a system training data with which it can learn to automatically improve without being explicitly programmed. This branch focuses on development of computer programs that can access data, learn it, and use it for themselves. There are three different categories in ML

  1.       Supervised machine learning
  2.       Unsupervised machine learning
  3.       Reinforcement machine learning

Let’s talk about each of them in a little more detail

Supervised Machine Learning:

In this type of ML, the input is a series of labelled or structured data, which is also known as training data. In this type, the system is given both the input and told what the output must look like.

For example, if the training data is 2+3=5, then 2 and 3 are the inputs and 5 is the target. Hence the machine knows the output must be the addition of the two numbers.

Two important concepts here are:

  1.         Classification
  2.         Regression


It predicts a continuous response value. For example,

      a. Predict a number which can vary from +infinity to –infinity.

      b. Price of a house in a city?

      c. Value of stock?

      d. How many total runs can be onboard?


It is a type of problem where we predict categorical response in which data can be separated into specific classes. For example,

  •         Is this mail a spam or not?
  •         Will it rain today or not?

Unsupervised Machine Learning:

In this type, the training data does not include targets. So we don’t tell the system what to do or where to go. The system has to understand this itself and come up with desired results. Only the results will be evaluated and the system will be “rewarded” with data to learn and correct itself. For example, random articles from various pages.

This data is unlabelled and unstructured.

Unstructured data    —> algorithm          —> conclusions/results

                                   (understands patterns in

                                   the data itself)

An important type in this is clustering, in which we don’t provide labels, the system understands this by itself. For example; given news articles, cluster them into different genres of news.

Reinforcement Learning:

In this type of machine learning, the system is allowed to interact with the environment using trial and error, learn from it, improve itself, and come up with the intended results

A popular example of this would be Atari games. Till now, Atari games were dominated by some of the smartest people around the world knows as ‘Atari masters’. Until, google’s Deepmind algorithm AlphaGo beat the best players around the world. How? By using millions or cycles of trial and error and back tracking the strategies that work and working out the best strategy to beat the game. The basis on which it was taught was,

Goal – beat the game (highest score)

State – raw game pixels

Action – up, down, left, right

Reward – game score  

(source: Reddit)

Deep Learning:

Deep learning is a subset of machine learning. It is inspired by the functioning and structure of a human brain, and on how neural networks work. In this, a machine is basically taught to do what comes naturally to humans.

It is a key technology that is used in various AI applications like self driving cars, differentiating between a stop sign and a lamp post, voice control devices etc.

There are two main reasons as to why deep learning is gaining a lot of traction these days and could not be implemented before;

  1. It requires a large amount of training data. For example, millions of pictures and thousands of hours of videos
  2. It requires a substantial amount of computing power. With the advent of high performance GPU’s combined with cloud computing for unlimited storage, it helps developers train the system faster.

you can also learn more about the basics of ML from this link

Intro to machine learning- GL4L

How It Works?

Deep learning models often utilise neural network architectures, which is also why they are referred to as deep neural networking.

The word deep is used because of the number of hidden layers in the neural network. A traditional neural network might consist of 2-3 hidden layers but a deep neural network consists of about 100-150 hidden layers.

These models are trained by using large sets of labelled data and neural networks that learn directly from the training data without any external human intervention.

One of the most popular types of deep neural networks is known as convolutional neural networks (CNN or ConvNet). A CNN convolves learned features with input data, and uses 2D convolutional layers.

So Let’s Summarise:

It is clear that the four terms, artificial intelligence, machine learning, reinforcement learning and deep learning are all related. But they aren’t the same thing and have substantial differences.


Leave a Reply

1 Comment threads
0 Thread replies
Most reacted comment
Hottest comment thread
1 Comment authors
안전놀이터 Recent comment authors
newest oldest most voted
Notify of