What is the differentiation between Machine Learning, Artificial Intelligence, and Deep Learning?

 

You know what, this is the most popular question asked by any machine learning/data science discussion. Newbies (and some specialists as well) tend to complicate around the relation between this trio.

I will keep the boring formal definitions of AI/ML for last. Let us start with an exciting character named Deep Blue (presented by IBM) who became eminent in 1997 when he has beaten the Chess champion - Gary Kasparov. For each 3-min move, Deep Blue could analyze 50 billion positions & take a decision based on pre-programmed software rules. This was an example of AI without ML. Two decades passed by and the spotlight shifts to Seoul in 2016 where an even more interesting character named AlphaGo created by Google defeated the Go world champion - Lee Sedol. This was an example of AI with ML. No rule was pre-programmed into Alphago! Not even AlphaGo's development team would be able to pinpoint exactly what set of final rules are used by AlphaGo to make its moves and why!

AI without ML - Humans provide the rules to the machines

AI with ML - Humans provide only the data. The machines learn the rules themselves.

You may have seen nice following circular graphs where ML is displayed to be a subsection of AI and DL is shown to be a subdivision of ML.

Well the fact is in recent times; ML is hijacking almost all AI space and there is very little non-ML AI development happening. So, you can say that most of the AI systems today run using ML.

AI began in 1953 when Claude Shanon at Bell labs hired two assistants named Marvin Minsky and John McCarthy setting in motion a chain of events that was to have wide-ranging implications for human-kind. They had a common interest in a quaint scientific field of those times called 'thinking machines'. Turing had a couple of years back proposed his now-famous Turing test - ‘a computer can be said to be intelligent if a human judge can't tell whether he is interacting with a human or a machine’ and it was a hot subject in those days. Anyway in 1958, the three of them came up with an interesting proposal requesting a break from regular work for 8 weeks and for funding of a 'series of brain-storming sessions' to discuss this new field which they formally titled 'Artificial Intelligence'. While I am sure that in modern-day this kind of proposal would raise eyebrows, it did get approved and the rest is history!

While John McCarthy gave the general definition of AI as “the science & engineering of making intelligent machines or machines that think the way humans think”, Arthur Samuel in 1959 defined Machine Learning (ML) as - “a field of study that gives computers the ability to learn without being explicitly programmed”.

In those days and for several decades afterward, ML was one of the (several) techniques by which AI (“making intelligent machines”) could be achieved. Starting in the late 90’s the face of AI changed as never before! The Internet era threw in an abundance of DATA. This fueled up ML-like never before because as we discussed in ML systems what happens is - Humans provide only the data. The machines learn the rules themselves. They don't need explicit programming.

So, most of the AI systems today are based on ML. As to what DL is, it is a subset of ML inspired by biological systems like the human brain which uses multiple layers to progressively extract higher-level features from the raw input. For e.g. when we see something, data is passed from our eyes to the brain to be interpreted. The brain identifies the object thru’ several layers of processing at first it will identify the edges and corners, while subsequent layers extract higher-level features, and finally, we see whole features like digits or letters or faces. The adjective "deep" in deep learning comes from the use of multiple layers in the network.

This is a crisp definition meant for the layman without going too deep.