Since the beginning of the year 2015, the world has experienced a boom in the realm of Artificial Intelligence, Machine Learning, and Deep Learning. These technologies are getting huge coverage in the media too, with some heralding them as futuristic and science fiction like, while some are a little wary of what the consequences of all this can be?
But there has been a big confusion among most people as to what these technologies actually are, as they all sound quite similar.
The short answer is that Artificial Intelligence is the ability of machines to think and make decisions like humans, while Machine Learning is one of the approaches in which AI can be achieved and Deep Learning is a technique for Machine Learning. This is simply summarized by this easy-to-understand Venn diagram.
Still not clear? Let’s discuss it in detail.
What is Artificial Intelligence?
Intelligence is the capability of someone to understand a problem, gather data related to it and take a decision to solve it. Humans are intelligent. Animals are intelligent too, but quite lesser than humans. Are machines intelligent too?
Some of them are, and that ability of machines to make decisions is called Artificial Intelligence. Whenever people hear the term AI, they immediately make a mental picture of "The Terminator". First of all, HE is not AI, his CPU or brain is AI, i.e. "The Terminator HAS Artificial Intelligence". Secondly, he has a "General AI", a very human-like intelligence that uses senses, thinks, takes its own independent decisions and uses actuators to perform actions.
This type of AI is still a far-fetched dream for us.
Although, what we have achieved is called a "Narrow AI".
These are things that can carry out specific tasks like recognizing faces of your friends in pictures or separating spam from your emails, just like humans or better than us.
How do you tell a machine to do a task? You give them an "algorithm", which is the set of rules that need to be followed to solve the given problem. Simple problems that have a fixed problem statement and the same set of data every time are solved with simple fixed algorithms and don't require a "learning" process.
For example, an ATM machine that dispenses cash doesn't require a learning process. It uses a fixed algorithm to dispense bills as per the input that you give (the amount of money you need) and the amount that it has and dispenses the right amount.
What if a new currency denomination is introduced? Does it learn by itself about it? No. You have to put it into its algorithm and then it'll work just fine. This is not an AI-driven machine.
But Facebook can identify the photos of many of your friends. You don't need to tell it the name of your friend; it automatically learns that from the data. It can even recognize them in dark photos or from multiple angles.
In a way, Facebook is intelligent enough to learn on its own! How is it possible?
It is possible through Machine Learning, which we'll discuss in the next section.
Machine Learning
Machine Learning is a type of approach to accomplish Artificial Intelligence. The word "learning" implies that the machine can actually be trained to do a particular task on its own, without the need to code it directly in the algorithm. In a way, the machine makes its own algorithm depending on the initial set of instructions and a huge set of data and examples that it learns from.
ML enabled AI can perform tasks even better and faster than humans, all depending on the data that is provided for its training. If the data is flawed, the learning and implementation will be flawed too.
Some applications of ML are email filtering, learning to rank, grammar checker, and computer vision.
Deep Learning
Deep learning is one of the techniques that can be used to implement machine learning. It is based on Artificial Neural Networks, which is our attempt at mimicking the functioning of the human brain. Although current deep learning algorithms are quite far from our biology, it is a step in that direction.
Deep learning networks have several layers, nodes, and interconnections as well as directions in which data can propagate. It utilizes humongous quantities of data, thanks to modern GPUs, to learn from them, and hence doesn't need programming to "define" items.
A popular example of Deep Learning based AI is what Andrew NG of Google did in 2012 with his "Car finding algorithm". Another great example is Google's AlphaGo, that trained itself to play the board game GO and finally defeated a South Korean master in it.
Conclusion
These three terms are used interchangeably in the media quite often but are different actually. To learn about these concepts in a better way, it is essential to identify what each actually means.
Still having any confusion or have anything to add to it? Leave it in the comments below.
You may also like:
