AI vs ML vs DL
AI vs ML vs DL, the very phrase brings out a raging sense of conflict to our minds, but this is far from reality. Actually, these are all deeply interconnected terms in their beings. Deep learning (DI) forms a part of Machine Learning (ML) which in turn is a component of Artificial Intelligence (AI).
So, in our process of understanding the interconnect features and at some points- contrasts of these aspects that are today being commonly used by programmers and software developers all across the globe, we first stop to understand their meanings.
Put in simple terms, anything intelligent in a computer can be considered as a part of Artificial Intelligence, it can be your basic switch/break statements or extremely complicated piece of code assorted in a most intricate manner. What is important in AI is its ability to imitate intelligent human behavior in computers. Machine Learning on the other hand, though is a part of AI, is so extensive on its own that it has become a subject of intense study and research in itself. It is described as the ability of the computers to use a given piece of information (it can be anything like input by the user, coder, experience) to teach itself something new. Arthur Samuel, who was considered to be a pioneer in this field himself called machine learning as a “field of study that gives computers the ability to learn without being explicitly programmed.” Even further to this we now enter the domains of deep learning, which forms a yet more complex and diverse part of the studies of computers even though it effectively remains to be a subset of Machine Learning. The purpose or objective of deep learning is the same as that of the machine learning but with the added factor of accuracy. This much is maybe eminent from the prefix “deep” used. Deep actually is a technical term and it indicating a multitude of neural networks in the machine, these networks interact with each other and learn about the features of any data in accordance with a phenomenon called “feature hierarchy” where neural networks of simple feature interact with each other to raise a more complex one.
AI is the intelligence most casually observed by you while interacting with computers, simple tasks like recognizing images, user patterns, learning to solve problems, etc. It is broadly classified as narrow and general. While the general AI boasts all the features of human intelligence, narrow AI is more specific in its application. A good example of your interaction with such would be- the Amazon shopping application recommending you certain items are all carried out through AI, the computer extracts your previous purchases, analyses it and then goes as far as to predict what you may want to buy next. This is done for each of its users and at times is pretty accurate in pinpointing to the users need. While this sounds pretty interesting, but to gain a window to its actual working, we shall have to refer to Machine Learning. It should be understood at this very point that AI can be achieved without using ML as it has in many software in the past, but this would a direct invitation to writing a vast amount of complex codes that could have been otherwise skipped.
Machine Learning is how we teach the computer to observe certain things and then start using this to teach itself of new abilities hence enhancing its further functioning. Its modus operandi is quite simple- the computer receives data with some human features, it is then trained to tune the relative importance of each feature and then finally to predict something new from this data. An example of this can be the feedback mechanism for a company with a huge online presence, the company shall provide the algorithm with a large amount of data about what is spam and what is not (this is called labeled data), this algorithm will further identify these common features in the data and then when run upon new feedback messages received they shall be able to predict whether they are spam or not. As can be concluded from this that ML requires a lot of human interference and input and its results are not always accurate.
It is these shortcomings more than anything else that extends the importance of Dl. While the idea, came as long ago as the 1950s it is the youngest in terms of development among these three. Deep learning is based on the neural model of the human brain and is used to enhance ML and in some cases achieve it. By using Artificial Neural Networks (ANN) they mimic the biological functions of the brain. Its multiple layering and network connections make an intricate and widespread development model that is the most accurate one among its kind. Each layer works on a specific feature and learns about it and hence combining all of these layers a certain depth is generated.
Thus, we can realize how even though being a part of each other and hence heavily connected, each of these features seems to be highly different from each other. When compared we can see an increasing sense of complexity, that may have something to do with the discrepancy in the amount of research done on each other. But not just that, the mode of operations, the basic models and the amount of human interference needed varies immensely as we go from outside to inside. All of this calls for a greater understanding of these concepts and with research making exponential progress we may very soon have a lot of data and hence a lot of utility is derived from these very concepts.
