What is AI? What are different techniques to make a Machine Learn ?

What is AI?

Let us learn AI simplistically. First, in short we will understand what AI means. As we know, AI stands for Artificial Intelligence, which means we are artificially trying to make a system intelligent by giving it different types of inputs or data, so that it understands the world or domain. You can relate this methodology to how a child learns things in this world: parents show them lots of things (input), teach them what the input is (process) and then ask them what it is (output). So, simplistically put, AI is all about making a system learn by giving it a variety of data, processing the data (teach the system about the input data or make it learn about the data) and generating the output. Using this technique we are making the machines learn about the domain (eco-system), and hence Machine Learning (ML) is a subset of AI.

Hope you got a simplistic idea on what AI is.

Now let's delve into our main topic:

Types of Machine Learning or AI Learning Models OR How to make a system (machine) artificially intelligent (AI) using different techniques.

There are various ways to train a machine. Let us try to understand them by drawing parallels with real life examples:-

1) Supervised Learning:

It is like a teacher/parent teaching a student/child or helping them perform a task. You show a child lots of objects and tell them what each object is. Then you ask them what the object is.

If we talk in technical terms, given an input A, supervised learning can decide the output is B. For e.g, if we give input as an image of an apple and label it as Apple, the machine will give the output as an Apple but it will require thousands of different types images of apples including those which are of different colors, sizes and pixels for it to correctly identify it as an apple. This essentially means, more the training data more accurate are the results. This also means that if we train an AI system with lots of data (e.g hundreds of thousands of words) we get an application like ChatGPT also called Large Language Models (LLMs).

2) Unsupervised Learning:

It is the opposite of Supervised Learning, i.e., learning without a teacher (supervisor). Here, it is difficult for a student to learn by himself/herself and hence outputs won't be so accurate. When we provide an image of an Apple, it will not be labelled as Apple like it is done in supervised learning.

In technical terms, when inputs are provided to an AI model, they are not labelled. So how would the machine infer what is the input for it to provide the output? It will use its program (algorithm) to process the data based on similarities, differences and patterns.

3) Semi-Supervised Learning:

It is a hybrid approach where some data is labelled and some is not and the data is then processed by the language model.

4) Reinforcement Learning:

Reinforcement means the process of encouraging or establishing a belief or pattern of behavior. This is something analogous to a child learning a cycle by interacting with the environment like the road, speed breakers, side walks, signals etc.

So Reinforcement Learning is a type of machine learning where an agent learns to make decisions by interacting with the environment. E.g. Robot.

I hope this article gives you a brief and simplistic idea on Artificial Intelligence and the different model training techniques.

Happy Learning!!

Source: https://aws.amazon.com/what-is/machine-learning/

For more interesting reads:

https://www.guninatech.com/#blog

https://www.guninatech.com/blogs/the-microsoft-global-outage