In the last lesson, we discussed that the goal of AI is to enable machines to think. We also understood that machine learning (ML) is a subset of AI. It teaches machines by providing them with data and allowing them to learn patterns and relationships within it.

‍

If we say that AI is about enabling a machine to think (and act) like a human, then a question that needs to be answered is how do we determine that machine has actually started thinking like a human. The answer to this question was first answered with the introduction of the concept Turing Test.
Dr. Alan Turing first introduced the concept of the Turing test in his influential 1950 paper, "Computing Machinery and Intelligence".

‍

The concept behind the Turing test was that if a human judge, interacts simultaneously with a human participant, and a machine participant, and cannot distinguish between them, then the machine is considered to have passed the Turing test.
The argument is that if a machine can successfully imitate human behavior in a conversation, demonstrating understanding, reasoning, and creativity, it would exhibit a level of intelligence comparable to that of humans.

‍

Alan Turing did acknowledge that creating such a machine would be a challenging task. He hypothesized that one way for a machine to pass the Turing Test, demonstrating human-like intelligence, would be to simulate the behavior and learning processes of a child.

This sparked a new wave of research to create a thinking machine, which is to learn how a child learns. throughout the next few modules, I will invoke the child or the parent within you. I request you to play along. The more deeply we understand how a child learns, the easier machine learning will become for us.

There are 3 primary ways of Learning Techniques that we will discuss in this module. Later, we will add a few more.

1. Supervised Learning: A machine learning technique where the algorithm learns from labeled data, consisting of input-output pairs.
2. Unsupervised Learning: A machine learning technique where the algorithm learns patterns or structures from unlabeled data without explicit supervision or guidance.
3. Reinforcement Learning: A machine learning technique where the algorithm learns by taking an action and receiving rewards, positive or negative.

‍

ML Technique 1: Supervised Learning
‍

A machine learning technique where the algorithm learns from labeled data, consisting of input-output pairs.

‍
What is labeled Data? To understand this let us go back and understand how a child learns. Suppose you want to teach your 12 months for the first time ‘What is an Apple’? How do you do that? Think about it before you read further.

You will show her an Apple or an image of an Apple. And say loudly this is an ‘Apple’.

The image of Apple in this case is ‘Input’. The right answer that you are saying loud, in this case, is ‘Label’. So you are giving the kid labeled input data. Label in this case is the ‘right answer’ to the input data.

‍

When you repeat this multiple times i.e. showing various images of Apple and saying ‘Apple’. The kid learns that it is an Apple. But now what happens when you show her for the first time Banana and ask what is it? Will the kid say ‘Banana’? Think about it.

No, the kids will not say Banana. Instead, it will say ‘Apple’. Why? Because the kid has no notion of Banana. She only knows ‘Apple’. Then you will have to tell her - This is ‘Not An Apple’. Now you will show multiple other images and some will be labeled as ‘Apple’ and some as ‘Not Apple’. When you repeat this enough times - eventually the kid will start saying ‘Apple’ when shown unseen images of Apple and ‘Not an Apple’ when shown unseen images of other fruits. We will say the kid is trained now.

This technique of learning in the Machine learning world is called ‘Supervised Learning’.  In other words, in the above example, the kid is learning under your supervision. We will see multiple examples of this in the next sections. But for now, I hope we understand Supervised Learning.

‍

ML Technique 2: Unsupervised Learning

A machine learning technique where the algorithm learns patterns or structures from unlabelled data without explicit supervision or guidance.

Image in the previous example, now you cannot say ‘Apple’ or ‘Not an Apple’. So now the input data is unlabelled. Can they learn what is Apple?

No. Because you have never said the word Apple or Not An Apple. The kid has no notion of these words. So what can you teach the kid? This is the problem that unsupervised learnings deal with as in many cases we cannot provide labels. Let us see what we can teach the kid to do.One thing that kids can learn to do on their own is finding patterns and grouping similar, based on some characteristic, things together. In Machine learning, this is called Unsupervised Learning.

‍

ML Technique 3: Reinforcement Learning

A machine learning technique where the algorithm learns by taking an action and receiving rewards, positive or negative.

If you remember your kids’ first walk. How did she learn to walk?

Imagine Shania, a one-year-old toddler, who's learning to walk. Shania's environment is her living room, filled with furniture, toys, and obstacles.

1. Exploration: Shania begins by exploring her environment, crawling around, and experimenting with standing up and taking steps while holding onto furniture for support.
2. Trial and Error: Shania tries different actions, such as reaching for a toy or attempting to stand independently. She experiences success when she manages to take a few steps without falling and reaches a desired object.
3. Feedback: Shania receives feedback from her environment and you, her caregivers. When she successfully takes a few steps without falling, she might receive a positive reward in the form of smiles, claps, and encouraging words from her parents. Conversely, when she falls or struggles, she does not get a clap but a sympathetic tone. This acts as a negative reward.
4. Learning from Experience: Over time, Shania learns from her experiences. She remembers which actions (e.g., shifting her weight, moving her legs) led to successful walking and which actions led to falling or stumbling. She gradually improves her motor skills, balance, and coordination through repeated practice and experimentation.
5. Generalization and Adaptation: As Shania becomes more proficient at walking in her living room, she begins to generalize her newfound skill to other environments, such as the backyard or playground. She adapts her walking technique to navigate different surfaces, slopes, and obstacles she encounters.

This technique, where Shania is learning from trial and error, receiving positive and negative signals from the environment or parents, is an example of Reinforcement Learning.  This is how robots learn to walk - By exploring the environment, trial and error, receiving positive and negative rewards, and learning from the experiences.

We will discuss this in much more detail later in section 4.

‍

How do the three forms of learning relate to Machine learning and Deep Learning

Now that we know the three primary forms of learning, in the next few modules, we will dive deeper into each one of them and gain a deeper understanding through some business applications, and foundational models within these learning techniques, and look at some simple hands-on exercises. It will be exciting and easy. Trust me.

‍