What is the Use Of Reinforcement Learning in Artificial Intelligence? ( Simply Explained )

Reinforcement learning and Its Role In AI

reinforcement learning explained

Many people tend to confuse themselves thinking reinforcement learning and AI(Artificial Intelligence) is different, some don't understand what is the reason for the existence of reinforcement learning! If you are in that place, don't worry this article will clarify it for you!

These are the topics I will be covering in today's Post: -
  • What is Reinforcement Learning?
  • How does Reinforcement Learning Work?
  • Examples of Reinforcement Learning
  • Conclusion and What's Next 

In the world of AI, the first two things that we have to keep in mind is the environment and the agent. The environment can be a place or a simulation in which our agent or AI is placed in for accomplishing a particular task.

In this article, we will explore a very simplistic approach of what is reinforcement learning and then eventually build-up to the more complicated approach in the next articles...

Want all this content and more in audio format? checkout our podcast!

What Is Reinforcement Learning?

According to Wikipedia,

Reinforcement learning (RL) is an area of machine learning concerned with how software agents ought to take actions in an environment in order to maximize the notion of cumulative reward.

Reinforcement learning is one of three basic machine learning paradigms alongside supervised learning and unsupervised learning.

for simplicity purposes...

Think of it like this, 

The AI is there in an environment. Then It takes action, after which it receives a reward that affects the algorithm of the agent which shows changes in the next action. 

Reinforcement learning workflow

This cycle continues until it attains mastery over the environment. This Cycle of learning from results by reinforcing the outcomes of the actions can be considered Reinforcement Learning.

How Does Reinforcement Learning Work?

Yes, most of it is already mentioned in the definition. 

The AI reinforces it's understanding from the outcome and changes its next action to improve the next action (what we call learning). 

Now let's understand some terms in Artificial Intelligence fundamentals

Here are some terms you need to know,

  • State: The current position/level/condition in an environment 
  • Reward: The outcome of an action (can be positive/negative)

Let's dive a little bit deeper... using this maze for better understanding.

Artificial Intelligence Solving Problems

Here let's say the red box is the unwanted result and let the green box be the wanted result ( the place you want the Artificial Intelligence to stop at )

Now the AI is offered with a set of options  (actions) that it can take. 

Then it is told to achieve a particular result, it's not given pre-coded actions or directions. It has to figure it out!
This is exactly why reinforcement learning sometimes does better than humans or programs we make.

The AI does some random actions and ends up with a result. If it goes to the green box, it wins! and that makes it want to repeat a similar action in the future and vice-versa for losing.

Making the program learn, just like training a dog.

Examples of Reinforcement Learning

1. Training a Dog

Let's say a dog begins training. It's commanded to sit and it doesn't react.

What does the trainer do?

He associates a negative feeling with it, in terms of AI it gets a negative reward (gets hit or no treat) The next time it sits for the same command. It gets a treat (positive reward).

The trained rewards it with biscuits or a bone.

This cycle is done for potty training a dog or teaching to guarding a house. It is very similar to how AI learns.

How is in the realm of AI?

However, in the world of AI, there is a reward that we create, which is a highly valued item. The speed of learning is always highly improvised because it's done on a computer.

The same happens here but,
it's a digital reward

+1 and-1

When something is trained through reinforcement learning it doesn't have a pre-program on how to do its work and that's why robot dogs have learned to play football but many many normal dogs have not!

The Bot or AI is only given the Goal and Desirable result when it starts then,

  • When it does not create a desirable result it's given a -1. Hence, does not repeat the mistake

  • When a desirable result is created the AI is given a +1

  • eventually, from this process, it learns how to reach a given goal.

This can be implemented in so many aspects and here is a list of them! I would like to think of them as a combination of things because AI doesn't stand alone.

1. Cooking with AI
2. Fishing with Deep Learning 
3. Gaming with Pattern recognition and prediction
4. Pattern Recognition with Machine Learning
5. Population Control and Facial Recognition
6. Stock Market with Machine and Deep Learning
7. AI with Cloud Computing

The list just goes on... however what do you want to use Artificial Intelligence for? let me know in the comments!

If you have any suggestions or questions feel free to comment below or reach out to me on Social Media!

Post a Comment