Artificial intelligence is all around you. Self-driving cars, spam filters, email classification: as soon as you get an email, Gmail would automatically classify it into different boxes, and fraud detection. As soon as you do a bank transaction, there is a fraud detection algorithm which is running behind the scenes. All of this work is based on AI, or artificial intelligence.
What is AI? According to the Oxford Dictionary, “AI is the theory and development of computer systems able to perform tasks normally requiring human intelligence.” Examples of such tasks are visual perception. A human can look at a picture and recognize different objects in it. A human can recognize if the people in the picture are happy or not. A human can recognize speech. When somebody is talking about something, if that’s a language I can understand, I’d be able to understand what they’re talking about. I can make decisions, and I can translate between different languages.
So, AI is nothing but the “theory and development of computer systems that are able to perform tasks that need human intelligence.” If we break it down further in simple terms, the goal of AI is to create machines that can simulate human-like intelligence and behavior. For example, play chess like Magnus Carlsen or Viswanathan Anand, make purchase decisions that humans usually make, what to buy, when to buy, drive a car like a human, and write an essay like a student.
So, what we’re talking about is the fact that the goal of AI is to create machines that can simulate human intelligence and behavior. And there are two different types of AI.
- Strong artificial intelligence or general AI
- Narrow artificial intelligence or weak AI
Strong artificial intelligence or general AI
The goal of strong artificial intelligence is to create machines that are as intelligent as humans. So, the intelligence of a machine is equal to the intelligence of a human.
What does that mean? It means creating a machine that can solve problems, that can learn, that can plan for the future, a machine that is expert at everything, including learning to play all sports and games, a machine that can learn like a child, building on its own experiences. A child can learn to do a lot of things just by observing others. The goal of strong AI is to create a similar machine which can learn like a child. The thing is, we are very, very far away from achieving this. The estimate is that it would be at least a few decades to never before we achieve this.
Narrow artificial intelligence or weak AI
Typically, whatever AI systems we see in action today belong to the category of narrow AI or weak AI. These AI systems focus on a specific task, for example, a self-driving car. A self-driving car can only drive. It cannot do other things. It cannot play chess, for example. A chess-playing machine just plays chess. Another example is a model that can predict a house price. It can only predict house prices.
So, what we’re talking about is the fact that the goal of AI is to create machines that can simulate human-like intelligence and behavior.
Artificial Intelligence and Machine Learning
The goal of AI is to create machines that can simulate human-like intelligence and behavior. And machine learning is a subset of artificial intelligence. It’s a particular approach to developing AI solutions.
How does machine learning differ from traditional programming? Traditional programming is based on rules. If this, do that. A good example is if you want to develop a program to predict the price of a home, you would design an algorithm that considers all the factors. You’d take the location of the home, the size of the home, the age of the house, the current condition of the house, the market, and the economy. You design an algorithm considering all these factors.
However, if you’re doing machine learning, it would learn from examples, not rules. Whatever we do in traditional programming is rules. We write a number of rules. We would design an algorithm, and we would write a program to calculate the house price. However, when it comes to machine learning, we are learning from examples. We would try and learn from millions of examples. If you want to predict the price of a home, what we would do is we would get all the historical prices for millions of homes, and based on this data, based on the data which we gather, We would create a model. and after that, we would make use of the model to make the predictions.
The most crucial difference between the traditional programming approach and machine learning is that machine learning tries to create models by learning from existing data and examples. Using these examples, we would create a model. The machine learning model would be used to make predictions.
What is deep learning?
Deep learning is a form of machine learning only, but the logic behind it is based on the concept of neurons that we humans have. If you think about it, right, how does the human body transmit information?
Our body has a network of neurons that are tied to each other. For example, when you see something from your eyes, these neurons transmit information from one to another. Eventually, it reaches your brain, which analyzes the object and tells you that what you are looking at is an apple or cat.
So, neurons are like multiple layers, with each one connected to the other, and they process information. They are transmitting it to the next stage. And this whole idea of having multiple stages, multiple layers, each one processing information and enhancing it, sending it to the next one, is what really makes humans so intelligent.
So the engineers thought that, okay, that is great. This is what makes humans awesome. So, how do we replicate it? What if we apply the same concept to machine learning? And that’s where they created something called neural networks.
As you can see, there are also multiple computational layers here. Each layer analyzes the information from the previous layer, learns from it, enhances it, and then passes it on to the next layer. This next layer again processes it, analyzes it, learns from it, and sends it to the next one, and so on.
So the idea here is that when we have a neural network like this that has multiple layers analyzing us, our accuracy on complex problems like text generation improves significantly because we are now passing the training data through many layers, and each layer is analyzing that data. It is doing computations on it; it is learning from it, predicting things. And that improves the overall output. That helps you achieve much, much greater accuracy. Actually, this neural network is the model that generative tools like ChatGPT are also using. And we will see it in more detail in the following modules.
However, the key takeaway here is that for complex problems like text generation or image generation, deep learning models like neural networks outperform other older machine learning models. And that is simply because of their multi-layered structure that you see over here. Another thing that you may have guessed by now is that this solution is very complex. It would require a lot of computational power to move the data, analyze this data, process it, enhance it, predict it, or whatever.
You are totally right. Neural networks do require a lot of computational power, which has limited their adoption and improvement in the past. But as I said earlier, with the recent advancements in computational power, GPUs, and whatnot, neural networks have become much cheaper to train and much cheaper to use, and that is driving their adoption, especially in areas of Generative AI.
So, what is Deep Learning? Let’s take a look at the definition now. It’s a subset of Machine Learning, and it processes data through a neural network. This results in more accuracy for complicated problems. With a sip of a cup of tea, let’s learn or study something.
What is generative AI?
The goal of artificial intelligence is to create machines that simulate human-like intelligence and behavior. The goal of machine learning
is to create artificial intelligence using a specific approach to learn from examples. Instead of writing complex algorithms and complex logic, you try and learn from examples. You’d have millions of rows of data, and you try and learn from that data. That is what machine learning focuses on.
Machine learning is a subset of artificial intelligence, so you can see that in the picture. So you have the big circle, which is artificial intelligence. Inside that, you have machine learning, which is a smaller circle, and the smallest circle that you see here is generative AI. Typically, most machine learning solutions that we talked about until now are used to learn from examples and make predictions.
However, in generative AI, the goal is to learn from examples and create new content. So, when we are using generative AI, what we are doing is machine learning, and the goal of this specific approach to machine learning is to create new content. That’s why you can see that this is a smaller circle here.
So artificial intelligence is the biggest circle here. You’d want to create machines that are really intelligent and that can do everything and anything that a human does.
Machine learning is about learning from examples, and generative AI is a specific category under machine learning where we learn from examples to create new content. So, the goal of generative AI is to generate new content.
Reference:
- https://www.roboticmarketer.com/what-does-ai-actually-mean-and-how-does-it-work/
- Sheehan, K. (2017). The ongoing audit transformation. Accountancy Ireland, 49(6), 54-55.
- Tombone’s Computer Vision Blog: Deep Learning vs Probabilistic Graphical Models vs Logic. https://www.computervisionblog.com/2015/04/deep-learning-vs-probabilistic.html?showComment=1458438261799
