The easiest way to think about AI is in the context of the human. After all, humans are the most intelligent creatures we know of.
The goal of AI is to create systems that can function intelligently and independently. Humans can speak and listen to communicate through language, this is a field of speech recognition. Much of speech recognition is statistically based hence it’s called statistical learning. Humans can write and read text in a language, this is the field of NLP or natural language processing. Humans can see with their eyes and process what they see, this is the field of computer vision. Computer vision falls under the symbolic way for computers to process information, recently there has been another way which I will come too later.
Humans recognise a scene around them through their eyes which create images of that world. This field of image processing which even though is not directly related to AI is required for computer vision. Humans can understand their environment and move around fluidly, this is the field of robotics. Humans have the ability to see patterns such as grouping of like objects, this is the field of pattern recognition. Machines are even better at pattern recognition because they can use more data and dimensions of data. This is the field of machine learning.
Now let’s talk about the human brain. The human brain is a network of neurons and we use these to learn things. If we can replicate the structure and the function of the human brain, we might be able to get cognitive capabilities in machines. This is the field of neural networks. If these networks are more complex and deeper and we use those to learn complex things that is the field of deep learning. There are different types of deep learning in machines which are essentially different techniques to replicate what the human brain does. If we get the network to scan images from left to right, top to bottom, it’s a convolution neural network. A CNN is used to recognise objects in a scene, this is how computer vision fits in and object recognition is accomplished through AI. Humans can remember the past, like what you had for dinner last night, well at least most of us. We can get a neural network to remember a limited past, this is a recurrent neural network.
As you see there are two ways AI works, one is symbolic based and the other is data based. For the data-based side called machine learning, we need to feed the machine lots of data before it can learn. For example, if you had lots of data for sales vs advertising spend you can plot that data to see some kind of a pattern. If the machine can learn this pattern, then it can make predictions based on what it has learned. While one, or two or even three dimensions is easy for humans to understand and learn, machines can learn in many more dimensions like even hundreds or thousands. That’s why machines can look at lots of high dimensional data and determine patterns. Once it learns these patterns it can make predictions that humans can’t even come close too.
We can use all these machine learning techniques to do one of two things: classification or prediction. As an example, when you use some information about customers to assign new customers to a group like young adults, then you are classifying that customer. If you use data to predict if they are likely to defect to a competitor, then you are making a prediction.
There is another way to think about learning algorithms used for AI. If you train an algorithm with data that also contains the answer, then it’s called supervised learning. For example, when you train a machine to recognise your friends by name you will need to identify them for the computer. If you train an algorithm with data where you want the machine to figure out the patterns, then it’s unsupervised learning. For example, you might want to feed the data about celestial objects in the universe and expect the machine to come up with patterns in that data by itself. If you give any algorithm a goal and expect the machine through trial and error to achieve that goal, then it’s called reinforcement learning. A robot attempt to climb over the wall until it succeeds is an example of that.
So there you go!