New light on AI will be shed for you

Artificial Intelligence (AI) is now becoming a bigger conflict than ever. Many people worry about robots taking over the world. The concept of AI scares people because we don't know how they work because we fear to create bots. What if I told you that the majority of statements you heard about AI were inaccurate? When I say inappropriate, I don't shy away. The media is very wrong when it comes to AI. How can you talk about AI when you don't know how AI works and have no experience with it? Even worse, most of the content produced by the media does not include testimony from experts in the field of AI. At this point, it still makes sense if you don't trust AI; I don't expect you to agree with me right off the bat. But hopefully, by the end of this article, new light on AI will be shed for you.

Let’s first talk about how AI is being implemented in today’s world. Many people think that AI is something that is yet to come in the future, but what they don't realize is that AI is very prevalent in today's society. Gmail uses AI to filter spam. Facebook uses AI to refer to friends. YouTube and Netflix use AI to recommend other videos and movies. Ever heard of data analytics? It is mostly machine learning, a specific area of   AI. Facial recognition uses AI on new iPhones. Most video games use AI these days. Simply put, AI is already a common theme in most of the technologies we use today. The truth is, people are very unfamiliar with AI; we don't care about it big. I understand that AI has some flaws. For example, if implemented in factories for the production of goods, AI could replace many employees, resulting in greater unemployment. But when is there anything that lacks flaws? Not only that, there are many scenarios where AI shows no risks, and AI can solve some of the biggest problems we face today. For example, AI has already been able to diagnose various diseases, especially cancer, with much greater accuracy than current methods. There was also a study in which 10 different radiologists were given different mammograms of breast tumors. These radiologists have shown that they have a 10-55% variation in their diagnosis. According to the American Cancer Society, 1 in 5 screening mammograms is misdiagnosed. On the other hand, experts have been able to diagnose Artificial Intelligence with 85-90% accuracy. Yet many people say they don't trust AI…

Why are people afraid of AI? Why do people claim that we do not understand? To address these issues, I will discuss some algorithms in the field of machine learning.

AI comes down to a lot of math and a lot of logic. Have you heard of linear regression (also known as the best fit line)? What if I told you that linear regression is an example of AI? Of course, it is; this may be one of the most basic algorithms in AI, but it is AI. You input data to create a model (best-fit line) and then make an estimate based on the line. Most of the AI   works the same way; the only difference between the algorithms is the generated model and the purpose of that model.

Machine learning is classified as supervised and unsupervised. Supervised learning is the practice of having labels for our data. Labels can be thought of as a result. For example, if we have a cancer dataset, each data point is labeled as a tumor, whether the tumor is malignant or benign. So, in supervised learning, we get these labels, but in unsupervised learning, we don't have those labels. Supervised learning can take the form of regression or classification. In regression, when categorizing you try to estimate the output value of the input, you try to estimate a certain class of input. Unsupervised learning often involves clustering algorithms; you plot the data and try to group the different parts of the data into groups to find relationships and patterns in the data. Regardless of the machine learning algorithm you use, all of this is based on math. It is the computer that does the most calculations, pulling the numbers into different equations. There is no intelligence behind this, hence the term Artificial Intelligence; it’s not real intelligence, and it sounds like that.

There is a specific type of machine learning called deep learning, which is based on algorithms called neural networks. There is a lot of controversy behind AI here. Neural networks are algorithms that simulate the brain; this can be confusing for many because they think that neural networks have their intelligence. They are difficult to understand for sure, but they are nowhere close to human-level intelligence. Neural networks have three types of layers: the input layer, the output layer, and the hidden layer between the input and output layers. Each layer contains a certain number of neurons defined by the programmer, and each neuron is connected to all the neurons in the previous layer and all the neurons in the next layer. Each neuron has weight; Neurons not in the input layer carry the weight of neurons in the previous layer. Each neuron also has a bias, which is added to this amount of weight to modify the inactivity of that neuron. This weighted sum is implemented by the activation function, which converts that weight into a number on a scale of 0 to 1. This process continues for each neuron with each layer until the output layer is reached. Let’s take the breast cancer diagnosis example I used earlier. If we create a neural network for this problem, we assume that the inputs are parameters related to the malignancy of the breast tumor and that the output layer contains two neurons. One neuron represents benign, and another indicates malignancy. We input the parameters of the tumor into the network, and the benign neuron returns 0.02, while the malignant neuron returns 0.98. Since the malignant neuron is overweight (1 means 100% weight), we assume that the tumor is malignant.

Now, going back to layers, in the input layer, you put your inputs, and in the output layer, you get the output. They work similarly to other types of machine learning algorithms. However, as of today, we don't understand hidden layers, but we do know that the purpose of the neural network algorithm is to find a set of weights that provide a very accurate neural network model. We're not sure about explaining what's going on in the hidden layers.

To illustrate this uncertainty, let us simulate a billiard with a neural network. The neurons in the input layer represent each ball. The neurons in the output layer represent each pocket. Each hidden layer represents a parameter that affects where the ball is going. A hidden layer represents other balls that can hit the ball. In the simple game of pool, we have a very easy time to sort out which ball goes into the pocket; we need to do very little calculations so that we can keep track of where the ball is going. We can tell how the ball bounces off the walls if the walls are flat, and how the ball bounces off the other balls if the balls are perfectly spherical. It is a very simple neural network. Now, we will add more balls. There are more variables where the ball travels, so it can be a little difficult to determine which pocket the ball is entering. This neural network is a bit more advanced than the previous one. But for now, we will add many more hidden layers. A hidden layer indicates the walls, but this time, the walls are not flat; they may be skewed or spikes. A hidden layer represents the surface of each ball; What if they are not perfectly spherical and are unevenly skewed? A hidden layer represents the earth; Instead of having flat land, the land is now skewed. At this point, you cannot predict where the ball is going. It is very sophisticated with many variables and many more computations.

This is a problem we have with neural networks. Most neural networks used to take up large datasets with very large parameters, so the neural network is very sophisticated to understand. We can try to break it down to understand what is going on, but it will take a long time. Training of neural networks can take anywhere from a few minutes to a week. We work so hard to handle everything and have trouble putting the pieces together. Like neural networks and other machine learning algorithms, there is no real intelligence; they are so complex that they seem to have their minds. The problem with neural networks is not our lack of control over them; this is their explanation. It is very difficult for us to understand how the algorithm gets its estimation correct. However, there is much research on how we can make more precise, understandable and efficient neural networks. Neural networks have lots of potentials. They can achieve high accuracy, with some exceeding 90% accuracy. They also can be used for very sophisticated tasks such as image classification/recognition and natural language processing (analysis of a text). Think about the potential applications of neural networks. Their implications are HUGE. For example, machine learning has become popular for cancer diagnosis. The implementation of machine learning in bioinformatics can help patients to get treatment at an early stage, saving many lives.

AI can solve many real-world problems. Not only that, but it’s been shown that AI can solve these problems more accurately than modern methods. We could make lots of progress thanks to AI. However, the false statements claimed by many on AI today are simply hindering this progress. That being said, it is important to take into account the drawbacks of AI. There are concerns that robots will take away jobs from people, resulting in more unemployment. & what if there was an Artificial Intelligence that could create AI? That’s scary, even for me. But AI isn’t the first thing we’ve had were too much of it is bad. Having too much medicine isn’t good because bacteria will develop antibiotic resistance, but not having medicine when it’s needed isn’t good either. Working too much isn’t good because you need enough sleep, but you also need to work to earn enough money. For AI to be used effectively and appropriately, there needs to be a balance. There should be general support for the use of AI, but there should also be regulations as to what can be made with AI. AI should be promoted, but we have to make sure that we don’t make AI that can create other AI because, at that point, we won’t have much control over the program. As long as we can pull that off, AI will be the future of technology, and it will help the world to become a happier and healthier place.