SIMPLEST Explanation of How Artificial Intelligence Works? No Jargon | What is AI? How AI works?

0:00 Should we be afraid of artificial intelligence? It is a question that keeps many of us up at night. Most people first heard about AI through 1990s movies like Terminator or The Matrix, where it was portrayed as a villain turning against humanity. And it was not just the movies. Renowned

0:18 figures like Steven Hawking have also warned about the possible dangers of AI rising against human beings. So back then it was completely natural to feel uneasy whenever we heard the term artificial intelligence. But for nearly 20 years after those movies, AI made little real world progress that could

0:38 cause serious concern. Things stayed relatively quiet. But today the situation has completely changed. AI is now part of our daily lives through AI cameras, Alexa, Siri, ChatGpt, Google Gemini, and many more tools. Even the YouTube algorithm that suggested this video to you works using AI. And it is

1:00 not just YouTube. On platforms like Facebook and Instagram, AI decides what you see on your feed. Another unsettling use of AI is in deep fake videos, which many of you may have already come across. In several countries, banks now use AI to approve loans. Insurance companies use it to calculate your

1:20 premium, and AI even plays a major role in stock market trading. In the medical field, AI is now being used for diagnosing diseases. In short, AI is being used in areas where serious decisions are being made. Yet despite all this, we have not seen AI behaving like the evil villain shown in those

1:42 science fiction movies. And that naturally leads to an important question. Is there a difference between the AI we saw in movies and the AI we are seeing today? What exactly is artificial intelligence? How does it work? How many types of AI are there? And could AI ever become a villain as

2:00 shown in the films? Let us explore the answers together in this video. Hi friends, welcome to a new video from science simplified for all. Most of us have been familiar with AI based personal assistants like Alexa, Siri and Google Assistant for some time now. But it was only after the arrival of Chat

2:22 GPT and Google Gemini that many people truly realized just how far artificial intelligence has evolved. Chad GPT seems to understand what we say in natural language and respond in a way that feels remarkably human. Because of this, many people unknowingly assume that chat GPT is a conscious being, something with

2:43 self-awareness. But we must always remember Chat GPT is just a software program specifically designed to mimic the way a human respond in conversation. It does not truly understand things the way a human does. To clearly grasp this idea, we first need a basic understanding of what artificial

3:01 intelligence actually means and how it works. This video will give a very simple explanation aimed at everyday viewers with no technical background or prior knowledge about AI. So if you are someone already familiar with the core concepts of AI, feel free to skip this part as it may seem like an

3:22 oversimplification. Let us begin with the term intelligence. The ability to learn new things, make logical decisions and solve problems. That is what we call intelligence in a human being. When a machine or computer begins to exhibit this kind of ability, we call it artificial intelligence. That

3:43 is the most basic definition of AI. But this definition alone does not give us the full picture. We already know that computers have been capable of doing mathematical calculations much faster than humans for decades. For example, if you ask a computer to multiply 1,230 by 2480, it will do it instantly, far faster than

4:07 any human ever could. In fact, a computer can perform millions of such calculations every second. In that sense, computers have always been ahead of us in solving arithmetic problems. But here is the key point. These calculations are all based on instructions that we the humans have

4:25 already given to the computer in advance. These instructions are what we call programs. A standard computer can only follow those predefined instructions exactly as given. If it encounters a situation slightly different from what we programmed it for, it will usually fail. Now, here is

4:44 the difference. When a computer is able to do something new, something we did not specifically teach it by learning from the data and recognizing patterns on its own, that is when we say the computer has artificial intelligence. Let us take an example. Suppose we create a program to identify birds and

5:02 upload it into a computer. Then we show the computer pictures of 10 different types of birds. A crow, an eagle, a sparrow, a parrot, and so on. We also tell the computer these are all birds. Now imagine we show the computer a new picture of a bird it has never seen before. If the computer is able to

5:24 recognize that the new image is still a bird even though we never showed it that specific example, that is a simple form of artificial intelligence. It has learned from the previous examples and applied that knowledge to something new. That is one of the key traits of AI. Of course, what today's artificial

5:44 intelligence is capable of goes far beyond this simple example. AI systems now handle much more complex and powerful tasks. But we are using this bird example here to make the core concept easier to understand. Now, let us take a closer look at how the artificial intelligence program in our

6:01 bird example actually works. We begin by defining the unique features that distinguish a bird from other objects. For example, we might say that birds usually have two legs, wings, feathers, and a beak. These are the features that we define in the program. Next, we assign a weightage to each of these

6:21 features. That means we decide how important each feature is in identifying a bird. For instance, two legs 20%, wings 30%, feathers 30%, beak 20%. Together that adds up to 100%. Now if we show the computer a picture of a crow which has all four features it would score 100% and the program would

6:47 confidently say yes this is a bird. But suppose we show it a picture of a penguin. A penguin does not have visible wings or feathers like other birds. Its wings look more like arms. So the program might only find 40% of the features it is looking for. Based on this low score, it may wrongly decide

7:08 this is not a bird. But we know that a penguin is a bird. So we correct the program and tell it that it made a mistake. When the program receives this correction, it begins to adjust the weights it had assigned earlier. It might reduce the importance of features like feathers and wings and increase the

7:26 weightage for features like a beacon two legs. It also learns that a perfect 100% match is not always necessary. Even if a creature matches only 80% of the key features, it could still be a bird. This program is written in such a way that it can adjust these weightages by itself based on feedback. As we continue to

7:47 show the program more and more images of different birds, it keeps refining its internal settings. Each time it makes a mistake, we correct it and the program learns from that error. In other words, it adjusts its internal parameters. Eventually, after seeing enough examples and making enough adjustments, the

8:07 program becomes capable of correctly identifying any bird, even one it has never seen before. This process is what we mean when we say an AI program is being trained. In reality, the working of an AI system is far more complex than what we just explained. Modern AI programs use many different parameters

8:28 and weight values and these are processed through multiple layers and stages before reaching a final decision. Still, we chose to present the concept in such a simple way for one important reason to help you understand three key points clearly. First, in the example we discussed, the AI does not actually

8:48 understand what a bird is. In fact, the concept of understanding or mind itself is quite vague. But even if we leave that aside, what a human understands when they see a bird is fundamentally different from what an AI sees. For the AI, a bird is just a group of numbers, weight values associated with different

9:08 features like a beak, feathers, and other such traits. That is all. Second, to train an AI, we need a massive amount of data. Even in the bird example, the AI needs to be shown thousands of bird images to learn what a bird looks like. And not just that, every time the AI makes a mistake, a human has to step in

9:30 and correct it, telling the AI that it was wrong. Without that feedback, learning will not happen. Third, even after all this effort, the AI only learns to answer one specific question. Is this a bird or not? If you want the AI to answer a different question like which bird it is, you need to write a

9:51 different program and train it separately. And if the image is not a bird at all and you want the AI to say whether it is a mammal or something else, that too requires a separate program and new training. In other words, each AI can only perform the specific task it was trained for. If you

10:09 want the AI to do a new job, it must be trained again, often from scratch. To put it simply, every specific task requires a specially trained AI system. Just because something is called artificial intelligence does not mean it can do everything. All these limitations we discussed so far explain why even

10:29 though science fiction movies talked about AI decades ago, the real development of artificial intelligence took much longer to happen. The rapid growth of AI that we see today became possible mainly because of two key factors. The first reason is the dramatic increase in computer processing

10:47 power. Back in the 1990s, computers simply did not have the capability to run AI programs. But over the last 20 years, the computational speed of computers has increased by several thousand times. This incredible improvement is what finally made it possible to run complex programs like

11:08 artificial intelligence on regular machines. The second major factor was the rise of social media. Let me give you an example. When you upload a photo of your pet dog to Facebook and tag it as dog, you're actually helping Facebook's artificial intelligence learn what a dog looks like. Every time

11:26 someone does this, the AI gets better at recognizing dogs, even when it sees a totally new photo of a dog later. With the explosion of social media, AI systems suddenly had access to huge amounts of data for training. Today, millions of people across the world upload photos every single day. Many of

11:46 those photos are used behind the scenes to train AI programs. The same is true for public messages, comments, and captions that we post. All of this content is used to help train artificial intelligence in understanding natural language. the way we humans actually speak. And this is important because the

12:06 way we speak is very different from what you find in books or dictionaries. Spoken language is filled with informal phrases, slang, and regional styles. Yet today, we have reached a point where AI can understand meaning even in informal speech. And that is largely thanks to the vast amount of training data AI has

12:27 received from language used on social media. By now you should have a general idea of how an artificial intelligence system is trained. But there are a few important things you need to know. When an AI is trained using millions of images and messages, the process is not done manually by humans. That would be

12:47 practically impossible. Instead, the training is handled automatically by powerful software systems designed for that purpose. And this leads to a major problem. By the time training is complete and the AI is released for public use, even the developers who created the system often have very

13:06 little idea of what exactly is happening inside the AI's decision-making process. This is because the training process changes the AI program so extensively that it becomes difficult to track how it arrives at its conclusions. Here lies the real issue. If such an AI makes a mistake, it is incredibly difficult to

13:27 figure out where it went wrong or why it made that mistake. This lack of clarity, often called the blackbox problem, is one of the biggest drawbacks of many modern AI systems. Let me give you a real world example. An AI system was trained to identify animals in pictures. But during testing, it started

13:47 mclassifying certain dog photos as wolves. At first, the reason for this behavior was unclear. Only after detailed investigation did researchers discover the actual cause. In the training data set, almost all wolf images had snow in the background. So, whenever the AI saw snow behind a dog,

14:07 it assumed the image must be of a wolf. In other words, it was giving more importance or weight to the snowy background than to the animal itself. This kind of issue happens because after training we no longer fully understand what the AI is focusing on internally. That is why it becomes so difficult to

14:27 trace and correct these kinds of errors. To address this problem, a new approach called transparent AI has been proposed. The idea is to make AI systems more understandable where we can see and interpret what is happening inside. However, transparent AI is still a developing concept and has not yet been

14:48 widely adopted in real world applications. There is one more crucial point to remember. The data used to train an AI must be completely accurate and unbiased. If the training data contains flaws or biases, those issues will reflect directly in the AI's behavior. Let me give you a real world

15:06 example. In one case, a company used an AI system to shortlist candidates for job interviews by analyzing job applications. But when the results came out, it was found that the AI had selected only male candidates. On investigation, the reason became clear. The data used to train the AI mostly

15:27 came from past hiring decisions. And in the past, the company had preferred hiring men for that specific job role. This bias in the historical data, even if unintentional, got passed on to the AI system. As a result, the AI learned to favor male candidates simply because that is what the past data showed. This

15:48 is a classic example of how bias in training data can lead to discrimination in AI decisions. And this can apply to other human biases as well, such as bias based on skin color, religion, or any number of other prejudices. That is why it is absolutely essential that the data used to train AI is carefully checked

16:09 for fairness and neutrality. Otherwise, our own biases will get transferred into the AI we build. And here is something we must never forget while using AI. When an AI makes a mistake, it has no idea that it made a mistake, nor does it feel any regret. That is because AI has no concept of understanding. It does not

16:31 know what is right or wrong. It simply follows the patterns it was trained on. So the responsibility to monitor AI behavior will always rest on us the humans. We must be the ones to watch to correct and to decide what is acceptable. Now let us look at the different types of artificial

16:49 intelligence used today and the kinds of jobs they are designed to do. The first major type is called natural language processing or NLP. This refers to AI systems that can understand and respond in human language, the way we actually speak. Examples you're already familiar with include Alexa, Siri, and Google

17:10 Assistant. But NLP is used in many other areas as well beyond just virtual assistants. The second type is generative AI. This is a kind of AI that can create new content, things that never existed before. For example, imagine an AI that has read thousands of novels. Now, if you ask it to write a

17:32 completely new novel, it can do that. Or if it has seen thousands of human faces, you can ask it to generate a picture of a face that does not belong to any real person and it will do that too. That is what generative AI does. Chat GPT and Google Gemini are examples of generative

17:52 AI. To be more specific, they are generative text AI, which means they create new text based on what they have learned from existing text data. Similarly, there are generative image AI, which can create new images from learned visual data. One such tool is Deli. For instance, you could ask it if

18:12 a famous painter from the past were alive today, what would their painting of a modern city look like? and the AI would generate an image based on that prompt. The third type is called computer vision AI. This kind of AI is used for image and face recognition. AI cameras which can identify people,

18:32 objects or license plates fall under this category. Beyond these, there are many other specialized types of AI available today. Robotic AI helps robots navigate and interact with the world. Speech recognizing AI converts spoken words into text. Explainable AI is designed to make AI decisions more

18:53 transparent. Planning and scheduling AI are used in logistics, project management, and more. In some cases, two or more types of AI work together to perform a more complex task. You might have seen deep fake videos where a famous actor's face is seamlessly swapped onto another person's body or a

19:13 politician appears to say something they never actually said. This is possible because two AI systems are working together in tandem. The first AI system is trained specifically to swap faces in a video. But if you look closely at those videos, you might notice something odd, some slight unnaturalenness that

19:34 makes you realize it is fake. Here comes the role of the second AI program. Its job is to detect flaws in the video. Anything that seems artificial or off. Once the flaws are identified, the first AI makes corrections and produces a new, improved version of the video. Then the second AI checks it again to see if any

19:55 new mistakes are still visible. This process continues in multiple rounds with both AIs competing. One trying to make the video more realistic, the other trying to detect any imperfections. Eventually, the outcome is a deep fake video so convincing that we can no longer tell it is fake. The deep fake

20:14 videos we have seen so far are not perfect. But today there are AI tools capable of generating even more realistic and convincing videos. This raises a disturbing truth. Such highly accurate deep fakes can potentially be used for fraud or other malicious purposes. So far we have been

20:32 classifying AI based on the kind of work it does. Image recognition, text generation, speech processing and so on. But there is another way to classify artificial intelligence based not on the task it performs but on its level of capability. This is where we come across the terms weak AI, strong AI and super

20:53 AI. And it is in this classification that the possibility of AI turning against humanity like in the movies enters the picture. The AI systems we have discussed so far are all designed to do one specific task. This kind of AI is known as weak AI or narrow AI. Almost all the AI that exists today falls into

21:17 this category. We humans on the other hand possess what is called general intelligence. We can learn a wide variety of things, adapt to different situations, make logical decisions, and solve many types of problems. If a computer ever develops that kind of broad capability, it would be called

21:36 general AI or artificial general intelligence. Sometimes simply called strong AI. But let us be clear, artificial general intelligence does not exist yet. It is still a theoretical concept. There are ongoing efforts to build such an AI, but no one knows how long it will actually take. Some experts

21:58 in the field who are very optimistic believe it might take around 20 years. Others say it could take at least 50 years or even more. As of today, general AI remains a future possibility, not a present reality. There is one more category of AI that is even more powerful than general AI. It is called

22:18 artificial super intelligence. This refers to a future AI that will be more intelligent than humans in every possible way. Right now, this kind of AI is purely speculative, but that does not mean it can never happen. Some believe that such an artificial intelligence might emerge by the end of this century,

22:38 but no one really knows how long it will take. The main concern is this. Once artificial super intelligence is created, it could potentially become smart enough to design even better versions of itself. And those versions could go on to create even more advanced versions and so on. If that happens, the

22:59 growth of AI would become exponential like a chain reaction. This kind of situation is what experts refer to as the technological singularity. It is this level of super intelligent AI that is often portrayed in movies as turning against humanity. And it is not just in fiction. Many respected individuals have

23:19 expressed concerns about this possibility. The late physicist Steven Hawking and Elon Musk, CEO of Tesla and Space X, have both openly warned about the potential dangers of uncontrolled AI. But there are also many experts who believe such fears are unfounded, at least for now. Their main argument is

23:38 this. Today's AI systems do not have consciousness or self-awareness. As we discussed earlier, when an AI identifies a bird, it does not truly understand what a bird is. It is simply processing numbers and patterns. It has no inner awareness or sense of meaning. In fact, even human consciousness is still a

23:59 mystery. We do not fully understand what consciousness is or how it arises in the brain. So, creating a conscious self-aware AI is far beyond our current capabilities. And without self-awareness, the idea of an AI deciding to take control or destroy humanity is not realistic. After all,

24:19 such desires for power and domination are very much human traits. Even if an AI becomes smarter than humans, it does not necessarily mean it will want to harm us. What we should actually be concerned about is not the AI itself, but the humans who might misuse it. AI is a very powerful tool and like any

24:39 powerful tool it can be used for good or for harm depending on who is using it. That is the real risk we must watch out for. In fact, some of this may already be happening. AI can be used to subtly influence the general public toward a particular political orientation or ideology. And with many of the AI

24:58 algorithms we interact with in daily life, achieving this kind of influence is relatively easy. Another common fear people have about AI is that it will take away jobs. But we have faced this situation many times in history. When electricity and machines were introduced, people feared job losses.

25:19 When computers came, there were similar fears. Yet each time humanity adapted. We found new ways to work and new types of jobs were created. Many experts believe the same will happen with AI. Artificial intelligence will not directly take away your job. But someone who learns to use artificial

25:39 intelligence effectively might out compete you in the same field. That is where we need to adapt to the changing world. We must learn how to work with AI, not against it. I hope this video helped you gain a clear and simple understanding of what artificial intelligence really is and what it is

25:55 not. If you found this video useful, give it a like and share it with someone who might find it interesting, too. And if you enjoy content that explains complex science in a way that anyone can understand, make sure to subscribe to the channel and of course tap the bell icon so you do not miss any of our

26:14 upcoming videos. We have some fascinating topics coming soon. Thank you.