AI Vs ML Vs DL for Beginners in Hindi — Transcript
Full transcript
- 0:00Hey guys, welcome to my YouTube channel
- 0:02. This is day two of 100 Days of
- 0:04Machine Learning. In this video, we're
- 0:07going to talk about a different topic:
- 0:11AI vs. ML vs. DL. Okay? So, I've always
- 0:13noticed that whenever beginners enter
- 0:16this field, this is the first curiosity
- 0:18they have. They ask, what's the
- 0:20difference? What is AI? What is machine
- 0:22learning? And what is DL? And if you
- 0:25Google it, if you Google AI vs. ML vs.
- 0:27DL, then mostly you'll get this picture
- 0:30where you'd see three circles, the
- 0:33outermost circle representing AI, and
- 0:35inside that, there's a circle
- 0:37representing ML, and inside that,
- 0:40there's a circle representing DL. Right
- 0:43? Now it's clear that the outermost set
- 0:46, the biggest thing, is AI. Within AI,
- 0:49a subset is machine learning, and
- 0:51within machine learning, a subset is
- 0:53deep learning, which is actually
- 0:55correct. But what I want to give you is
- 0:57a very good intuition as to why this is
- 1:00so. Okay? So let's discuss that. Let's
- 1:03first discuss AI. AI stands for
- 1:05artificial intelligence, or simply,
- 1:08infusing intelligence into machines.
- 1:11Right? And this idea has been around
- 1:13for a long time. Probably for 100 years
- 1:15. We humans have been wondering, what
- 1:18if we could make machines as
- 1:19intelligent as we are? Right? Now, if
- 1:22you think carefully, what is
- 1:23intelligence? Right? We humans are
- 1:26better than other biological species,
- 1:28we are advanced, we are evolved.
- 1:31Because of our intelligence. Right? But
- 1:33what exactly is intelligence? So, if
- 1:35you study and research intelligence a
- 1:37little, you'll understand it's a
- 1:39complex thing. And it's made up of
- 1:41multiple things, right? So, there's
- 1:43pattern recognition, which we use on a
- 1:45day-to-day basis. Be it coding or
- 1:47solving puzzles. These kinds of things
- 1:50are what you might call IQ. It also
- 1:52includes creativity and imagination. It
- 1:55also includes emotional intelligence.
- 1:57Right? So, our intelligence is a very
- 1:59complex thing, encompassing many other
- 2:02things. Right? But if you talk about AI
- 2:04, at this point, the AI we human beings
- 2:07are pursuing is a subset of actual
- 2:09intelligence. We're trying to create a
- 2:12general intelligence that behaves
- 2:14exactly like us. But it's actually very
- 2:16difficult. So, what we people are
- 2:18trying to do is we're trying to go a
- 2:20specific path, and that is pattern
- 2:22recognition. So, most of the AI stuff
- 2:24you'll see around here is mostly
- 2:26related to pattern recognition. We
- 2:28can't go for creativity and imagination
- 2:30at this point, and emotional
- 2:32intelligence at this point because
- 2:34that's hard to quantify. We don't know
- 2:36the true results. We don't even know
- 2:38what the equation of love is yet, or
- 2:40how to measure creativity. Right? So,
- 2:43for us, what's measurable, what's
- 2:44quantifiable, is pattern recognition.
- 2:47And that's why, at this point, whatever
- 2:49you see in AI is mostly related to
- 2:51pattern recognition. Okay? Now, let's
- 2:54talk about AI. Let's take a look at all
- 2:57the work that's been done so far. So,
- 2:58this idea started floating around
- 3:00around the 50s. At that time, Alan
- 3:02Turing and all those people started
- 3:03thinking about whether intelligence
- 3:04could be instilled in machines. And at
- 3:07that point, intelligent systems began
- 3:11to emerge. In fact, at that time, a new
- 3:14wave of intelligence emerged, a new
- 3:16wave of AI called symbolic AI. What
- 3:18happens there? What do you do in
- 3:20symbolic AI? You build a massive
- 3:22knowledge system using if-else
- 3:24conditions. Basically, you write down a
- 3:26logic for each condition. Right? And
- 3:29from that symbolic AI, expert systems
- 3:31began to emerge. For example, if you
- 3:34ever played chess against a computer as
- 3:36a child, that was actually an expert
- 3:38system where an expert was seated, all
- 3:41the knowledge was extracted from him,
- 3:43converted into code, and then we play
- 3:45chess against that code. Right? Now,
- 3:48what happens in expert systems? Expert
- 3:50systems have a knowledge base that we
- 3:52take from an expert, and then there's
- 3:54an influence engine, a program that
- 3:56makes decisions based on that knowledge
- 3:58base. Right? So in the 80s, everyone
- 4:00started believing that expert systems
- 4:02were the future of AI, and AI and
- 4:04expert systems were synonymous at that
- 4:06point in time. But later, people
- 4:08realized that expert systems have some
- 4:10flaws. What are those flaws? Expert
- 4:12systems work on very closed problems.
- 4:15Meaning, on very specific problems like
- 4:18chest detection or, I don't know, lung
- 4:20detection, for medical purposes. But if
- 4:23you give it a problem that is a little
- 4:25fuzzy, whose logic is a little fuzzy.
- 4:27It is called fuzzy logic, for example,
- 4:30I don't know, there is a dog in an
- 4:32image. There is a dog, it has to be
- 4:34detected. Now, this is fuzzy logic
- 4:36because there are so many breeds of
- 4:38dogs. And all of them have different
- 4:40features. So if you sit down to write
- 4:42if else, you can't write all the
- 4:45conditions so as to build an expert
- 4:47system which can actually classify dogs
- 4:50in an image. Right? So, for such
- 4:52problems where fuzzy logic was used,
- 4:54image classification was done, voice
- 4:56recognition was done, and apart from
- 4:58this, there were many other areas where
- 5:00this symbolic AI logic was not
- 5:02applicable. And that is why expert
- 5:04systems were left behind, and then came
- 5:05machine learning, which solved this
- 5:07problem. So, what we are discussing
- 5:09next is machine learning. So now let's
- 5:11talk about ML. Where does ML fit in? So
- 5:14if you have seen the last video, there
- 5:16I told you the definition of machine
- 5:17learning. Machine learning is a branch
- 5:20of computer science where you use
- 5:23statistical techniques to find patterns
- 5:25in your data, or more effectively, we
- 5:28can say that the technique of learning
- 5:31patterns in data is called machine
- 5:33learning. Right? The most important
- 5:36difference between machine learning and
- 5:38the symbolic AI that was previously
- 5:40used is that in machine learning, you
- 5:42never use explicit programming.
- 5:43Explicit programming means rule-based
- 5:45programming. You write logic for
- 5:47everything. Right? You don't do that in
- 5:49machine learning. Instead, what you do
- 5:51is you have data. Data has both inputs
- 5:54and outputs. You feed that data to your
- 5:56system, and your system's job is to
- 5:58derive rules or patterns. To derive
- 6:00mathematical patterns. So, when you get
- 6:03new input, now that you know the rules,
- 6:05you can produce a new output or
- 6:07prediction. Right? Now, why is this
- 6:09approach better than symbolic AI? This
- 6:12is very easy to understand. Where was
- 6:14symbolic AI failing? In fuzzy logic, in
- 6:17fuzzy logic-based problems. For example
- 6:19, in an image classification problem
- 6:21where you have to classify dogs. Now,
- 6:23you can't write rules to identify every
- 6:27dog breed. Right? But in machine
- 6:30learning, you don't have to write rules
- 6:31. Your system will generate the rules.
- 6:34All you have to do is provide the
- 6:35correct data. So now what I'll do is
- 6:37I'll provide my system with a lot of
- 6:40photographs of a lot of different dog
- 6:42breeds. Right? I'll show a lot of
- 6:44photographs. I'll tell it, "This one is
- 6:45a dog. This one is not a dog. This one
- 6:47is a dog. This one is not a dog." Now,
- 6:48what the system will do is it will
- 6:50automatically try to find the
- 6:52underlying representation of the data.
- 6:54This is called learning. It's trying to
- 6:56learn patterns, to recognize them, and
- 6:58that's what's called learning. And if
- 7:00you show it a lot of photos, eventually
- 7:03it will be ready to classify whether a
- 7:06given image has a dog in it or not.
- 7:09Right? So that's the main idea. It's a
- 7:11simple one. The idea is, why write
- 7:13rules? Show them data, and your system
- 7:16will automatically understand what's
- 7:17considered a dog and what's not. In a
- 7:21way, we work like this. As children,
- 7:25we're also taught, look, this is a dog,
- 7:25this isn't. Right? And that's how we
- 7:27learn. So, in a way, that learning is
- 7:29inspired, in part, by our own evolution
- 7:32of intelligence. Right? So this was
- 7:34machine learning. And guess what? The
- 7:36simple idea that rather than writing
- 7:38rules for everything, you can simply
- 7:40provide data, and the rules will be
- 7:42generated automatically has changed the
- 7:44industry. In the last 30 years, machine
- 7:46learning may have come in the 90s, it's
- 7:48only 2020. It has been just 30 years,
- 7:50and this thing has been revolutionary.
- 7:52Especially in the last 15, 10 years,
- 7:54when data is so abundantly available,
- 7:57and good hardware is available, machine
- 7:59learning has taken off and is flying
- 8:01high. Right? Many people used to say,
- 8:03in fact, that machine learning is
- 8:05nothing but statistics. But to be
- 8:07honest, it's not purely statistics. Yes
- 8:09, we are using statistical tools. We
- 8:11are using mathematical tools. But there
- 8:13isn't much mathematics involved. There
- 8:15is mathematics only to a very limited
- 8:16extent. It's highly practical
- 8:18engineering-based. Okay? So, it's
- 8:21inspired by statistics, but not exactly
- 8:23derived from it. Okay? So, I guess you
- 8:25understand machine learning. Now comes
- 8:27the question: what is deep learning? If
- 8:30machine learning could do everything so
- 8:32well, why is deep learning needed?
- 8:34Right? Let's discuss that. So, let's
- 8:37discuss deep learning in two parts.
- 8:39First, let's discuss what deep learning
- 8:41is, and then let's discuss why deep
- 8:43learning is needed. Okay? So, deep
- 8:46learning is very recent. Like, already,
- 8:48work on it has been going on for a long
- 8:50time. But it wasn't until 2012 that we
- 8:52reached the peak of deep learning,
- 8:54which is why we're hearing so much
- 8:56about it now. Okay? So, deep learning
- 8:59is basically a class of algorithms.
- 9:02First, it's machine learning. It's not
- 9:04different. Meaning, never have the
- 9:06notion in your mind that deep learning
- 9:07is something else. You use the same
- 9:09procedures, the same process flow, in
- 9:11deep learning as you do in machine
- 9:12learning. You give the same data, train
- 9:15the algorithm, and reduce the error
- 9:17during training. Eventually, your model
- 9:20is trained. Then, you give it new data
- 9:22and make predictions on it. The exact
- 9:23same process occurs in machine learning
- 9:25. The exact same process occurs in deep
- 9:26learning. So, deep learning is
- 9:27essentially machine learning. It's just
- 9:29that the algorithms you use here are
- 9:31slightly different. And in a way, you
- 9:34can say that it's inspired by biology.
- 9:36Neurons are inspired by our biology. I
- 9:38wouldn't say that it works completely
- 9:40like our brains. Deep learning is a
- 9:42different concept. Deep learning is
- 9:44inspired by our biology. But we can't
- 9:47say that deep learning algorithms work
- 9:49exactly like our brains. To be honest,
- 9:52we don't yet understand how our brains
- 9:53work. Deep learning is a mathematical
- 9:56model. It's math-based. It's just that
- 9:58its core unit, the smallest unit, is
- 10:00called a neuron perceptron. It's based
- 10:02on biological neurons. So that's why
- 10:04there's this biological connection.
- 10:06Many people are confused that deep
- 10:08learning works the way the brain does.
- 10:10It doesn't. Okay? So that's about deep
- 10:12learning. So why did deep learning come
- 10:15into use? Or why? Deep learning is so
- 10:17popular today. It's such a buzzword in
- 10:19the industry. The reason is that there
- 10:21are certain things that machine
- 10:23learning can't do, or you can say,
- 10:25can't do as well, and that's where deep
- 10:27learning shines. Okay? Let me tell you
- 10:29where. So, a big problem with machine
- 10:31learning is that in machine learning,
- 10:33you have to provide features. You have
- 10:35to tell the input you provide what the
- 10:37input will contain. For example, if you
- 10:41want to classify between dogs and cats
- 10:43in an image, saying, "This one contains
- 10:45a dog, this one contains a cat," you'll
- 10:48have to specify the features that
- 10:50belong to dogs. And you'll have to
- 10:51specify the features that belong to
- 10:53cats. The ears are pointed. It has to
- 10:55be a dog. Right? It has a mustache. It
- 10:59has to be a cat. Right? The tail is
- 11:00curled up a lot. It has to be a dog.
- 11:03Right? This is how you'll have to
- 11:04define features. If you're in the
- 11:06machine learning space, right? But if
- 11:08you're in the deep learning space, what
- 11:10happens in deep learning is that your
- 11:12system automatically creates or detects
- 11:14these features. You don't have to
- 11:17create them, right? And if you're
- 11:19working with fuzzy logic-based problems
- 11:21where you actually don't know what
- 11:23features might exist, deep learning can
- 11:25be very useful there. I'll give you one
- 11:28more example. Like, we were trying to
- 11:30build a system in our startup where we
- 11:32could look at resumes and decide
- 11:34whether someone would be placed or not.
- 11:36We were trying to build a system like
- 11:38this. Now, if you work in machine
- 11:40learning, you'll have to create
- 11:42features that you'll feed to the
- 11:44machine learning model. You'll have to
- 11:46create those features, and you'll only
- 11:48be able to do that if you have
- 11:50knowledge about the data. Right?
- 11:51Someone who knows a lot. Probably,
- 11:53someone working in a TMP cell will know
- 11:55that if they have more than two
- 11:57certifications, there's a chance of
- 11:59getting placed. If they have a
- 12:01percentage between 90 and 95%, there's
- 12:03a chance of getting placed. If there
- 12:05are backlogs, there's no chance of
- 12:07getting placed. All of these are
- 12:09individual features, right? But if
- 12:10you're working in the deep learning
- 12:12space, what you have to do is simply
- 12:14provide the raw text, and it will
- 12:15automatically create features. So
- 12:17that's one of the big differences
- 12:19between machine learning and deep
- 12:21learning. In deep learning, you don't
- 12:23have to be concerned about the features
- 12:25. What deep learning will do is
- 12:27automatically extract those features.
- 12:30In machine learning, you have to create
- 12:31those features. And that's a difficult
- 12:34task: understanding which features will
- 12:35be more accurate. Right? Second, in
- 12:38deep learning, as you add new layers of
- 12:44neurons, the more your classification
- 12:47or regression task becomes better. Or,
- 12:50more simply, as you add layers of
- 12:54neurons, the system's efficiency and
- 12:57prediction power improves. Because with
- 13:01each layer, you can uncover deeper and
- 13:04deeper what's hidden in your data. If
- 13:06you look at this image, you'll
- 13:08understand that gradually, if we're
- 13:10trying to classify something, say, a
- 13:12digit, the first layer will tell you
- 13:14the edges. The second layer will
- 13:16combine the digits into bits and pieces
- 13:18, and the next layer will combine those
- 13:19digits to tell you which digit it is.
- 13:21Right? So that's how deep learning
- 13:23works. Right? And this is one of the
- 13:25main reasons why deep learning was used
- 13:28where machine learning wasn't
- 13:29performing as well. And deep learning
- 13:32gave good results. The second reason is
- 13:34that deep learning is something else.
- 13:36It's because of the algorithms that the
- 13:40more data you feed it, the more its
- 13:42performance improves. Right? If you
- 13:45look at this graph, you'll understand
- 13:47that machine learning models improve
- 13:50their accuracy up to a point with
- 13:51increasing data, but after that, they
- 13:54stabilize. This isn't the case with
- 13:55machine learning and deep learning. If
- 13:57you keep feeding it more data, your
- 13:59performance will continue to improve.
- 14:02Right? So, for these two or three
- 14:04reasons, deep learning has found a
- 14:07niche where it's used a lot, like image
- 14:10classification tasks, basically
- 14:12image-related tasks, and text-related
- 14:15tasks. And these are some of the
- 14:17applications where deep learning is
- 14:20better than machine learning. But then
- 14:23you might ask, why use machine learning
- 14:24if deep learning is so much better? The
- 14:26answer is that where you need a needle,
- 14:29you wouldn't use a sword, right? So,
- 14:31deep learning is a sword. It's a
- 14:32powerful thing, it's a powerful
- 14:34technology, but you wouldn't use it
- 14:36everywhere. Machine learning is really
- 14:38superior in some areas. You use deep
- 14:40learning on big data. You get good
- 14:43results. But with small data, you don't
- 14:45have a lot of data available. So you
- 14:47won't use deep learning because machine
- 14:49learning will give you better results
- 14:50on less data. And guess what? 90%of the
- 14:53world's data is small data. Big data is
- 14:56all you use, like Facebook and WhatsApp
- 14:58. So, you might think everyone has a
- 15:00lot of data. But that's not true. If
- 15:01you ever work in industry, you'll find
- 15:03that banking companies, insurance
- 15:05companies, and sports analytics
- 15:07companies don't have a lot of data. So,
- 15:09machine learning is still very useful
- 15:11in such scenarios. Okay? So, that's the
- 15:13complete point. Let me summarize it
- 15:15once. The biggest term is AI. We're all
- 15:19after AI. Now, the most efficient
- 15:21system for creating AI is machine
- 15:23learning. Right? And even in machine
- 15:25learning, a method has emerged to solve
- 15:27some very complex problems, which we
- 15:30call deep learning. And in today's date
- 15:32both Deep Learning and Machine Learning
- 15:34are very famous. Now you never know how
- 15:36to build a general AI, after 10 years
- 15:37something completely out of the box new
- 15:39may come. But if at this point you are
- 15:42planning to learn to build AIs, Machine
- 15:44Learning and Deep Learning is the right
- 15:46path. Okay? So I hope you understood
- 15:48this entire video guys. If so please
- 15:50consider subscribing. I will try to put
- 15:53a new video every day. This is our goal
- 15:56100 days of Machine Learning. So or
- 15:58that's it. Thanks for watching.
About this transcript
This page contains the full transcript of AI Vs ML Vs DL for Beginners in Hindi by CampusX, generated from the public captions YouTube serves with the video. The transcript has 2,855 words across 437 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.