What is the connection between artificial intelligence & machine learning?

Applications in these areas and they’re getting better over time super is futuristic sooji right now only exists in Hollywood movies and fiction novels and the fact one of the quotes that I really like about you know this whole debate about whether AI will take over the planet in their arguments would be delegated as slaves were given the Andrew ng one of the foremost thinkers on this area and he mentioned it that he worries about soup earlier in the same way that he should worry about overpopulation in Mars right.

It is that far Thornton right but let that not distract you from the application and the reality of AI that surrounds all of us today on that note let me take you one level ahead one level deeper right and let me try and explain to you three terms that are often used interchangeably and often quoted out of context right but one of the things that I thought I should try and do is also demystify some of these since we all right.

So we’ve spoken about a yeah that what does AI really is essentially you’re talking about developing algorithms computing systems and machines in such a way that they can understand context understand data and give output whether that is inside whether that is information any of the output that the system is supposed to give at a rate that is comparable to humans or at an accuracy rate its competitive with comparable to humans or slightly better than humans right so that is the broad new world of AI now there are two other associated terms one is machine learning and the other is deep learning first.

Let’s talk about machine learning so machine the science of machine learning’s it’s a complete field of computer science right so it’s an entire field of study and that is something that we all want you to appreciate right it’s a very vast and deep field the field of machine learning is all about creating algorithms using certain techniques in such a way that you can train these algorithms over the period of time as more data gets ingested right so the algorithms start learning and start becoming smarter thereby their output starts improving over a period of time and machine learning and AI are very iterative because machine learning is one way of implement in Ai

Mlis a set of techniques to implement AI right.

The connection between the two so obviously the question that may come up in your mind is that can you do AI without a machine gun and the answer the theoretical answer to that is yes you can write it you can implement it without machine learning but implement AI without machine learning the effort and time and investment that’s going to go in you know commune maintaining millions of lines of code and constantly updating it and you know deploying it is a lot of lot more than that is why it is not feasible which is why the only practical way of implementing AI is through machine learning right.

You really cannot advance an AI unless you know machine learning because that is the only way that hey I will get implemented in a scalable and sustainable fashion right because as I said what you’re trying but the systems that you’re building the engines that you’re building by using machine learning are in such a room okay that they improve beautiful times let your effort does not increase non linearly as the scale of data increases then it’s not sustainable right which is my machine learning is important to machine learning then further breaks down into multiple categories whether it is suitable earning unsupervised learning reinforcement learning all these are a bucket of technique is depending upon the problem you’re trying to solve but all of these are means to an end and the end is essentially implementing in a burrow di solution all right so what is important for you to understand is that machine learning is a way or a means of implementing AI.

What is deep learning is an alternate set of techniques that you will deploy at some point in time when you want to take the accuracy of your system even higher when machine learning models can only take it to a certain scale and then you deploy deep learning techniques to increase it even further and I’ll explain that with an example but the broad difference here is that deep learning the whole science and field of deep learning is all about creating algorithms of creating systems in such a way that they mimic the human brain right and obviously there is a lot of science behind that like when you think about the human brain you know that its capacity to analyze to consume information to give insights is practically limitless.

It can do all of this gathering the context that exists around it right so deep learning is when you are actually trying to build solutions right when you’re trying to build recruiting system in such a way that you can enhance what machine learning has already achieved and you’re trying to take it even further some for every problem that you face it is not as if you need to build a solution using Duplin but it really depends upon the level of accuracy and the complexity of the problem the level of accuracy that you want to achieve and the complexity of the problem that you’re trying to solve in today’s world there are a far there are far more applications of machine learning and that’s why there are so many jobs in machine learning and deep learning is something that is now beginning to kick in because for a lot of these organizations that are implementing machine learning solutions.

They’re one of their focus areas is also how do you enhance on top of that and that is when you know areas such as deep learning come into prominence let me go back to the example that I was explaining to you which is image recognition you know the deep freeze algorithm by Facebook right so image recognition is typically about the first you know examples or case studies or walkthroughs that you will come across when you begin your journey in learning machine learning it’s a classic machine learning problem so when Facebook.

Now if he came back about the journey that you know the machine learning engineers in the air engineers would have made at Facebook to bring the face to life obviously what is the problem that they’re trying to solve the problem that you’re trying to solve is that they have to build an algorithm that recognizes human beings based on pictures that they have right now typically when you try to solve such a problem it is not as if we immediately jump into one solution and tribal to start building it right.

You will first start building a model on a smaller data set that gives out a certain set of results in a certain at certain predictability right now go back in time and all the times when they were actually tagging people manually what you were doing was that you were training the algorithm the data you were saying that with this picture belongs to this friend of mine who’s by this name right and this other picture also belongs to this friend of mine who’s by the snake right.

What is driving this whole world towards an AI?

The first talk about what really is AI and you know what is driving this whole world towards an AI revolution as everybody keeps saying like what are the trends to it what does it mean for professions like to and also try and attempt at demystifying some of the terms or jargons that you may have heard where there are many of you may not be very clear where what the distinctions are so terms such as artificial intelligence machine learning deep playing data science and so on.

I’ll attempt to try and explain to you in very simple terms in terms of what – each of these mean and also give some practical examples and walkthroughs so that the distinction can be clear your mind the second half of this webinar shall focus on the program the postgraduate program and AI and ml audition intelligence and machine learning by bay TX and Stewart School of Business why is it such a unique program what really is its benefits you know what are you likely to learn what can be your learning outcomes and you know more features about program towards the end all right.

On that note let’s get started so I’m sure that you know you all have heard artificial intelligence this is a term right but let me ask you a question what do you really envision when you hear artificial intelligence what’s the first thought that comes to your mind and I’ve been doing this you know pop quizzes with various audiences across the past one year and you know here are some of the most common responses that I get and I’m sure that you know how many of you there would be the first image that comes to your mind for many of them and many of those people when they think about AI or intelligently minds immediately races towards some super-intelligent humanoid robot the kind that you see in sci-fi movies alright a crisis of the machines.

So forth another image that really comes up to all of us is the driverless car a self-driving car right again it’s been in the news if experiments are happening there pilot is happening so that that is a very distinct reality is not so far future over to the mold enhanced you know prospects and folks who’ve been reading about this a lot I’m sure alpha goo is an image that comes in the mind comes to mind we’re in you know for the first time a machine actually defeated a human consistently in a game of go but there is one common aspect to all these images right which is that there we are looking at a world where machines or computing systems.

If I may call them in more simplistic terms you know computing systems are actually making decisions like human beings – and that really is the crux of artificial intelligence right if we remove all the fluff around it if you remove all the jargons around it what AI really at the core of it means is how do you make machines behave and take positions like human beings with better accuracy and better predictability alright so that is what all systems are trying to achieve today as we speak and that is the world that we are moving into brought clean when you think of AI how will you think of artificial intelligence there are broadly three categories of you know that these fall into the first is what is known as narrow.

AI narrow AI is where you know you have developed an algorithm or your machine or computing system is extremely capable of doing one task and performing it at an accuracy like humans are better than humans right but it is only for that one task right so a very common example that you may all relate to here is you know image recognition so there was so one of the pioneers of one of the companies that have done really very well when it comes to image recognition based recognition is Facebook and I’m sure all of you if not most of you are Facebook users so you guys know what I’m talking about you know when you upload pictures on Facebook nowadays you know the Facebook algorithm doesn’t really ask you to tag them.

It can identify who these people are there was a time until about 12 months back when you had to actively tag people right all there were suggestions that were coming and nowadays for most cases that do not happen right so that is an AI algorithm that they built which they call as deep face and the deep faced accuracy of recognizing faces, by the way, is 97 percent it is currently higher than human beings have an average accuracy for 92% the deepest algorithm has a killed 50 97 percent so that is a classic case of narrow AI where that one algorithm right the only task it has to do is recognize faces obviously it’s easier said than done but that is one task that it has to do and that it performs at comparable or overhead better accuracy than human beings right.

The other broad category of AI is generally I eye is where there is also context it is not just a task but it’s about a series of adjacent tasks and now the machine can perform those tasks right to relay that to an example a perfect example would be any of your voice assistants you know whether it is Google assistant or whether it is City or whether it is Alexa right in all of these cases it is not as if you’re restricting your query you only do one aspect right it’s not as if you li ask Alex of whether to today right or you only ask Google assistant about you know what’s my schedule for the day right you are asking these systems these machines a bunch of questions a bunch of adjacent questions.

Where there is context and the algorithm is working and learning in such a way that it is it is aspiring to improve its accuracy and responses overtime right and for any of you who are active users of any of the voice assistance or chat boards you would see that over a period of time the accuracy improves over a period of time the system starts learning what you really mean right and starts giving you better this one so essentially in many ways you are training the machine and that is what he is all about but the machine is learning right to respond better so this since this is a classic example of generality I were the use case is not restricted to one specific task or one specific area but there are adjacent areas and there is a context around it.

The algorithm of the machine has to be trained enough to pick up that context and thereby give responses or give insights or answers that are more meaningful the third category of AI is super AI as the world calls it so the soup when you talk of super AI it basically you know you talk hang about the world that has built machines that are just much smarter than human beings and everything that humans – all right so imagine you know we like matrix coming to life, for example, right so it it is a world like that and obviously this today isn’t the realm of fiction whether it will happen and when it will happen is a different matter altogether and I don’t think that is what we want to talk about today but this is far from reality right now Natalie I am generally I am where a lot of work is happening a lot of evidence is already pulled out and you already have systems.

How can you relate Deep Learning with Artificial Intelligence?

How is it evolving what do we understand when we talk about some of these terms and yeah in today’s day and age the kind of use cases that are evolving for artificial intelligence and machine learning are evolving very rapidly so we’ll talk about that right.

I’ll give you a brief overview of what does it mean for all of us in terms of what kind of impact in terms of jobs creation in terms of the roles is it creating and then we’ll take I’ll take you through the program that we offer great learning very briefly towards the end hopefully everyone can hear me if not please drop a note and our team will work it out cool okay so so this is a question which I ask you know people.

I meet them for the first time and then you know when we’re talking about artificial intelligence is that what do they envision when they hear the word artificial intelligence right it’s a fairly broad term people are using it now MMM utilizing the storm for many things but what does it mean I’m usually the answers that I get is are let’s say the first chance that I get easily is in a major four machines which are super intelligent you know which kind of is similar to what we have been seeing in movies and like terminator etc a smart humanoid machines who can also think and make decisions like us.

So that’s one of the responses which I get late some of the responses which I get out of the driverless cars right and we have seen this technology evolve very rapidly in the past five-six years right it used to be something which you know people used to foresee would happen but the kind of progress it has made in the last three-four years has been tremendous right we have we are now seeing working prototypes and also production cars which are very autonomous right so in some and they’re only limited by regulations at the moment there are cars which can drive by themselves but it’s just that the regulations and some of the developed countries don’t allow a car to be fully autonomous.

That’s the reason they have you know the systems are built in where a human driver will have to be present and we’ll have to keep touching the steering wheel now and then to tell the Machine that it is to tell the car that a human driver is a present right but for all practical purposes we now have the technology we’re in a car can self-navigate then all of that so a wonderful example.

This is something that we but still it remains very popular we one of the things when you talk about the, is that we are our understanding of AI is kind of impacted by some of these popular themes right so if I need somebody who has a slightly more nuanced understanding of AI like the answer that I get is let’s say an example of alpha go right, so alpha Solo is a game which is mean practice for you know it’s a very complicated game and it has multiply it has like billions of combinations of moves which are possible.

Now we have an AI which has beaten a human player the human champion in the game of going right if you remember I think in the 90s there was an algorithm which beat which had beaten the chess players the leading chess players at that point of sign right, but there has been a sea change since then the kind of algorithm that was which you know had beaten the written in chess versus this algorithm which has now beaten players and go these two algorithms are very very different in chess the algorithm versus trying to predict the moves it was just trying to get to predict or think three or four moves ahead by looking at all permutations and combinations right.

It was a very different kind of algorithm with going what happens is and it was more algorithmic so when chess was it was more rule-based the chess algorithm right so it was human intelligence which was codified into a machine and then the machine had the power to just compute all the combination and looking more steps I had more moves ahead than a human player so it was a very different kind of algorithm going go the algorithms have changed completely Co is powered by deep learning is powered by artificial intelligence like what goes does is then Gotham here does what alpha does is it learn, so there are a lot of games which go into training this regarding.

So it learns from those games automatically there is no human intelligence which is provided to the silicon it will just play enough games and then understand from that what kind of patterns are emerging how should it play the game and from there then it learned and learned and then reached a point where it started beating you know the top players in this game, so this was this is a sea change in the way intelligence is defined I mean this also is is a good example to showcase how we have been defining intelligence right earlier it used to be something which we codify we take our intelligence and codify into a machine but now it is more about how we learn itself.

Now being codified hate again a good example a good analogy which I give here is you know how a human child learns right so some of you may have children right I also have a son right and he is only two years old but now and he started to speak now right but I never or we never you know went and told him about the grammar about how to form sentences it says that he observed from all the conversations which were happening in the household and then he learned automatically so he just looked for patterns and those patterns have given him the power to speak and that speech is going becoming better and better as we go along so similar is the algorithms that we have developed.

That’s the reason we talk about algorithms like neural networks right the whole neural network the term has come in because we have modeled algorithm in a way the human brain functions so so these are paver domes which are now very and that’s the kind of intelligence now which is developing notice when we say AI artificial intelligence these are the kind of algorithms that we are talking about okay.

Now let’s understand even further what people mean when we say AI and what are the stages of AI the first stage is narrow AI which is a machine that learns enough to do one specific task well enough right, so a good example is that say in let’s say recognition right so humans have been so one and the other thing which has developed is the cognitive ability for algorithms.

Earlier we used to talk about more task-oriented things and whatever we used to whatever was in the realm of cognitive understanding like for example looking at his photo and identifying which people are there in the photo that is a cognitive skill that we associate more with humans because it requires a lot of contexts it requires a lot of learning right but now, for example, Facebook has an algorithm called D to face which can look at a photo and INA fight which people are there in the photo so some of you who have been using Facebook for some years now if you remember if you go back to three years right you would remember that when we used to upload photos to Facebook to three years ago it would ask us to tag those photos with which people are presenting the photos right.