 Good evening everyone. Thanks for joining. Happy to see you all here. My name is Rithvik. I'm currently working at LinkedIn. You can know more about me and my work at the given portfolio link here, which is Rithvik12.github.io. Before we start, I would like to give you all a minute to think about AI. What comes to our mind when you think about it? If you all are currently thinking about it, so was it smart, intelligent, or just a well-trained agent using data? Let's start from the very beginning of AI when the term was itself coined. The whole idea of AI was based on the cognitive abilities of the brain. Most of the AI and ML frameworks works either on training or data provided to them. All algorithms revolves around the idea of learning by an AI algorithm rather than cognitive and intellectual growth. Now just try to imagine if an AI can listen to a language and learn it, if it can read a book and reason it. Do not just memorize the knowledge, but also have an ability to apply that knowledge just as a human would do. How great that would be. If you think about it, it takes years for a person to learn about a subject and then apply that in research and development. This can be done simply by having an AI which can reason and brainstorm with the person the same way our research partner will. We know so little about this universe, how it was made, why things are in a certain way. Human lifetime is so small to explore this vast universe. By the time someone is near to something productive from their research, usually their time is up. That's sad to hear, but that is true. Very few scientists were able to do something extraordinary in their life. And that is also because of their peers or ancestors or earlier scientists who made it possible. With their work and with their learnings, they were able to do and continue on their research. Very, very few will be there who were able to start something and compete it in their whole lifetime. Otherwise, it takes centuries for some theory to be proved. Think if we had an AI who can read all the books or research papers ever written. If we can pass on all the knowledge to an AI in just few minutes or few days or even few months will work. Rather than a person who is studying and reading about all the knowledge from their research paper and books for years and years and trying to understand how things work. In that we will only have to ask the right questions to the AI and we can get many answers in very short duration of time. The closest we have ever been to human intelligence in AI is in the form of artificial intelligence, artificial neural networks. Though there are big differences in that approach too. Let's talk about human brain alphabet. Human brain consists of different parts, mainly the four lobes. The frontal lobe, which is associated with reasoning, problem solving, planning, movement, emotions. The perietal lobe is associated with recognition, orientation, perception of stimuli. The occipital lobe, which is mainly responsible for visual processing. Temporal lobe, which is there for memory, speech perception and recognition. If you think about it closely, information about the world is gathered through senses such as vision, sound, smell, touch or taste. Whereas when we talk about AI, in neural net, the neurons are just numbers in the code, typically ranging from 0 to 1. The connection between the neurons also have numbers associated with them and those are called weights. These weights tell us how much the information from one layer matters for the next layer. As simple as that, the value of neurons and weights or the connections are simply the free parameters of the network. By training the network, we want to find those values of the parameters that minimize a certain function called loss function. So it's really an optimization problem that neural nets solve. Think about when a baby is born, what do we expect from them? I guess nobody from us will be expecting them to be like this. So they don't need books or us to teach them how to breathe, cry, eat or drink. There are some things that they get from their genes as part of heredity. The moment they are born, they are able to breathe, cry, move some of their parts and they can use all their vital organs which are essential for living. So there are things that we don't need to teach them and there are things that we need to teach them. The moment a baby is born, they start using their vital organs as normal as just an adult will do. Few organs will take some time to develop but there isn't any knowledge or anything to pass on to use those organs. But when it comes to an AI, we solely depend on data. Think of telling a child about a dog. We can show the photos of different dogs or we can just take the child near a dog. A child needs to see only a couple of times to grasp main features of a dog. And the next time they see a dog, they know what it is and how to behave or act around it. Whereas an AI needs millions of examples of dog photos as data. Without it, it can never understand what a dog is. To an AI, everything works as an object. It tries to grasp main features from the images and then tries to depict from what it learned with those millions of photos. There could be times when those images consist of backgrounds. For assumption, an AI might classify a wolf standing in snow as a dog because the data given to the AI for training consisted of dog images standing in snow. So the AI thinks of snow as an important feature of a dog and predicts anything closer to a dog such as wolf or jackals in a snow as a dog only. That might look like a confusion but to an AI that was what we wanted it to do. Let's look at this. This part is controlled by an AI which came up with the design of legs and then figured out how to use them to get past these obstacles. But when this experiment was set up at Google Brain, they had to set it up with very, very strict limits on how big the AI was allowed to make the legs. You can see here that few of the agents or boards are having longer lengths, few are having very short lengths. Because without that, the AI actually did what we asked it to do here but in an astonishing way of its own. The danger of AI is not that it is going to rebel against us, it's that it is going to do exactly what we asked it to do. Most of humanity is afraid of advanced AI as it might be the cause of humanity's self-destruction. As soon as it can think and reason like us, in a way this could be true but only if we let it. Because we just saw that the current AI does not do things on its own but tries to do exactly what we asked it to do. Developing an AI which can have cognitive capabilities is like growing a child. It depends on what we feed to it and how we do it. It ain't gonna just pop up from somewhere and start doing bad things for no reason. Let's talk about approach. The fastest man-made object which we can think of is NASA's Parker Solar Probe. It can travel as fast as 140,000 meters per second, which is 535,000 kilometers per hour. It can do that with the help of Sun's gravity. Without that, it won't be able to attain such a high speed. That's blindingly fast, right? Yet, it is the only 0.05% of the speed of light. No matter how hard we try, it is gonna take us very long time to reach near the speed of light with the same approach of engineering and science. The same goes for the true AI which we seek. By getting better data or lots of data, we are not going to get what we want. We will have to think about totally different approaches. One of the totally different approaches that we came to was artificial neural network when we tried to think about human intelligence, how it works, and try to know more about the brain. While improving the current best approaches that we have, we also need to look for different approaches. With that, we might be able to attain what we seek for. I know that a lot of people might already be working on it and some day, somewhere, someone will do it for sure. So with that, have a good day, everyone.