Overview of AI ML for Geodata

Namaskar. Good afternoon, everyone, and I welcome all of you in this 144th course, ISRO distance learning course, and the topic of the course is AI ML for geodata analysis. The course is beginning today on... August 19th and it will be ending on August 23rd. This is a one week course. In totality, there will be five sessions in this course. And I'm happy to inform everyone that we have received immense response from the users for this particular course. And for this course, we have more than 550,000 registrations because of which I request all of you for your cooperation. so that everyone gets benefited by this course. So first of all, the participants who have registered through the Nodal Center, I request all the coordinators of the Nodal Center to encourage your registered participants to attend the program from the common point or common login ID. I request all the coordinators, please discourage the individual attendance through the eClass portal. because we are having a large number of registrations for this course. In case you are unable to log in, then you can ask your participants, the students to join the live session through IARS YouTube channel. The address of the URL of the channel you all already have, and it is also getting flashed in IARS website also. The participants who are... One more thing, one means for the participants who have registered through the Nodal Center, the coordinators are requested to update the attendance of the participants later on through their login credentials. Then the second set of participants are the one who have registered as an individual. So all the individual participants are advised to log in. to join the live session through the YouTube link. The link has already been shared with you and also it is being flashed on IAR's website. Please take the link from there. The individual participants are advised to appear in the daily quiz anytime after the session in ISRO LMS through the registration credentials. The quiz will be open from five o'clock five 5 30 after the ending of the session uh till the next day and the quiz entire quiz will be open till saturday so you are all requested to please appear uh in the daily quiz because the certificates will be issued certificates of the individual participants will be issued based on your participation in the quiz so um as i have already told you that this is us you five session course it's a one week course in the first session will be on introduction to ai machine learning and distance and deep learning which will be taken by dr hina pandey the second session which will be tomorrow it will be on methods of machine learning the supervised unsupervised semi-supervised and reinforcement learning it will be taken up by me I'm Dr. Poonam Seth Tiwari. The third session, which will be on 21st of August, will be on deep learning concepts and the common deep learning algorithms will be discussed in that particular session. That also I'll be taking. On 22nd of August, we'll have a session on machine learning through Google Earth Engine, which Dr. Kamal Pandey will be taking up. And lastly, on Friday, we have a session on Python for Machine and Deep Learning Models, which Mr. Ravi Bhandari will be taking up. So now I welcome Dr. Hina Pandey for this particular session, and I wish all the participants a very fruitful learning experience. I also request cooperation from all the participants and coordinators during this program. Thank you. I now request Dr. Hina Pandey to start the session. Thank you, Dr. Poonam. A very warm welcome to all the participants in this first session of the course on AI ML for Geospatial Data Analysis. As already told to you all by Dr. Poonam, that we have had an overwhelming response for this course. We are very glad to have you all and we are very glad about the interest that you have shown in this program. I hope all of you are able to benefit from this and you have a great learning experience ahead. So in this first session of the five-day program, we will start with the basics of artificial intelligence. I'll walk you through the outline of what artificial intelligence means. what are its advantages, what could be its limitations, how we should ethically use this technique for the benefit of humankind, and what would be its potential applications. So, in today's modern lifestyle, we have been exposed to this technique of artificial intelligence in everyday life in several ways. So when we talk about artificial intelligence, sorry for the interruption. So artificial intelligence would be a branch of science, to be precise computer science, that would use models which are based on computational principles and it would be trained to perform tasks which previously required or would be done through human intelligence only. So basically it is a method which involves using several disciplines of science. computer science, data analytics, some hardware and software engineering concepts and so on. club them with biological concepts like neuroscience, philosophy, psychology, and come out with a system which would let the machines behave intelligently, or they could mimic based on certain rules that are fed to them, or training data as we call them, which is fed to them, they would be able to mimic human behavior and perform tasks which In the previous days, we expected only human beings to be able to do such things. So now coming to the technical description or definition, it would be a field of science concerned with building computers and machines. As I said already, that we would be coming up with a tool which is using computers or machines. And these machines are trained to reason, learn and act. or perform activities in such a way that would normally require human intelligence, or it would involve data whose scale would exceed. Now when they function these tools or these artificially intelligent machines, when they start analyzing data and giving us results, the scale of this data or the scale at which they can work exceeds what humans can analyze by far. Now we can then say that it would be artificial intelligence or AI, as we know it, would be a broad field that would encompass many different disciplines, right? We've talked about that. So they would club the biological sciences with the engineering sciences, the mathematical sciences, and together produce for us a tool which would be able to do several things for us at a very fast pace, looking at data which is vast in nature, which human beings would take maybe eons to process and so on. So it would be a set of technologies that are primarily based on machine learning and deep learning. Now these are two terms which you will constantly keep coming across in the next five days and later when you wish to you know build up on the knowledge that you've gained in this program. So machine learning and deep learning are two sub sets of AI or artificial intelligence, which are used extensively for this data analytics, predictions and forecasts, object categorization, applications like natural language processing, giving you recommendations, then intelligent data retrieval and many, many more such applications. Now, these AI systems learn and improve through exposure to vast amounts of data. So, the larger the amount of data... that we feed to them to make them learn the concept for which they are supposed to give a solution. Then they identify patterns based on these data and establish relationships that sometimes human beings also may miss out on. So their logical capability depends upon, or their capability to give us answers depends upon the training data or the amount of data that is fed to them to make them learn about the solution. Now this learning process involves algorithms which are sets of rules or instructions that guide the AI's analysis and decision making capability. Now as we see in this figure, artificial intelligence is the umbrella term and if we put it simply we would be able to say that it would be the ability of a machine to imitate human intelligence. Then as I said there are two subsets. machine learning and deep learning. In machine learning, the algorithms are trained on labeled or unlabeled data. Labeled data would mean that we are classifying it, and unlabeled data would be, as the name suggests, it would not have these classified fire labels on it to make predictions or categorize. The categories would not be there, so the categorizing would be done by the algorithm. So this data would make predictions or categorize information into classes. Deep learning utilizes something called artificial neural networks with multiple layers. So these layers, there are many, many layers to train this network. That is why it is called deep, deep learning algorithm. So these layers enable it to process information and thereby mimicking or imitating the structure and function of the human brain. as you are aware in the biological system the human body is made up of neurons which are the basic units which pass messages from the brain to the other parts of all the other organs of the body, helping it to perform daily functions or for our reactions and so on. So when an attempt is made to mimic this or to imitate this in as an artificial set of network, we call it the artificial neural network, when there are many such neural networks connected to each other, passing on their information and learning backwards and forwards. We call this a deep learning network. So these are algorithms that mimic human brain to incorporate intelligence into the machine. Now, through continuous learning and adaptation, the AI systems become increasingly adept or used to performing specific tasks. which can range from recognizing images to translating languages and beyond many such applications. We will discuss those applications which are possible through artificial intelligence in just a little while. So the core goal of artificial intelligence or creating an artificially intelligent machine is to emulate human intelligence by machines and the closer we go to this the better. decision making the machine is able to do. Now this entire process of emulating the human intelligence would involve tasks like reasoning, that is analyzing information and drawing logical conclusions from it, then learning, that is acquiring new knowledge and skills from the data that we already have to be able to do problem solving, then identifying and then solving the problems in a goal-oriented way, which will involve the learning which we've had in the previous step. Then once we are done with this, the final step would be to be able to give a decision for the solution, which will be on the way to the solution. So decision-making, that is to evaluate options and to make choices which are based on the available information. The aim is to make intelligent choices. and be able to choose the best solution and go forward with that. Now, this artificial intelligence or machines based on artificial intelligence are based on a core set of concepts and technologies that would enable machines to perform tasks that typically require human intelligence, as we've been discussing. So, the sub... sets of this would be machine learning, which we've just talked about. This is basically the backbone of the AI systems where algorithms learn from data without being explicitly programmed. It involves training an algorithm on a dataset and allowing it to improve over time and make predictions or decision based on new data. The algorithm is provided a set of data, it learns from that, and then gives us a solution. When more data is supplied to it, it improves and makes better predictions and data by learning from the more the data it has, better learning process will happen. Then we have neural networks. These are inspired by the human brain and these are networks of algorithms, many layers stacked together. So layers of algorithms or networks of algorithms imitate the way the neurons interact with each other. Now this allows the computers to recognize patterns and solve common problems in the fields of AI, machine learning and deep learning. Deep learning is a subset of machine learning. Deep learning uses complex neural networks with many layers, therefore it is called deep. So it learns from a set of several connected neural networks or layers of neural networks to analyze various factors of the data or various parameters of the data. This is instrumental in tasks like image analysis and speech recognition. Then there is another branch called the natural language processing NLP. This involves programming computers to process and analyze large amounts of natural language data. enabling interactions between computers and humans using natural languages, the languages that we are used to conversing in. Then robotics. While often associated with AI, robotics merges AI concepts with physical components to create machines capable of performing a variety of tasks, which could range from assembly line systems in manufacturing to complex surgeries and so on. So we could use AI-enabled robots or bots to do several tasks which humans otherwise would not be able to do or if they are able to do it. they would not be able to do it in that many numbers or they would not be able to do it with that precision with which the robot or the bot would be able to do enhanced with AI software. Then we have something called cognitive computing. This AI approach mimics or imitates the human brain processes to solve complex problems which say for example could often involve pattern recognition, then natural language processing that we discussed earlier, and data mining. Data mining from large amounts of data. People are able to locate the data of their choice or the data that they need for a particular problem-solving activity. Then we have something called expert systems. These are artificially intelligent systems that emulate the decision-making ability of a human expert. applying reasoning capabilities to reach conclusions. So the expert system is designed for a particular concept. So we could be using an expert system, say, for medical problems in the medical field. So the system would be designed that the knowledge related to the medical science would be fed into it. Algorithms would be related to that and different reasoning would take place. place to reach conclusions or solutions related to that particular field. Then if we are talking about the geospatial field, then we would be looking at data images, other kind of the ancillary data, all fed into the expert system along with the rules, rule sets, and then the reasoning would be applied and we would be getting solutions to the problems that we have at hand. Now each of these concepts helps to build systems that can automate, enhance and sometimes they could also outperform human capabilities in specific tasks. Like I said, when we have a huge amount of data, big data so to speak, then the analytical capabilities of the computer systems combined with the reasoning and learning capabilities of the human intelligence. combined together they would outperform singular capabilities and give us better decision-making abilities. So this figure would summarize what we've been talking about. We could do predictive analysis, we could convert text to speech and vice versa, speech to text, image recognition, machine vision. language processing so you could classify translate and extract data we have expert system as we've just talked about then planning and optimization and then we have robotics which then strengthened or enabled with ai software or algorithms would do certain things which humans would not be able to do as well as these computer systems and algorithms together would be able to do for us Now, let us just go through how the artificial intelligence field has evolved. What is the history? And we already know that it is not a new concept. And the development of artificial intelligence started in the early 50s, 1950s. And today it has come a long, long way from where it started. Now, the timeline that we have with us is that in about 1943. The idea of a model that would imitate brain cells was introduced. Then in 1950, the well-known scientist Alan Turing developed a test to see if machines could think like humans. And the first computer based on neural networks was built, artificial neural networks. Then in 1956, the term artificial intelligence was called by John McCarthy at the Dartmouth. conference and in 1963 the first AI research lab is set up at Stanford University. The first chatbot was created in 1966 called ELISA. The first productive expert system was developed at Stanford, and they were called Dendral and Mycin. This was done in 1969. Then there came a period, or a decade almost, where interest in AI dropped. And this led to a period that we know as AI winter. Not much. Research was done in this field during this period from about 1972 to 1980 and therefore it is called AI winter. Then in 1980 the research in this field of artificial intelligence picked up again with the development of successful business applications. Now to enable computers to learn from data, machine learning techniques like neural networks and genetic algorithms were created. So the interest in this field was renewed and people started working on algorithms which would you know build new techniques in this arena. In 1997 IBM created its supercomputer Deep Blue. We all of us are aware of this and this machine defeated the then chess champion Gary Cosgrove. In the 2000s Google makes brick tools in speech recognition and this was due to advances in machine learning, natural language processing and computer vision. All of them put together and we are now very familiar with speech recognition software and we several of us use it in everyday working also. Then the decade from 2010 to 2019 computers were now capable of doing tasks like speech and image recognition with previously unheard of levels of accuracy. And this was, all this was owed to the deep learning techniques like convolutional neural networks and recurrent neural networks. Now, today's discussion will not explain these terms in detail. This will be done by Dr. Poonam in her next two sessions, where she will be giving you an overview of machine learning, what are its concepts. what are the algorithms that are used and how they are used and what are their advantages. Then she'll be moving on to deep learning. She will take up convolution neural networks and recurrent neural networks and explain what the two algorithms do specifically in the field of geospatial data analysis. Then we come to 2020 to about the present day. So we now have all several applications which are AI-enabled, self-driving cars, virtual assistants, we use them for several applications, medical diagnostics. This is a field which uses this technique heavily, and also drug development. These are just a few of the many present uses of AI. Several companies unveiled their conversational AIs like the Googlebot, Microsoft Bing, the ChatGPT by OpenAI, and so on. So we are all used to today's field is very, very dynamic and we have unheard of technologies available and we are all using these in everyday functioning as well. So as you see here, we have a little more details of from right from 1943 onwards to the present day and where we have seen that you have. realistic humanoid robots capable of displaying human expressions and they interact with human beings and the chart GPT of course where which is capable of processing data and both images and text and produce complex and go through complex tasks and give us solutions. We are also used to virtual assistants like Alexa and Siri and so on, where we give instructions through voice instructions and several things like playing music, updating your shopping list and so on and so forth are done for us through these AI-enabled virtual assistants. Now, let us go and see what are the types of artificial intelligence. capabilities available. So the classification is based on capabilities and functionalities. So based on capabilities we can say we have three types. It's been divided into three sections, narrow AI, general AI and super AI. Then based on functionalities, we could have four such types, reactive machines, limited memory, self-awareness and theory of mind. We will just discuss these in brief. So this is categorized. on four kinds based on the kinds and level of difficulties of the tasks the system is capable of performing. So reactive machines are the most basic kinds of AI. This is called the reactive technology, which is built just to respond to inputs and it has no memory or capacity for learning from previous events. It works only on the data that is provided to it in the present scenario. So whatever data it has, it works on that. So we see spam filters, chat box, recommendation engines, say from Spotify or Netflix and so on. Then limited memory. This has the capacity to learn from past data and make decisions based on such data. So speech recognition software and recommendation engines are again two such examples. For example, generative AI tools like chat box, self-drive. cars, virtual assistants and so on, as they have only limited memory, right? And the capacity to learn from these is based on this data. Then the third kind would be the theory of mind. Now, this understands the need of other intelligent entities also. So, it is this kind of AI should be able to communicate socially and comprehend ideas and emotions, much like the human beings do. So, the theory of mind describes a machine's capacity to perceive and anticipate the thoughts and feelings of other people. This is how we human beings react, right? We look at a person and what kind of emotions they are displaying, and then we change or act accordingly. So right now, it is still very early for this kind of artificial intelligence, and it is still in the research phase and has... been in the development phase, so it has not much developed yet, but work is going on and research people are researching in this area so that social interaction etc. in human-like, more human-like intelligence is incorporated into the machines. Then the fourth kind is self-awareness, where artificially intelligent machines that have the capacity to acquire a feeling of self and consciousness. So then these are termed or referred to as self-aware AI. So with the present technology again this is a purely theoretical idea that is yet not feasible. So the aim is to have to evolve these machines so that they have human-like intelligence and they are self-aware, aware a kind of consciousness. inside of them so that they can react accordingly. But again this is right now a theoretical idea and does not have any available applications. Then if we talk about artificial intelligence machines which are categorized on the basis of their capabilities, then we have ANI that is narrow intelligence, AGI which is general intelligence, intelligence and super intelligent AI which is ASI. So for example, ANI would be machine learning which specializes in one area and solves one problem. Machine intelligence or the second stage that is general intelligence, this refers to a computer which is smart enough and it acts like a comes across as a human when you converse with it or when you interact with it. Then the third one which is super intelligent machines or super intelligent artificial machines, then here we talk about machine consciousness which means that an intellect that is smarter than the best human brains in practically every field has been developed. So let us see a comparative analysis between these three kinds. A and I is also known as weak AI. Its counterpart, general intelligence, is called strong AI or human-level AI. And superintelligence, ASI, represents a level of intelligence significantly more advanced than even the most brilliant human minds put together. And AI refers to AI systems which are designed and trained to perform specific or narrow range of tasks. So several such systems are known today and are being used for. several applications. The ADI refers to systems, artificial intelligence systems that can understand, learn and apply knowledge across a board to a range of tasks at a level which is comparable to human intelligence. Then the superintelligence ASI refers to systems that surpass or are beyond human intelligence and capabilities in all aspects including creativity, general wisdom, and problem solving. Now, AGI systems operate under limited predefined set of functions and cannot generalize knowledge beyond their specialized domain. So, they work for particular domains, they are made for specialized domains. The AGI systems are characterized by generalization capabilities, which allows them to perform any intellectual task that a human being can do. So they can do several tasks, they can be trained to do many tasks at the same time. Now potential characteristics of the ASI systems would be that they have superior problem solving abilities, enhanced creativity and innovations and much advanced decision making skills and strategic planning. Now examples of narrow intelligent intelligence-based artificial machines would be voice assistants like Siri and Alexa, recommendation algorithms on platforms like Netflix, Amazon, Spotify, and so on. They could also be chatbots and some image recognition systems. We have been using these examples and the AGI systems are able to mimic human cognitive functions, including reasoning, problem solving, and understanding complex concepts. Now, they are also able to adapt to new and unfamiliar tasks without specific training for each new challenge. So, they are able to, you know, learn from the data that they have, and they are able to understand and then learn and then apply that knowledge across several such broad range of tasks. Now, ANI is one of the most common forms of AI which is in use today. AGI still remains largely theoretical but the entire gamut of functions has not been achieved as yet. The concept of ASI or superintelligence is still highly speculated and it remains a topic of theoretical exploration still. Now, the characteristics of the first category that is ANI is that it is highly specialized in one area. It is unable to perform tasks outside its training scope and it operates under predefined parameters and tools. The implications of the AGI, that is the second category, could be once the theoretical concepts have been achieved. could be that it would revolutionize industries by automating complex decision-making processes, which are unheard of till now. However, it also raises some questions about its ethical use and safety concerns regarding control, ethical use, and impact on employment. So there are still some implications where people are, human beings are worried at present that how will these affect once they are totally functional, how would these affect these fields? Similarly, the implications of the super-intelligent machines are that they could lead to unprecedented advancements in science, medicine, technology, and many other fields. However, again, like the AGI, general intelligence, when it reaches that level, it would pose... significant concerns about ethical, philosophical and existential risks, including concerns about alignment with human values. potential unintended consequences because after all it would be a machine and would it be able to you know learn as much or have the consciousness of a human being and be able to interact and react in those ways or would there be certain risks in terms of the ethics and would it be able to align with human values and so on so it's still unknown people are exploring it but there are concerns which need to be addressed. Now let us see what the differences between artificial intelligence and human intelligence would be. What would be the basic differences? The first point to discuss would be the information processing. Now AI processes data using algorithms and makes objective data-driven decisions. There are no ifs and buts. Based on the data, it makes objective decisions. Human beings combine emotional and logical reasonings. Their decisions are also, you know, influenced by their emotions and some logic that they have. So cognitive reasoning, adding depth to their decision-making. The efficiency in learning. So the second point is how efficient are. the two systems in their learning process. To learn efficiently, AI systems require large data sets to perform that learning, to acquire that knowledge, whereas human beings can learn from a few examples through abstract thinking and reasoning. So the amount of data that is required to train artificially intelligent systems or machines is very very high. A large amount of data is needed for it to be able to make proper decisions. Whereas when compared to this, human beings require very, very little, less amount of data. Adaptability, how adaptable are the two systems? Human beings easily generalize knowledge across tasks. Whereas it has been observed that AI often struggle which with the jobs or tasks that it wasn't specifically designed for. So even if there is a small change in the process and it has to make a decision, it falters sometimes or it gives us a not so correct answer or maybe an altogether wrong decision or solution when it is used for something which there is a change in, sudden change in. in the problem that is provided to you and if it wasn't specifically designed or trained for it. Whereas human beings are not so, they are easily able to generalize their learning or knowledge across different tasks. Now, emotional understanding, these two systems differ in this emotional concept. So, humans naturally interpret emotions. whether somebody is sad or happy or depressed and so on and so forth. So these are emotions that we are used to interpreting and this helps us in the complex decision making process. So we change our decisions or solutions based on these emotional inputs. Artificial intelligence, because it is a machine, it has to perform based on the rule set that is provided to it, lacks emotional intelligence totally at present and struggles with social nuances. So, as you are aware, as human beings we interact or react socially and in different social contexts, we change our behavior and behave accordingly. Whereas this is not possible for the artificial machines or artificial AI-based systems. The fifth very important element of difference between AI and human intelligence or human beings would be ethical reasoning. Human beings have moral reasoning, they have moral backgrounds and are also capable of ethical reasoning. So, they also have ethical concepts which they conform to and which are very, very important for them. Now, AI follows the ethical framework which is set by its creator. So if we put in those rules or the creator or the person who has generated these rulesets puts them as input for it to learn then it follows though that framework with potentially perpetuating biases otherwise it has no such qualms about being ethical or morally correct and so on. It will only go on processing the data, learning from the data and giving you solutions or giving you answers. of performing tasks according to the data that has been fed to it. So currently the level of difference between AI and human based intelligence is quite a lot. People are striving, the researchers and engineers are striving to achieve a balance between both these systems and trying to make machines. more like humans and perform tasks with, they are trying to incorporate emotional understanding and reasoning and by providing them a set of big making them adaptable like human beings, the machine, the AI based systems. Now let us see what are the advantages and the limitations of these artificial intelligence machines or systems. Now let us go through the advantages first. The advantages are that there are less chances of errors. AI can perform tasks faster with greater accuracy, more accuracy than human beings. This leads to improved performance and productivity in many areas. Now, there will always be a possibility of inaccuracies in jobs or tasks, which requires precision when people are involved. because they sometimes make subjective decisions. Increased safety. Now using AI-powered machines like robots etc can help in saving human lives and it can be used at places which are high risk for example nuclear plants, coal mines etc or collecting scientific data from inaccessible areas like glaciers, volcanic craters and so on. The AI systems help us in repeat performing repetitive tasks so there are several repetitive tasks that we do in daily life and they take a lot of our time so these tasks can be assigned to machines for example which would then help us in reducing time as well as cost for example industries which have an assembly line production could have automated could automate several tasks and then they would save on time and cost Now these AI enabled systems are available 24-7 unlike human beings who require to be rested to be able to perform to the fullest capability. AI machines can run 24-7, 365 without getting tired and it can run non-stop and provide continuous service and support. So this is a very big advantage. Then what are the limitations? it is very important to know the limitations in the current scenario. So these systems are expensive, they cost a lot because AI system development and implementation is complex and it involves a cost. This needs a lot of resources and knowledge. So currently they are not so easily available, they are still rather expensive. Then the biggest limitation is that they can't think what we human beings know as thinking. This is a certain limitation. It can only perform tasks and operations that have already been defined. So it cannot think for itself and perform additional tasks. So if it encounters a difficulty in between or the situation is changed midway that it is performing the task, it might not be able to align itself and give us a proper solution. is not capable of thinking like human beings. Also, it has no emotions, so it is not emotionally intelligent. Machines can perform tasks better and faster than human beings and can also do it with much more accuracy and precision. However, the emotional connection or the emotional connect, which is very important while working with each other, the machines lack this at this moment. Now another very big question that or point which is people are concerned with is that the incorporation of AI-based machines freely into the system can may lead to unemployment. So because AI would be able to do the job of several human beings put together and faster and quicker and with much more precision, people might not prefer employing human beings, especially for repetitive tasks. tasks which require long hours on duty and so on. So this has definitely made the AI technique has made life easier but it has also taken up jobs that were previously done by human beings and this can lead to large-scale unemployment problems especially for areas or countries where we have a large amount of human population this could be a very very big problem in the future. Now after having gone through all these concepts or what is artificial intelligence, what are its subsets and how it is advantageous and what are its limitations, let us see why it is important. Now the basic points that can be said is that it improves efficiency, it solves complex problems and it makes better decisions in certain cases of course. So increased efficiency and productivity because it automates repetitive tasks, frees up human time and resources for more strategic endeavors. Imagine AI powered robots handling assembly lines in factories or chatbots managing customer service inquiries. This will allow human employees to focus on innovation or complex problem solving activities. Then if there is enhanced decision making. then AI would be able to analyze vast amounts of data to identify patterns and trends that human beings might miss altogether. Now this would allow for data-driven decision making in different fields, especially finance, healthcare, marketing. For instance, AI could analyze financial data to predict market trends. It is already doing that. It predicts the trends in the stock market, or it could go through patient medical data to suggest personalized treatment plan. People are using it for such applications now. Then for innovation and progress, it could accelerate scientific discovery and technological advancements. AI powered research tools can analyze complex scientific data. So we could go through big data and make informed decisions, simulate experiments through AI tools and identify promising areas for further exploration. Then it would also be able to give an improved quality of life. So it has the potential to revolutionize various sectors which would lead to better lifestyle, a better quality of life for certain people. For example, if we are using self-driving cars, it could improve transportation safety. AI-powered prosthetics could enhance mobility for those. who have disabilities in their limbs and so on. So it could be a life-changing experience for them. Then it could also address global challenges. So it could be a powerful tool for tackling challenges like climate change, resource management. It could optimize energy use, enable us to smartly use the available energy and sustainably use it, predict weather patterns. analyzed environmental data to support sustainable practices, which is the call of the day or the need of the hour. These are all global challenges, and AI-enabled technology or systems would greatly help us in tackling these challenges effectively. Now, in a nutshell, let us see what it is that makes this technology so useful. It provides us or offers us several critical benefits, which makes it an excellent tool. The first one being automation. The second one being enhancement. That is, it enhances the products and services effectively by improving experiences for end users and delivering better product recommendations. Then it has the capability of analysis and accuracy. So the analysis that it makes is much faster and more accurate. than what the rate at which human beings would be able to do it and it can also use its ability to interpret data to give us better decisions because it will be able to go through more high amount of data, huge amounts of data in a small amount of time and give us decisions based on that. Now, another very, very important point is to go through ethical considerations and challenges in the AI field. One that the AI field, the AI enabled machines or systems would have. they would handle biases and would they be fair. So the bias and fairness would be defined as the systematic and unfair discrimination that is often a reflection of the prejudices that exists in its training data design. Now because it has does not have the capability of thinking or cannot make logical decisions so these systems are biased based on the input data or any prejudices would be reflected. in the training data or the way that the training data or the design of the system has been done. And this would affect its fairness or equity. Now, the challenges are the first one being in data quality of the AI systems, the input data, that is the training data. If this data contains biases, the AI output... is most likely to contain or carry these biases. Then the algorithm design. Algorithms may also contain implicit biases introduced during the design process by the developers. So the design of the algorithm is again of utmost importance. And then the impact. The biased AI can lead to unfair treatments across various sectors. Suppose we are using an AI decision-making tool for criminal justice. or as a hiring tool or maybe for lending impact. This could impact individuals based on race, gender and other characteristics if the input data and the algorithm design is biased. So very, very important to consider the bias and the fairness aspects. Then there would be privacy issues. So the AI systems often require vast amounts of data to function effectively. Now, which can include sensitive personal information, for example, in the medical field. So ensuring the privacy of this data is a significant ethical concern. Now, challenges here would be data collection. So collecting large data sets could lead to overreach into personal privacy. And this has to be properly managed. And the aspect of personal privacy has to be taken care of. This can be done by... data storage and access. So how it is stored and who accesses it are crucial in maintaining privacy. Then surveillance. So AI technologies like facial recognition, behavior prediction could lead to increased surveillance, raising concerns about individual privacy and autonomy. So these are, again, very sensitive issues and need to be dealt with as such very, very carefully. The third point of concern is job displacement. Many people are discussing this, and this is a problem that we could be left with. So AI and automation can perform tasks more efficiently than humans in many, many industries. And this is leading to fears of widespread job displacement or job loss. So the challenges here are economic impact. So while it would increase productivity, it would also pose a risk to jobs in specific sectors, especially those in manufacturing, transportation. and maybe administrative support and so on. Now, skill gaps. So there is a growing need for retaining and reskilling workers, provide them with skills which would help them deal with this and prepare them for the jobs of the future, where human insight and interaction with AI systems will become very critical. Then social inequality. Now, job displacement would widen the economic gap. between those who have the skills to work with AI systems and those who do not. So these are again very very critical points and need to be addressed and taken care of. Then the safety and security, as these systems become more prevalent they become more in use ensuring that safety and security against malfunctions, misuse, hacking is of paramount importance. So system failures need to be seen. Sometimes they can be very, very catastrophic, for example, in areas like autonomous vehicles or healthcare. Now if that system has been hacked into, there would be accidents on the road. health care systems if they are hacked or if they don't if they malfunction then you can imagine what would be the scenario then if somebody takes it into their mind for malicious use use it for harmful purposes such as developing autonomous weapons or cyber attacks and so on again it is unimaginable what might happen then security measures have to be taken so protecting ai systems from being hacked and used against the interests of users requires, again, very, very robust security protocols. Now, I will quickly take you through how we can measure the artificial intelligence system, how close is it to acting like a human. So there were a few tests which were designed to see if the AI system is behaving like a human or is it not. So there are four such tests, one which is the Turing test. tests whether the system is acting humanly. Then the second one is the rational agent approach. So the rational agent approach decides whether the system is acting rationally. The third is the cognitive modeling approach and this decides whether it is thinking humanly or not the system. And the law of thought approach is a system where the rational thinking of the system is analyzed. Now the Turing test, the basis of this test is that the artificial intelligence entity should be able to hold a conversation with a human interrogator. So as you see here in the figure, the human interrogator is having a conversation with a human responder and with a computer. So ideally, the human agent should not be able to conclude whether they are talking to artificial intelligence. So then the level at which it reaches decides how closely it is able to mimic the human intelligence. Then the co-operative modeling approach. So here it has to test introspection, psychological experiments by observing human behavior and brain imaging. MRI, etc. medical techniques are used to observe how the brain functions in different scenarios and replicates it through a code. Then the law of thought gives us a set of rules that will help think which can be used to program artificial intelligence and rational agents use a strategy where the AI tries to make the best decision based on its current situation rather than following a fixed set of rules. rules. Now we come to the last section where we will quickly take you through the applications or potential fields where AI is being used. It is being used in a wide area right from healthcare to education to automobile to finance surveillance in the social media and entertainment. We are very used to using these space exploration, gaming, robotics, agriculture, e-commerce, you name it. and AI has pervaded that field. So use cases could be from speech recognition where automatically one can convert spoken speech into written text and vice versa. Image recognition, it is used to identify and categorize various aspects of image data. It is very, very widely used in the geospatial field, as you will see later in the different sessions tomorrow. the forthcoming sessions. Translation, you can translate written or spoken words from one language into another. Very widely used in predictive modeling, mine the data. Data mining can be done to forecast specific outcomes with a high degree of granularity. Analytics, one can find patterns and relationships in data for example to be used in business intelligence. Cyber security, so autonomously scanned networks for cyber attacks and threats and so on. So these are just a few. Then for business intelligence, you can use it for data analysis, which would help you in giving you insights and recommendations for decision making. You can use it in healthcare for disease diagnosis, treatment development, and giving personalized care for the patients. AI in education. Here we can use this technique to personalize learning, improve the student engagement. and automate administrative tasks so that human beings are able to engage in more productive jobs. Then in the finance field we can do risk and fraud detection, personalized recommendations can be made for products and services, then document processing can be done very quickly through AI enabled software. tools. Manufacturing, we've been discussing this in the previous examples as well, that it would improve efficiency, productivity and quality. So very, very widely it has been started to be used in the manufacturing field, especially in the assembly line systems where repetitive tasks are to be performed. Then for other additional applications could be AI use in retail sector where it can be used to personalize shopping experience, recommend products and manage inventory. Transportation, again, self-driving cars, improve traffic management, flights running on autopilot and so on. Energy usage, we can use AI tools to improve energy efficiency and also predict the demand for the energy. In government sectors where to improve public safety, crime detection. provide several citizen-centric services and so on. Also, in the entertainment field, it can be used to recommend music, movies, and so on. So, online streaming companies are able to use this to provide suggestions to their customers based on their previous interests or preferences. Now, I will only give you a very short overview of how AI is being used. for the geospatial arena and what is the big picture. So AI, machine learning, deep learning are all being used, specifically machine learning and deep learning. And when clubbed with artificial intelligence, a term is coined called geo-AI. So we can use GIS, that is geographic information system, imagery data or spatial data, club it with... artificial intelligence can make informed decisions. So big data can be analyzed very quickly. Geographic data can be analyzed using these when combined with spatial principles. So the steps would be for geospatial analysis when we use artificial intelligence would be, one would be the data collection from various sources. So we could collect data from satellite-based images, earth observation sensors, that is from laser sensors for 3D data, UAV-based sensors and so on, GPS, and then use it for real-time monitoring, simulation of scenarios for multi-objective optimization and smart decision support is enabled, and then the scenarios are assessed, managed, and plan designs and implementation is done. stakeholders are engaged and web-based geo-visualization and early warning systems are generated which are used for monitoring, evaluating and management plan adjustments. So this is how geospatial analysis can be done. All these will be dealt with in detail in the next four sessions. Some practical demonstrations will be given on how to use available resources like the Google Earth Engine which is a vast repository of these kind of geospatial data on the web. It's a cloud-based platform. It gives you both data and the tools to process it. So all this will be discussed in the next section. Now the general workflow here would be data acquisition and preparation. So if we talk about image data or any other kind of data, so image data and label data is labeled. for specific locations of the feature that we want to identify. For example, if we want to identify buildings, then we need the image of the target area, that is the building, and the label of the vector outline of that building, and then we generate training data in the corresponding format. Then the next step would be building the model building and management. So you construct the model and manage it. To train the neural network model based on the training data samples generated on the previous data. preparation process. So the data acquisition and preparation here we prepare the data which is then input. to train the neural network based on the samples, different kinds of features that we want to identify. Now, iteratively we evaluate the training model through the validation data set and the test data set to achieve the actual application accuracy and precision requirements. So this is how this is this is done at the preparation step and when we build the model we iteratively evaluate it when we input the training data by splitting it into training into test data and validation data. All these again I stress these will be explained how it is done in detail in the next two sessions. Then we come to the last step that is the inference. So the trained model is used to predict or infer the testing samples and comprises of a similar forward pass as training to predict the values. Then at the last step, it applies the learning that it has acquired in the previous two steps and gives us a solution or the predicted values. So the general framework would be something like this. You input data, if we are working with images, and then perform the network learns. and it gives us outputs labels either classification object recognition and other understanding so image use cases when we are dealing with images could be image classification object detection semantic segmentation or instance segmentation now these are terms that you will be discussing in the next four sessions just to give you an example of how geospatial data can be used say for identifying crime hotspots, generating risk maps, yes, and using for crime analytics by using machine learning models and overlaying it with hotspot zones and so on. The second example deals with identifying types of roofs. So the identification of built up area is done and then buildings are identified and in those buildings the second step is to identify features like trees roads buildings and then in the building data set the type of rows is classified all based on a training data set so model training is done here and then then sample tiles or build chips of different types of routes is given as input and then it searches the system searches and gives us two types of out that are available in the system. Road identification can be done, different types of roads, et cetera, based on AI training. Again, tree classification has been done here. So here it is used using two kinds of data, one which is image data and the other which is height data. So the trees are also classified based on their height. What are the kinds of parent? This just gives you an idea of what are the kinds of training inputs you can give. so height difference, change rate, intensity, number of returns, anisotropy, planimetry, and so on. This can be done with data like laser data which gives us 3D point clouds and all these can be used as input parameters for training the model. Now I've overshot my time but I hope you have had a fruitful learning experience and I have tried to give you an overview of what the general artificial intelligence systems look like, what they are, how they are defined, what are their pros, what are the points that we need to be aware of, what are the limitations, and how they can be used for different applications. Then the last three or four slides we discussed about specific application in the geospatial field where we can use geospatial data. club it with AI decision-making tools and perform. With this, we come to the end of this session. Thank you very much. I'll be happy to take up questions if there are any. Otherwise, we can do the query questions and then we can send you answers later on. has posted that as seen in some generative AIs, hallucinations are observed in some complex scenarios. Can a similar false positive negative be observed in GOI? Well, there is a large chance of observing such false positives and negatives in GOI as well. As I said, the response of the AI system will be biased in terms of the training data. So if we ensure that we have comprehensive training data which is as free from biases as we can make it, then the output is likely to be unbiased and such false positives or negatives can be minimized. Do we have to use Google Colab or ArcGIS? You mean in this course or in otherwise, whenever you are using it? You can use either. RTIS has certain tools catering to specific applications. So you can use that for those specific applications, whereas you can use tools like Google Colab, et cetera, and several other tools to generate your own model. Now, in this five-day program that we have, If you're asking specifically about that, then I don't think we will be having time to be able to do things practically and guide you to do those hands-on. Though there is a session on Google Earth Engine, but they will only be demonstrating to you how that tool can be used. However, as you have asked, Google Colab, ArcGIS, and several more tools are available. You can use all those. Okay, Akash, this is a very, very interesting question and I am sure people are working on it. Can we use biological neurons and somehow embed them to a biotech so that we can train the neurons? Something like Neuralink that is happening today. There are, you know, we might yet be unaware, but various sections or the technological people are in. They are extensively researching these kinds of applications. Of course, we have to keep in mind the ethical part of the considerations that we need to do along with this. We might launch ourselves into a totally unpredictable domain if we do that, but I'm sure it would be possible. And there are people researching these. No concrete. solutions have been come up or a functional system has been made but yes people are doing it Okay, how does ISRO integrate AI and ML techniques in geodata analysis for real-world applications? There are several applications and many, many solutions underway. I would suggest that you wait until tomorrow to be able to see examples of these applications. So one uses it in climate change prediction. They are using it in... say, land use, simple applications like land use, land cover detection, and so on, chain detection, and so on, to climate change applications and several other geospatial data analysis. Tomorrow and day after, and in the Google Earth Engine class, you will be exposed to all these examples and applications. What possible hallucinations can AI generate while working on geodata analysis? Well, what possible biases etc. would be dependent on the training data. So, as I said, if our training data is flawed or if it is biased to a particular point, then that is how the bias would be How AI helps to find risky places or accidental places? Well, I think you are wanting to ask how we can use it in the place where there is a risk and so on. So suppose there's been a forest fire and some rescue operations need to be done, or some data is to be collected from such areas, or maybe scientific data is to be collected from glaciers, medicines, etc. have to be dropped. All these can be done through a robot. deep sea data collection, deep space data collection, where it is nearly impossible for humans to survive. So all these areas can be done through AI. ANI is widely used as compared to AGI. Well, because AGI and the super intelligence phase is still under development, the machines have not yet been able or the systems have yet not been able to achieve the level of intelligence or the decision-making capabilities and work is still on. Can we train our model to identify space patterns related to stars and pre-planet constellations? Yes, definitely we can do that depending upon the data that we input. I think we can stop now. Any other questions that you may have can be posted to the distance learning team platform and we can answer them collectively from there. I hope you had a fruitful learning experience and an enjoyable session. Thank you very much. We meet again tomorrow at four o'clock in the evening. Thank you. Thank you. Thank you. Thank you. Thank you.

This is a one week course. In totality, there will be five sessions in this course. And I'm happy to inform everyone that we have received immense response from the users for this particular course. And for this course, we have more than 550,000 registrations because of which I request all of you for your cooperation.

so that everyone gets benefited by this course. So first of all, the participants who have registered through the Nodal Center, I request all the coordinators of the Nodal Center to encourage your registered participants to attend the program from the common point or common login ID. I request all the coordinators, please discourage the individual attendance through the eClass portal. because we are having a large number of registrations for this course.

In case you are unable to log in, then you can ask your participants, the students to join the live session through IARS YouTube channel. The address of the URL of the channel you all already have, and it is also getting flashed in IARS website also. The participants who are... One more thing, one means for the participants who have registered through the Nodal Center, the coordinators are requested to update the attendance of the participants later on through their login credentials. Then the second set of participants are the one who have registered as an individual.

So all the individual participants are advised to log in. to join the live session through the YouTube link. The link has already been shared with you and also it is being flashed on IAR's website.

Please take the link from there. The individual participants are advised to appear in the daily quiz anytime after the session in ISRO LMS through the registration credentials. The quiz will be open from five o'clock five 5 30 after the ending of the session uh till the next day and the quiz entire quiz will be open till saturday so you are all requested to please appear uh in the daily quiz because the certificates will be issued certificates of the individual participants will be issued based on your participation in the quiz so um as i have already told you that this is us you five session course it's a one week course in the first session will be on introduction to ai machine learning and distance and deep learning which will be taken by dr hina pandey the second session which will be tomorrow it will be on methods of machine learning the supervised unsupervised semi-supervised and reinforcement learning it will be taken up by me I'm Dr. Poonam Seth Tiwari.

The third session, which will be on 21st of August, will be on deep learning concepts and the common deep learning algorithms will be discussed in that particular session. That also I'll be taking. On 22nd of August, we'll have a session on machine learning through Google Earth Engine, which Dr. Kamal Pandey will be taking up. And lastly, on Friday, we have a session on Python for Machine and Deep Learning Models, which Mr. Ravi Bhandari will be taking up. So now I welcome Dr. Hina Pandey for this particular session, and I wish all the participants a very fruitful learning experience.

I also request cooperation from all the participants and coordinators during this program. Thank you. I now request Dr. Hina Pandey to start the session. Thank you, Dr. Poonam.

A very warm welcome to all the participants in this first session of the course on AI ML for Geospatial Data Analysis. As already told to you all by Dr. Poonam, that we have had an overwhelming response for this course. We are very glad to have you all and we are very glad about the interest that you have shown in this program. I hope all of you are able to benefit from this and you have a great learning experience ahead.

So in this first session of the five-day program, we will start with the basics of artificial intelligence. I'll walk you through the outline of what artificial intelligence means. what are its advantages, what could be its limitations, how we should ethically use this technique for the benefit of humankind, and what would be its potential applications. So, in today's modern lifestyle, we have been exposed to this technique of artificial intelligence in everyday life in several ways.

So when we talk about artificial intelligence, sorry for the interruption. So artificial intelligence would be a branch of science, to be precise computer science, that would use models which are based on computational principles and it would be trained to perform tasks which previously required or would be done through human intelligence only. So basically it is a method which involves using several disciplines of science. computer science, data analytics, some hardware and software engineering concepts and so on. club them with biological concepts like neuroscience, philosophy, psychology, and come out with a system which would let the machines behave intelligently, or they could mimic based on certain rules that are fed to them, or training data as we call them, which is fed to them, they would be able to mimic human behavior and perform tasks which In the previous days, we expected only human beings to be able to do such things.

So now coming to the technical description or definition, it would be a field of science concerned with building computers and machines. As I said already, that we would be coming up with a tool which is using computers or machines. And these machines are trained to reason, learn and act. or perform activities in such a way that would normally require human intelligence, or it would involve data whose scale would exceed.

Now when they function these tools or these artificially intelligent machines, when they start analyzing data and giving us results, the scale of this data or the scale at which they can work exceeds what humans can analyze by far. Now we can then say that it would be artificial intelligence or AI, as we know it, would be a broad field that would encompass many different disciplines, right? We've talked about that. So they would club the biological sciences with the engineering sciences, the mathematical sciences, and together produce for us a tool which would be able to do several things for us at a very fast pace, looking at data which is vast in nature, which human beings would take maybe eons to process and so on.

So it would be a set of technologies that are primarily based on machine learning and deep learning. Now these are two terms which you will constantly keep coming across in the next five days and later when you wish to you know build up on the knowledge that you've gained in this program. So machine learning and deep learning are two sub sets of AI or artificial intelligence, which are used extensively for this data analytics, predictions and forecasts, object categorization, applications like natural language processing, giving you recommendations, then intelligent data retrieval and many, many more such applications.

Now, these AI systems learn and improve through exposure to vast amounts of data. So, the larger the amount of data... that we feed to them to make them learn the concept for which they are supposed to give a solution. Then they identify patterns based on these data and establish relationships that sometimes human beings also may miss out on.

So their logical capability depends upon, or their capability to give us answers depends upon the training data or the amount of data that is fed to them to make them learn about the solution. Now this learning process involves algorithms which are sets of rules or instructions that guide the AI's analysis and decision making capability. Now as we see in this figure, artificial intelligence is the umbrella term and if we put it simply we would be able to say that it would be the ability of a machine to imitate human intelligence. Then as I said there are two subsets. machine learning and deep learning.

In machine learning, the algorithms are trained on labeled or unlabeled data. Labeled data would mean that we are classifying it, and unlabeled data would be, as the name suggests, it would not have these classified fire labels on it to make predictions or categorize. The categories would not be there, so the categorizing would be done by the algorithm. So this data would make predictions or categorize information into classes.

Deep learning utilizes something called artificial neural networks with multiple layers. So these layers, there are many, many layers to train this network. That is why it is called deep, deep learning algorithm. So these layers enable it to process information and thereby mimicking or imitating the structure and function of the human brain.

as you are aware in the biological system the human body is made up of neurons which are the basic units which pass messages from the brain to the other parts of all the other organs of the body, helping it to perform daily functions or for our reactions and so on. So when an attempt is made to mimic this or to imitate this in as an artificial set of network, we call it the artificial neural network, when there are many such neural networks connected to each other, passing on their information and learning backwards and forwards. We call this a deep learning network. So these are algorithms that mimic human brain to incorporate intelligence into the machine. Now, through continuous learning and adaptation, the AI systems become increasingly adept or used to performing specific tasks.

which can range from recognizing images to translating languages and beyond many such applications. We will discuss those applications which are possible through artificial intelligence in just a little while. So the core goal of artificial intelligence or creating an artificially intelligent machine is to emulate human intelligence by machines and the closer we go to this the better.

decision making the machine is able to do. Now this entire process of emulating the human intelligence would involve tasks like reasoning, that is analyzing information and drawing logical conclusions from it, then learning, that is acquiring new knowledge and skills from the data that we already have to be able to do problem solving, then identifying and then solving the problems in a goal-oriented way, which will involve the learning which we've had in the previous step. Then once we are done with this, the final step would be to be able to give a decision for the solution, which will be on the way to the solution. So decision-making, that is to evaluate options and to make choices which are based on the available information. The aim is to make intelligent choices.

and be able to choose the best solution and go forward with that. Now, this artificial intelligence or machines based on artificial intelligence are based on a core set of concepts and technologies that would enable machines to perform tasks that typically require human intelligence, as we've been discussing. So, the sub... sets of this would be machine learning, which we've just talked about.

This is basically the backbone of the AI systems where algorithms learn from data without being explicitly programmed. It involves training an algorithm on a dataset and allowing it to improve over time and make predictions or decision based on new data. The algorithm is provided a set of data, it learns from that, and then gives us a solution.

When more data is supplied to it, it improves and makes better predictions and data by learning from the more the data it has, better learning process will happen. Then we have neural networks. These are inspired by the human brain and these are networks of algorithms, many layers stacked together.

So layers of algorithms or networks of algorithms imitate the way the neurons interact with each other. Now this allows the computers to recognize patterns and solve common problems in the fields of AI, machine learning and deep learning. Deep learning is a subset of machine learning.

Deep learning uses complex neural networks with many layers, therefore it is called deep. So it learns from a set of several connected neural networks or layers of neural networks to analyze various factors of the data or various parameters of the data. This is instrumental in tasks like image analysis and speech recognition. Then there is another branch called the natural language processing NLP. This involves programming computers to process and analyze large amounts of natural language data.

enabling interactions between computers and humans using natural languages, the languages that we are used to conversing in. Then robotics. While often associated with AI, robotics merges AI concepts with physical components to create machines capable of performing a variety of tasks, which could range from assembly line systems in manufacturing to complex surgeries and so on. So we could use AI-enabled robots or bots to do several tasks which humans otherwise would not be able to do or if they are able to do it. they would not be able to do it in that many numbers or they would not be able to do it with that precision with which the robot or the bot would be able to do enhanced with AI software.

Then we have something called cognitive computing. This AI approach mimics or imitates the human brain processes to solve complex problems which say for example could often involve pattern recognition, then natural language processing that we discussed earlier, and data mining. Data mining from large amounts of data. People are able to locate the data of their choice or the data that they need for a particular problem-solving activity. Then we have something called expert systems.

These are artificially intelligent systems that emulate the decision-making ability of a human expert. applying reasoning capabilities to reach conclusions. So the expert system is designed for a particular concept.

So we could be using an expert system, say, for medical problems in the medical field. So the system would be designed that the knowledge related to the medical science would be fed into it. Algorithms would be related to that and different reasoning would take place.

place to reach conclusions or solutions related to that particular field. Then if we are talking about the geospatial field, then we would be looking at data images, other kind of the ancillary data, all fed into the expert system along with the rules, rule sets, and then the reasoning would be applied and we would be getting solutions to the problems that we have at hand. Now each of these concepts helps to build systems that can automate, enhance and sometimes they could also outperform human capabilities in specific tasks.

Like I said, when we have a huge amount of data, big data so to speak, then the analytical capabilities of the computer systems combined with the reasoning and learning capabilities of the human intelligence. combined together they would outperform singular capabilities and give us better decision-making abilities. So this figure would summarize what we've been talking about. We could do predictive analysis, we could convert text to speech and vice versa, speech to text, image recognition, machine vision.

language processing so you could classify translate and extract data we have expert system as we've just talked about then planning and optimization and then we have robotics which then strengthened or enabled with ai software or algorithms would do certain things which humans would not be able to do as well as these computer systems and algorithms together would be able to do for us Now, let us just go through how the artificial intelligence field has evolved. What is the history? And we already know that it is not a new concept. And the development of artificial intelligence started in the early 50s, 1950s.

And today it has come a long, long way from where it started. Now, the timeline that we have with us is that in about 1943. The idea of a model that would imitate brain cells was introduced. Then in 1950, the well-known scientist Alan Turing developed a test to see if machines could think like humans.

And the first computer based on neural networks was built, artificial neural networks. Then in 1956, the term artificial intelligence was called by John McCarthy at the Dartmouth. conference and in 1963 the first AI research lab is set up at Stanford University.

The first chatbot was created in 1966 called ELISA. The first productive expert system was developed at Stanford, and they were called Dendral and Mycin. This was done in 1969. Then there came a period, or a decade almost, where interest in AI dropped. And this led to a period that we know as AI winter.

Not much. Research was done in this field during this period from about 1972 to 1980 and therefore it is called AI winter. Then in 1980 the research in this field of artificial intelligence picked up again with the development of successful business applications. Now to enable computers to learn from data, machine learning techniques like neural networks and genetic algorithms were created. So the interest in this field was renewed and people started working on algorithms which would you know build new techniques in this arena.

In 1997 IBM created its supercomputer Deep Blue. We all of us are aware of this and this machine defeated the then chess champion Gary Cosgrove. In the 2000s Google makes brick tools in speech recognition and this was due to advances in machine learning, natural language processing and computer vision. All of them put together and we are now very familiar with speech recognition software and we several of us use it in everyday working also. Then the decade from 2010 to 2019 computers were now capable of doing tasks like speech and image recognition with previously unheard of levels of accuracy.

And this was, all this was owed to the deep learning techniques like convolutional neural networks and recurrent neural networks. Now, today's discussion will not explain these terms in detail. This will be done by Dr. Poonam in her next two sessions, where she will be giving you an overview of machine learning, what are its concepts.

what are the algorithms that are used and how they are used and what are their advantages. Then she'll be moving on to deep learning. She will take up convolution neural networks and recurrent neural networks and explain what the two algorithms do specifically in the field of geospatial data analysis.

Then we come to 2020 to about the present day. So we now have all several applications which are AI-enabled, self-driving cars, virtual assistants, we use them for several applications, medical diagnostics. This is a field which uses this technique heavily, and also drug development. These are just a few of the many present uses of AI. Several companies unveiled their conversational AIs like the Googlebot, Microsoft Bing, the ChatGPT by OpenAI, and so on.

So we are all used to today's field is very, very dynamic and we have unheard of technologies available and we are all using these in everyday functioning as well. So as you see here, we have a little more details of from right from 1943 onwards to the present day and where we have seen that you have. realistic humanoid robots capable of displaying human expressions and they interact with human beings and the chart GPT of course where which is capable of processing data and both images and text and produce complex and go through complex tasks and give us solutions. We are also used to virtual assistants like Alexa and Siri and so on, where we give instructions through voice instructions and several things like playing music, updating your shopping list and so on and so forth are done for us through these AI-enabled virtual assistants.

Now, let us go and see what are the types of artificial intelligence. capabilities available. So the classification is based on capabilities and functionalities.

So based on capabilities we can say we have three types. It's been divided into three sections, narrow AI, general AI and super AI. Then based on functionalities, we could have four such types, reactive machines, limited memory, self-awareness and theory of mind.

We will just discuss these in brief. So this is categorized. on four kinds based on the kinds and level of difficulties of the tasks the system is capable of performing.

So reactive machines are the most basic kinds of AI. This is called the reactive technology, which is built just to respond to inputs and it has no memory or capacity for learning from previous events. It works only on the data that is provided to it in the present scenario.

So whatever data it has, it works on that. So we see spam filters, chat box, recommendation engines, say from Spotify or Netflix and so on. Then limited memory.

This has the capacity to learn from past data and make decisions based on such data. So speech recognition software and recommendation engines are again two such examples. For example, generative AI tools like chat box, self-drive.

cars, virtual assistants and so on, as they have only limited memory, right? And the capacity to learn from these is based on this data. Then the third kind would be the theory of mind. Now, this understands the need of other intelligent entities also. So, it is this kind of AI should be able to communicate socially and comprehend ideas and emotions, much like the human beings do.

So, the theory of mind describes a machine's capacity to perceive and anticipate the thoughts and feelings of other people. This is how we human beings react, right? We look at a person and what kind of emotions they are displaying, and then we change or act accordingly. So right now, it is still very early for this kind of artificial intelligence, and it is still in the research phase and has...

been in the development phase, so it has not much developed yet, but work is going on and research people are researching in this area so that social interaction etc. in human-like, more human-like intelligence is incorporated into the machines. Then the fourth kind is self-awareness, where artificially intelligent machines that have the capacity to acquire a feeling of self and consciousness. So then these are termed or referred to as self-aware AI. So with the present technology again this is a purely theoretical idea that is yet not feasible.

So the aim is to have to evolve these machines so that they have human-like intelligence and they are self-aware, aware a kind of consciousness. inside of them so that they can react accordingly. But again this is right now a theoretical idea and does not have any available applications. Then if we talk about artificial intelligence machines which are categorized on the basis of their capabilities, then we have ANI that is narrow intelligence, AGI which is general intelligence, intelligence and super intelligent AI which is ASI. So for example, ANI would be machine learning which specializes in one area and solves one problem.

Machine intelligence or the second stage that is general intelligence, this refers to a computer which is smart enough and it acts like a comes across as a human when you converse with it or when you interact with it. Then the third one which is super intelligent machines or super intelligent artificial machines, then here we talk about machine consciousness which means that an intellect that is smarter than the best human brains in practically every field has been developed. So let us see a comparative analysis between these three kinds. A and I is also known as weak AI.

Its counterpart, general intelligence, is called strong AI or human-level AI. And superintelligence, ASI, represents a level of intelligence significantly more advanced than even the most brilliant human minds put together. And AI refers to AI systems which are designed and trained to perform specific or narrow range of tasks. So several such systems are known today and are being used for. several applications.

The ADI refers to systems, artificial intelligence systems that can understand, learn and apply knowledge across a board to a range of tasks at a level which is comparable to human intelligence. Then the superintelligence ASI refers to systems that surpass or are beyond human intelligence and capabilities in all aspects including creativity, general wisdom, and problem solving. Now, AGI systems operate under limited predefined set of functions and cannot generalize knowledge beyond their specialized domain. So, they work for particular domains, they are made for specialized domains.

The AGI systems are characterized by generalization capabilities, which allows them to perform any intellectual task that a human being can do. So they can do several tasks, they can be trained to do many tasks at the same time. Now potential characteristics of the ASI systems would be that they have superior problem solving abilities, enhanced creativity and innovations and much advanced decision making skills and strategic planning.

Now examples of narrow intelligent intelligence-based artificial machines would be voice assistants like Siri and Alexa, recommendation algorithms on platforms like Netflix, Amazon, Spotify, and so on. They could also be chatbots and some image recognition systems. We have been using these examples and the AGI systems are able to mimic human cognitive functions, including reasoning, problem solving, and understanding complex concepts.

Now, they are also able to adapt to new and unfamiliar tasks without specific training for each new challenge. So, they are able to, you know, learn from the data that they have, and they are able to understand and then learn and then apply that knowledge across several such broad range of tasks. Now, ANI is one of the most common forms of AI which is in use today. AGI still remains largely theoretical but the entire gamut of functions has not been achieved as yet.

The concept of ASI or superintelligence is still highly speculated and it remains a topic of theoretical exploration still. Now, the characteristics of the first category that is ANI is that it is highly specialized in one area. It is unable to perform tasks outside its training scope and it operates under predefined parameters and tools. The implications of the AGI, that is the second category, could be once the theoretical concepts have been achieved. could be that it would revolutionize industries by automating complex decision-making processes, which are unheard of till now.

However, it also raises some questions about its ethical use and safety concerns regarding control, ethical use, and impact on employment. So there are still some implications where people are, human beings are worried at present that how will these affect once they are totally functional, how would these affect these fields? Similarly, the implications of the super-intelligent machines are that they could lead to unprecedented advancements in science, medicine, technology, and many other fields. However, again, like the AGI, general intelligence, when it reaches that level, it would pose... significant concerns about ethical, philosophical and existential risks, including concerns about alignment with human values.

potential unintended consequences because after all it would be a machine and would it be able to you know learn as much or have the consciousness of a human being and be able to interact and react in those ways or would there be certain risks in terms of the ethics and would it be able to align with human values and so on so it's still unknown people are exploring it but there are concerns which need to be addressed. Now let us see what the differences between artificial intelligence and human intelligence would be. What would be the basic differences? The first point to discuss would be the information processing. Now AI processes data using algorithms and makes objective data-driven decisions.

There are no ifs and buts. Based on the data, it makes objective decisions. Human beings combine emotional and logical reasonings. Their decisions are also, you know, influenced by their emotions and some logic that they have. So cognitive reasoning, adding depth to their decision-making.

The efficiency in learning. So the second point is how efficient are. the two systems in their learning process.

To learn efficiently, AI systems require large data sets to perform that learning, to acquire that knowledge, whereas human beings can learn from a few examples through abstract thinking and reasoning. So the amount of data that is required to train artificially intelligent systems or machines is very very high. A large amount of data is needed for it to be able to make proper decisions.

Whereas when compared to this, human beings require very, very little, less amount of data. Adaptability, how adaptable are the two systems? Human beings easily generalize knowledge across tasks.

Whereas it has been observed that AI often struggle which with the jobs or tasks that it wasn't specifically designed for. So even if there is a small change in the process and it has to make a decision, it falters sometimes or it gives us a not so correct answer or maybe an altogether wrong decision or solution when it is used for something which there is a change in, sudden change in. in the problem that is provided to you and if it wasn't specifically designed or trained for it. Whereas human beings are not so, they are easily able to generalize their learning or knowledge across different tasks. Now, emotional understanding, these two systems differ in this emotional concept.

So, humans naturally interpret emotions. whether somebody is sad or happy or depressed and so on and so forth. So these are emotions that we are used to interpreting and this helps us in the complex decision making process.

So we change our decisions or solutions based on these emotional inputs. Artificial intelligence, because it is a machine, it has to perform based on the rule set that is provided to it, lacks emotional intelligence totally at present and struggles with social nuances. So, as you are aware, as human beings we interact or react socially and in different social contexts, we change our behavior and behave accordingly. Whereas this is not possible for the artificial machines or artificial AI-based systems.

The fifth very important element of difference between AI and human intelligence or human beings would be ethical reasoning. Human beings have moral reasoning, they have moral backgrounds and are also capable of ethical reasoning. So, they also have ethical concepts which they conform to and which are very, very important for them.

Now, AI follows the ethical framework which is set by its creator. So if we put in those rules or the creator or the person who has generated these rulesets puts them as input for it to learn then it follows though that framework with potentially perpetuating biases otherwise it has no such qualms about being ethical or morally correct and so on. It will only go on processing the data, learning from the data and giving you solutions or giving you answers. of performing tasks according to the data that has been fed to it. So currently the level of difference between AI and human based intelligence is quite a lot.

People are striving, the researchers and engineers are striving to achieve a balance between both these systems and trying to make machines. more like humans and perform tasks with, they are trying to incorporate emotional understanding and reasoning and by providing them a set of big making them adaptable like human beings, the machine, the AI based systems. Now let us see what are the advantages and the limitations of these artificial intelligence machines or systems. Now let us go through the advantages first.

The advantages are that there are less chances of errors. AI can perform tasks faster with greater accuracy, more accuracy than human beings. This leads to improved performance and productivity in many areas.

Now, there will always be a possibility of inaccuracies in jobs or tasks, which requires precision when people are involved. because they sometimes make subjective decisions. Increased safety.

Now using AI-powered machines like robots etc can help in saving human lives and it can be used at places which are high risk for example nuclear plants, coal mines etc or collecting scientific data from inaccessible areas like glaciers, volcanic craters and so on. The AI systems help us in repeat performing repetitive tasks so there are several repetitive tasks that we do in daily life and they take a lot of our time so these tasks can be assigned to machines for example which would then help us in reducing time as well as cost for example industries which have an assembly line production could have automated could automate several tasks and then they would save on time and cost Now these AI enabled systems are available 24-7 unlike human beings who require to be rested to be able to perform to the fullest capability. AI machines can run 24-7, 365 without getting tired and it can run non-stop and provide continuous service and support.

So this is a very big advantage. Then what are the limitations? it is very important to know the limitations in the current scenario.

So these systems are expensive, they cost a lot because AI system development and implementation is complex and it involves a cost. This needs a lot of resources and knowledge. So currently they are not so easily available, they are still rather expensive.

Then the biggest limitation is that they can't think what we human beings know as thinking. This is a certain limitation. It can only perform tasks and operations that have already been defined. So it cannot think for itself and perform additional tasks.

So if it encounters a difficulty in between or the situation is changed midway that it is performing the task, it might not be able to align itself and give us a proper solution. is not capable of thinking like human beings. Also, it has no emotions, so it is not emotionally intelligent. Machines can perform tasks better and faster than human beings and can also do it with much more accuracy and precision.

However, the emotional connection or the emotional connect, which is very important while working with each other, the machines lack this at this moment. Now another very big question that or point which is people are concerned with is that the incorporation of AI-based machines freely into the system can may lead to unemployment. So because AI would be able to do the job of several human beings put together and faster and quicker and with much more precision, people might not prefer employing human beings, especially for repetitive tasks.

tasks which require long hours on duty and so on. So this has definitely made the AI technique has made life easier but it has also taken up jobs that were previously done by human beings and this can lead to large-scale unemployment problems especially for areas or countries where we have a large amount of human population this could be a very very big problem in the future. Now after having gone through all these concepts or what is artificial intelligence, what are its subsets and how it is advantageous and what are its limitations, let us see why it is important. Now the basic points that can be said is that it improves efficiency, it solves complex problems and it makes better decisions in certain cases of course. So increased efficiency and productivity because it automates repetitive tasks, frees up human time and resources for more strategic endeavors.

Imagine AI powered robots handling assembly lines in factories or chatbots managing customer service inquiries. This will allow human employees to focus on innovation or complex problem solving activities. Then if there is enhanced decision making. then AI would be able to analyze vast amounts of data to identify patterns and trends that human beings might miss altogether.

Now this would allow for data-driven decision making in different fields, especially finance, healthcare, marketing. For instance, AI could analyze financial data to predict market trends. It is already doing that. It predicts the trends in the stock market, or it could go through patient medical data to suggest personalized treatment plan.

People are using it for such applications now. Then for innovation and progress, it could accelerate scientific discovery and technological advancements. AI powered research tools can analyze complex scientific data.

So we could go through big data and make informed decisions, simulate experiments through AI tools and identify promising areas for further exploration. Then it would also be able to give an improved quality of life. So it has the potential to revolutionize various sectors which would lead to better lifestyle, a better quality of life for certain people. For example, if we are using self-driving cars, it could improve transportation safety. AI-powered prosthetics could enhance mobility for those.

who have disabilities in their limbs and so on. So it could be a life-changing experience for them. Then it could also address global challenges.

So it could be a powerful tool for tackling challenges like climate change, resource management. It could optimize energy use, enable us to smartly use the available energy and sustainably use it, predict weather patterns. analyzed environmental data to support sustainable practices, which is the call of the day or the need of the hour.

These are all global challenges, and AI-enabled technology or systems would greatly help us in tackling these challenges effectively. Now, in a nutshell, let us see what it is that makes this technology so useful. It provides us or offers us several critical benefits, which makes it an excellent tool. The first one being automation. The second one being enhancement.

That is, it enhances the products and services effectively by improving experiences for end users and delivering better product recommendations. Then it has the capability of analysis and accuracy. So the analysis that it makes is much faster and more accurate.

than what the rate at which human beings would be able to do it and it can also use its ability to interpret data to give us better decisions because it will be able to go through more high amount of data, huge amounts of data in a small amount of time and give us decisions based on that. Now, another very, very important point is to go through ethical considerations and challenges in the AI field. One that the AI field, the AI enabled machines or systems would have. they would handle biases and would they be fair. So the bias and fairness would be defined as the systematic and unfair discrimination that is often a reflection of the prejudices that exists in its training data design.

Now because it has does not have the capability of thinking or cannot make logical decisions so these systems are biased based on the input data or any prejudices would be reflected. in the training data or the way that the training data or the design of the system has been done. And this would affect its fairness or equity.

Now, the challenges are the first one being in data quality of the AI systems, the input data, that is the training data. If this data contains biases, the AI output... is most likely to contain or carry these biases. Then the algorithm design.

Algorithms may also contain implicit biases introduced during the design process by the developers. So the design of the algorithm is again of utmost importance. And then the impact.

The biased AI can lead to unfair treatments across various sectors. Suppose we are using an AI decision-making tool for criminal justice. or as a hiring tool or maybe for lending impact.

This could impact individuals based on race, gender and other characteristics if the input data and the algorithm design is biased. So very, very important to consider the bias and the fairness aspects. Then there would be privacy issues.

So the AI systems often require vast amounts of data to function effectively. Now, which can include sensitive personal information, for example, in the medical field. So ensuring the privacy of this data is a significant ethical concern.

Now, challenges here would be data collection. So collecting large data sets could lead to overreach into personal privacy. And this has to be properly managed.

And the aspect of personal privacy has to be taken care of. This can be done by... data storage and access. So how it is stored and who accesses it are crucial in maintaining privacy. Then surveillance.

So AI technologies like facial recognition, behavior prediction could lead to increased surveillance, raising concerns about individual privacy and autonomy. So these are, again, very sensitive issues and need to be dealt with as such very, very carefully. The third point of concern is job displacement.

Many people are discussing this, and this is a problem that we could be left with. So AI and automation can perform tasks more efficiently than humans in many, many industries. And this is leading to fears of widespread job displacement or job loss.

So the challenges here are economic impact. So while it would increase productivity, it would also pose a risk to jobs in specific sectors, especially those in manufacturing, transportation. and maybe administrative support and so on. Now, skill gaps.

So there is a growing need for retaining and reskilling workers, provide them with skills which would help them deal with this and prepare them for the jobs of the future, where human insight and interaction with AI systems will become very critical. Then social inequality. Now, job displacement would widen the economic gap. between those who have the skills to work with AI systems and those who do not. So these are again very very critical points and need to be addressed and taken care of.

Then the safety and security, as these systems become more prevalent they become more in use ensuring that safety and security against malfunctions, misuse, hacking is of paramount importance. So system failures need to be seen. Sometimes they can be very, very catastrophic, for example, in areas like autonomous vehicles or healthcare.

Now if that system has been hacked into, there would be accidents on the road. health care systems if they are hacked or if they don't if they malfunction then you can imagine what would be the scenario then if somebody takes it into their mind for malicious use use it for harmful purposes such as developing autonomous weapons or cyber attacks and so on again it is unimaginable what might happen then security measures have to be taken so protecting ai systems from being hacked and used against the interests of users requires, again, very, very robust security protocols. Now, I will quickly take you through how we can measure the artificial intelligence system, how close is it to acting like a human. So there were a few tests which were designed to see if the AI system is behaving like a human or is it not.

So there are four such tests, one which is the Turing test. tests whether the system is acting humanly. Then the second one is the rational agent approach. So the rational agent approach decides whether the system is acting rationally. The third is the cognitive modeling approach and this decides whether it is thinking humanly or not the system.

And the law of thought approach is a system where the rational thinking of the system is analyzed. Now the Turing test, the basis of this test is that the artificial intelligence entity should be able to hold a conversation with a human interrogator. So as you see here in the figure, the human interrogator is having a conversation with a human responder and with a computer. So ideally, the human agent should not be able to conclude whether they are talking to artificial intelligence.

So then the level at which it reaches decides how closely it is able to mimic the human intelligence. Then the co-operative modeling approach. So here it has to test introspection, psychological experiments by observing human behavior and brain imaging.

MRI, etc. medical techniques are used to observe how the brain functions in different scenarios and replicates it through a code. Then the law of thought gives us a set of rules that will help think which can be used to program artificial intelligence and rational agents use a strategy where the AI tries to make the best decision based on its current situation rather than following a fixed set of rules. rules.

Now we come to the last section where we will quickly take you through the applications or potential fields where AI is being used. It is being used in a wide area right from healthcare to education to automobile to finance surveillance in the social media and entertainment. We are very used to using these space exploration, gaming, robotics, agriculture, e-commerce, you name it. and AI has pervaded that field.

So use cases could be from speech recognition where automatically one can convert spoken speech into written text and vice versa. Image recognition, it is used to identify and categorize various aspects of image data. It is very, very widely used in the geospatial field, as you will see later in the different sessions tomorrow. the forthcoming sessions. Translation, you can translate written or spoken words from one language into another.

Very widely used in predictive modeling, mine the data. Data mining can be done to forecast specific outcomes with a high degree of granularity. Analytics, one can find patterns and relationships in data for example to be used in business intelligence. Cyber security, so autonomously scanned networks for cyber attacks and threats and so on.

So these are just a few. Then for business intelligence, you can use it for data analysis, which would help you in giving you insights and recommendations for decision making. You can use it in healthcare for disease diagnosis, treatment development, and giving personalized care for the patients. AI in education.

Here we can use this technique to personalize learning, improve the student engagement. and automate administrative tasks so that human beings are able to engage in more productive jobs. Then in the finance field we can do risk and fraud detection, personalized recommendations can be made for products and services, then document processing can be done very quickly through AI enabled software.

tools. Manufacturing, we've been discussing this in the previous examples as well, that it would improve efficiency, productivity and quality. So very, very widely it has been started to be used in the manufacturing field, especially in the assembly line systems where repetitive tasks are to be performed. Then for other additional applications could be AI use in retail sector where it can be used to personalize shopping experience, recommend products and manage inventory.

Transportation, again, self-driving cars, improve traffic management, flights running on autopilot and so on. Energy usage, we can use AI tools to improve energy efficiency and also predict the demand for the energy. In government sectors where to improve public safety, crime detection.

provide several citizen-centric services and so on. Also, in the entertainment field, it can be used to recommend music, movies, and so on. So, online streaming companies are able to use this to provide suggestions to their customers based on their previous interests or preferences. Now, I will only give you a very short overview of how AI is being used. for the geospatial arena and what is the big picture.

So AI, machine learning, deep learning are all being used, specifically machine learning and deep learning. And when clubbed with artificial intelligence, a term is coined called geo-AI. So we can use GIS, that is geographic information system, imagery data or spatial data, club it with... artificial intelligence can make informed decisions.

So big data can be analyzed very quickly. Geographic data can be analyzed using these when combined with spatial principles. So the steps would be for geospatial analysis when we use artificial intelligence would be, one would be the data collection from various sources. So we could collect data from satellite-based images, earth observation sensors, that is from laser sensors for 3D data, UAV-based sensors and so on, GPS, and then use it for real-time monitoring, simulation of scenarios for multi-objective optimization and smart decision support is enabled, and then the scenarios are assessed, managed, and plan designs and implementation is done. stakeholders are engaged and web-based geo-visualization and early warning systems are generated which are used for monitoring, evaluating and management plan adjustments.

So this is how geospatial analysis can be done. All these will be dealt with in detail in the next four sessions. Some practical demonstrations will be given on how to use available resources like the Google Earth Engine which is a vast repository of these kind of geospatial data on the web.

It's a cloud-based platform. It gives you both data and the tools to process it. So all this will be discussed in the next section.

Now the general workflow here would be data acquisition and preparation. So if we talk about image data or any other kind of data, so image data and label data is labeled. for specific locations of the feature that we want to identify. For example, if we want to identify buildings, then we need the image of the target area, that is the building, and the label of the vector outline of that building, and then we generate training data in the corresponding format. Then the next step would be building the model building and management.

So you construct the model and manage it. To train the neural network model based on the training data samples generated on the previous data. preparation process.

So the data acquisition and preparation here we prepare the data which is then input. to train the neural network based on the samples, different kinds of features that we want to identify. Now, iteratively we evaluate the training model through the validation data set and the test data set to achieve the actual application accuracy and precision requirements. So this is how this is this is done at the preparation step and when we build the model we iteratively evaluate it when we input the training data by splitting it into training into test data and validation data.

All these again I stress these will be explained how it is done in detail in the next two sessions. Then we come to the last step that is the inference. So the trained model is used to predict or infer the testing samples and comprises of a similar forward pass as training to predict the values. Then at the last step, it applies the learning that it has acquired in the previous two steps and gives us a solution or the predicted values. So the general framework would be something like this.

You input data, if we are working with images, and then perform the network learns. and it gives us outputs labels either classification object recognition and other understanding so image use cases when we are dealing with images could be image classification object detection semantic segmentation or instance segmentation now these are terms that you will be discussing in the next four sessions just to give you an example of how geospatial data can be used say for identifying crime hotspots, generating risk maps, yes, and using for crime analytics by using machine learning models and overlaying it with hotspot zones and so on. The second example deals with identifying types of roofs.

So the identification of built up area is done and then buildings are identified and in those buildings the second step is to identify features like trees roads buildings and then in the building data set the type of rows is classified all based on a training data set so model training is done here and then then sample tiles or build chips of different types of routes is given as input and then it searches the system searches and gives us two types of out that are available in the system. Road identification can be done, different types of roads, et cetera, based on AI training. Again, tree classification has been done here. So here it is used using two kinds of data, one which is image data and the other which is height data.

So the trees are also classified based on their height. What are the kinds of parent? This just gives you an idea of what are the kinds of training inputs you can give.

so height difference, change rate, intensity, number of returns, anisotropy, planimetry, and so on. This can be done with data like laser data which gives us 3D point clouds and all these can be used as input parameters for training the model. Now I've overshot my time but I hope you have had a fruitful learning experience and I have tried to give you an overview of what the general artificial intelligence systems look like, what they are, how they are defined, what are their pros, what are the points that we need to be aware of, what are the limitations, and how they can be used for different applications. Then the last three or four slides we discussed about specific application in the geospatial field where we can use geospatial data.

club it with AI decision-making tools and perform. With this, we come to the end of this session. Thank you very much. I'll be happy to take up questions if there are any.

Otherwise, we can do the query questions and then we can send you answers later on. has posted that as seen in some generative AIs, hallucinations are observed in some complex scenarios. Can a similar false positive negative be observed in GOI?

Well, there is a large chance of observing such false positives and negatives in GOI as well. As I said, the response of the AI system will be biased in terms of the training data. So if we ensure that we have comprehensive training data which is as free from biases as we can make it, then the output is likely to be unbiased and such false positives or negatives can be minimized. Do we have to use Google Colab or ArcGIS? You mean in this course or in otherwise, whenever you are using it?

You can use either. RTIS has certain tools catering to specific applications. So you can use that for those specific applications, whereas you can use tools like Google Colab, et cetera, and several other tools to generate your own model. Now, in this five-day program that we have, If you're asking specifically about that, then I don't think we will be having time to be able to do things practically and guide you to do those hands-on.

Though there is a session on Google Earth Engine, but they will only be demonstrating to you how that tool can be used. However, as you have asked, Google Colab, ArcGIS, and several more tools are available. You can use all those.

Okay, Akash, this is a very, very interesting question and I am sure people are working on it. Can we use biological neurons and somehow embed them to a biotech so that we can train the neurons? Something like Neuralink that is happening today.

There are, you know, we might yet be unaware, but various sections or the technological people are in. They are extensively researching these kinds of applications. Of course, we have to keep in mind the ethical part of the considerations that we need to do along with this. We might launch ourselves into a totally unpredictable domain if we do that, but I'm sure it would be possible. And there are people researching these.

No concrete. solutions have been come up or a functional system has been made but yes people are doing it Okay, how does ISRO integrate AI and ML techniques in geodata analysis for real-world applications? There are several applications and many, many solutions underway. I would suggest that you wait until tomorrow to be able to see examples of these applications. So one uses it in climate change prediction.

They are using it in... say, land use, simple applications like land use, land cover detection, and so on, chain detection, and so on, to climate change applications and several other geospatial data analysis. Tomorrow and day after, and in the Google Earth Engine class, you will be exposed to all these examples and applications.

What possible hallucinations can AI generate while working on geodata analysis? Well, what possible biases etc. would be dependent on the training data. So, as I said, if our training data is flawed or if it is biased to a particular point, then that is how the bias would be How AI helps to find risky places or accidental places? Well, I think you are wanting to ask how we can use it in the place where there is a risk and so on.

So suppose there's been a forest fire and some rescue operations need to be done, or some data is to be collected from such areas, or maybe scientific data is to be collected from glaciers, medicines, etc. have to be dropped. All these can be done through a robot. deep sea data collection, deep space data collection, where it is nearly impossible for humans to survive. So all these areas can be done through AI.

ANI is widely used as compared to AGI. Well, because AGI and the super intelligence phase is still under development, the machines have not yet been able or the systems have yet not been able to achieve the level of intelligence or the decision-making capabilities and work is still on. Can we train our model to identify space patterns related to stars and pre-planet constellations? Yes, definitely we can do that depending upon the data that we input.

I think we can stop now. Any other questions that you may have can be posted to the distance learning team platform and we can answer them collectively from there. I hope you had a fruitful learning experience and an enjoyable session.

Thank you very much. We meet again tomorrow at four o'clock in the evening. Thank you.

Thank you. Thank you. Thank you. Thank you.

Transcript for:Overview of AI ML for Geodata

Transcript for:
Overview of AI ML for Geodata