Welcome to AI and Disaster Management, the third course in the AI for Good specialization. In the first two courses, you learn about how AI can play a role in FSA's big challenges in public health and climate change. In this third and final course of the specialization, you look at case studies of natural disasters and how AI can aid in disaster response and management. I'm here again with Robert Monarch, who will again be your instructor for this course. In addition to developing many AI products in Silicon Valley, Robert is also a deep expert in disaster response, and he has been involved in numerous disaster response and recovery efforts over the past few decades. Thank you, Andrew. It's great to be here and to speak about an area that I really care a lot about. So before I moved to Silicon Valley, I was working in post-conflict development for the United Nations in Sierra Leone and Liberia. And then even after moving here, I've continued to work in disaster response. So here in the U.S. for cases like Hurricane Sandy, where the company I was running did aerial imagery analysis for FEMA, immediately following landfall, through to other projects worldwide. So following an earthquake in Haiti in 2010, following floods in Pakistan at the same time, and most recently working with a number of organizations responding to the COVID-19 outbreak. I feel like there are many people whose lives you have impacted that probably did not realize your technical skills or even your pre-AI days, that that work was making recovery from some of these tragedies around the world much better for many people. Yeah, I appreciate that. And for me, at least, I worked in disaster response for a long time. Having studied AI as an undergraduate and not initially even thinking that there was much of an overlap there. And it's been interesting to see, as time has gone on, how ubiquitous AI has become in the world in general, which also means that there are cases where, when people are the most at risk following a disaster, AI can be part of a solution for them to help themselves recover from that disaster. I remember one of the first projects you worked on was on language, because you realized that with cell phones coming online, a lot of AI was done in English, but not necessarily in the native language that would allow governments and systems and disaster recovery organizations to understand native languages to serve the needs in a disaster response scenario. Yeah, I think a very important fact about the world's languages is that there are so many, so about 7,000 languages in the world. And if you speak one of the less widely spoken languages, then you are more likely to be the victim of a natural or a man-made disaster. And so for a lot of the problems that we face in disaster response scenarios, as international responders, we're going into an area where there's almost no one who speaks that language from among the international response community. And so some of the most important things we can do are enabling communications between responders and their affected populations, and also allowing the affected populations to themselves discover information, to differentiate real news from disinformation, and to help share whether it's market information or information about education or personal health care in their own language so that they can help impact their own recovery. In fact, I know that sometimes people still wonder, can AI make a difference? How can AI help people and save lives? I think Robert's work and what you learn in this course is one very clear, exciting, sort of tragic in some ways, but really exciting and promising application of these AI technologies. So the first project that you see in this course, you illustrate important concepts using the application of using satellite imagery to assess damage and help with disaster recovery after a hurricane. And I think one of the most interesting things I heard from you was paying attention to privacy, the fact that satellite imagery can be very sensitive data and being respectful of privacy, even while we're trying to be responsive and quick in a disaster response scenario. Yeah, this is something we think about quite a lot in disaster response. A lot of the time, we are recording information about people when they're at their most at risk. So in the case of Hurricane Sandy, the company I was running at the time, coordinated aerial imagery analysis. In this case, it wasn't satellite images. It was images taken by the Civil Air Patrol. And we saw that these were high enough resolution that they could be used to identify individual properties. And so after the response period, we shared the data with just one other company or an organization who could evaluate the quality. And then everybody deleted the data. So at no point was it made public and at no point was it stored long term. And this is something that's really important to think about in any use case. Similarly, in the first use case here, in the case of Haiti, we deleted all messages that could be used to identify an individual. So it's not the full data set, but it's a small enough set that the Haitian community looked at and agreed that we could share with the broader world. I think these are great case studies in how to address important challenges to individuals while also being thoughtful about the privacy implications and being very respectful of that, too. Yeah, that's exactly right. And as we'll go over a few times in this course, and you've seen already in the earlier courses, we have this principle of do no harm. So when we try to release any technology, whether it's AI or otherwise, in critical situations like public health and in disaster response, we want to ensure that no one is at risk of being harmed who would not otherwise have been harmed. So we're not looking like maybe in an academic paper where the net impact of an algorithm is higher accuracy. We are looking for impact where there is no one losing out as well. And then in the second project of this course will be one involving understanding text messages sent in the aftermath of a disaster. Do you want to say more about that, Albert? Yeah, so this case is one that I worked on myself. So following an earthquake in Haiti in 2010, more than 100,000 people were killed immediately. And as the international response community came in, we only had English as a common language, while most people in Haiti only spoke Haitian Creole. So I organized about 2,000 members of the Haitian diaspora who could do real-time translation between Haitian Creole and English, predominantly from emergency messages sent from the population in Haiti. And the people among the diaspora would also categorize and geolocate these messages. And so one of the things that was important to me following this disaster was understanding what could have we've done more with machine learning than we did. At the time, we used submachine translation, but we didn't automate the categorization or the analysis of the data. And so this actually became part of my PhD research, applying methods like LDA to analyze these messages, to understand how does a population communicate following a disaster versus a few weeks later. And so this is what we'll be doing in this lab. So you'll be using LDA to analyze these messages and to get an understanding of how the topics of these messages change over time. Very inspiring work. And so I'm excited for you to jump into this third and final course of the specialization to learn about these important key studies in disaster response and recovery. So please go on to the next video and learn about these important concepts.