#1 AI for Good Specialization [Course 1, Week 1, Lesson 1]

You. Welcome to AI for Good. If you're interested in how AI and machine learning can be part of the solution for real world challenges, things like public health or climate change or disaster management. This specialization will give you a sense when it comes to addressing complex real world problems. The potential solutions are often very complex as well. It might involve many different stakeholders. It may involve logistical constraints, sometimes data privacy issues and other things like that. So we've designed these courses to give you hands on experience working with AI applications so you can see how the AI technology piece fits into the broader context of addressing so some big challenges in the world. I'm delighted to introduce your instructor for this specialization, Robert Monarch, who is an expert in building machine learning systems and how to fit them into human workflows. He's founded his own AI startups and Rob has also built AI products at the biggest tech companies including Google, Amazon, Microsoft, Apple and Robert's worked for over 20 years in applying AI to addressing critical problems in the areas of disaster management and public health around the world. He holds a PhD from Stanford University and is the author of Human in the Loop Machine Learning. It's a pleasure to have you teach this specialization, Robert.

All right, thanks, Andrew. I'm really excited to be here. I look forward to sharing a lot of the experience that I've had both in industry and as a disaster responder to help you think about the ways that AI can, or maybe in some cases should not be used to help with areas like public health, climate change and disaster response.

So you called yourself a disaster responder and you're an AI machine learning person. Do you want to share how you wound up blending those two things together in your career?

Yeah, yeah, happy to. So I worked in machine learning and in disaster response separately for almost a decade. Immediately before I moved here to Silicon Valley to get my PhD, I was working in post conflict development for the United nations in Sierra Leone and in Liberia, and I was working on electrical systems there. So I was working in post conflict environmental development, installing solar power systems at schools and clinics, supporting refugee camps. And then following my time there, I came here to Silicon Valley, to Stanford, where we first met to start my.

Ph.D. many, many years ago.

Many, yeah. And some things that had really stuck with me from my time as a disaster responder, where in a relatively short amount of time in the early 2000s, most of the world started getting access to a cell phone. But even the AI that we took for granted back then, like search engines and speech recognition, didn't work in the majority of the world's languages. And so I thought, well, this is an interesting problem that is being faced not just in disaster response, but in public health and in industry worldwide. The world is coming online, but in order to access and interact with people, a lot of the support in AI wasn't there. And for a lot of these languages, it still isn't today, almost 20 years later. So it was when I was at Stanford that I continued initially to work as a disaster responder in, in parallel with studying natural language processing, and then eventually saw that there were some areas where AI and disaster response overlapped and was able to combine them.

Yeah. And so I think with the rise of cell phones and data, the data became available to do a lot of things, both in corporate, product, business settings and in solving these societal challenges. And sometimes when there's a disaster, even when cell phones are around, a neural network may or may not be the solution to what's happening out there in the world.

Yeah, that's right. And certainly there are good times and bad times to roll out new technology. Often following a disaster is the worst time to roll out something untested. So I think there is a synergy here in that a lot of what we can use to help people following a disaster is best built in preparation for disasters, maybe in preparation with industry partners who can help us road test and ensure technology works well so that we can understand its behavior when we deploy it to help some of the most at risk people worldwide.

So one of the things I'm most excited in this specialization is that in the context of public health and climate change and disaster response, it will step through how to fit in machine learning, AI in the context of a bigger project. And I think these tools, these techniques are useful for these grand societal challenges, but also for building a product in a company. And so I've seen in many contexts where machine learning engineer will say, oh, I do well on the test set, which is fantastic. When you do well on your test set, congratulations. But sometimes that bridging that gap from doing well on the test set to, you know, modeling out some part of climate change or solving some concrete business application, there can be a big gap. And as a machine learning engineer, I think it's part of my job to help bridge that gap. And I think the specialization will go through a lot of how to think about that.

Yeah, I think that's a good way to think of it. You can have a good machine learning model, but that doesn't necessarily mean that just because it improves in accuracy, that the downstream use case will also be improved. And in fact, in this course, transparently, I start with a case in public health where we had deployed a system helping maternal health. We could see the offline accuracy go up, but the end users who were healthcare professionals didn't feel that benefit and ultimately we sunsetted that project. And so to your point, I think a lot of what we learned from industry about making machine learning important to help people in human tasks is a lot of what we see when we're focusing on public health and disaster response and, and other projects which are more specifically focused on near term good.

So just for clarity, this is one of the least technical, maybe the least technical course that DeepLearning AI has offered so far. And you don't need to have significant coding knowledge or even any meaningful AI knowledge in order to get through the specialization successfully. But if you do know about machine learning errors, that that's great. You see how learning algorithms fit into bigger projects. And when you run through some of the code samples, which we fully provide, so you don't need to write code, but provide code that you can run even if you've never coded before, you'll be able to see how machine learning pieces can fit into solving these very important societal problems.

Yeah. So like we'll say throughout here, you often don't want to be innovated on machine learning in most critical situations. So not going so deep on the technical side here is aligned with practice. And I'm particularly excited that a lot of my colleagues from the disaster response industry might be able to take this course. So people who work in disaster response and public health run most of their data in spreadsheets and I think a lot of these notebooks are the right next step for someone that's been working with spreadsheets to think about how they can do a little bit more sophisticated exploratory data analysis and started evaluating machine learning models in a, in a coding environment.

Yeah. In fact, I liked how in approaching a brand new problem, you lay out a systematic framework that often starts with, you know, visualizing the data, kind of EDA or exploratory data analysis. And then that can feed into a thoughtful assessment of should you even use machine learning for this? Sometimes machine learning is fantastic and sometimes it's not quite the right fit to then, you know, build, building up the model, running the project, but to put that into a systematic framework for approaching complex problems.

Yeah, and I love the framework that we're presenting here, which is similar to the frameworks I've used in industry as well. So you're thinking about what is the real world problem you're trying to solve. And if you can define that real world problem without mentioning machine learning, so much the better. And then you're looking at the data and getting an intuition for, well, can machine learning help before you start sitting down and implementing the model and evaluating your results?

And one thing I'm very excited about is I do see many people starting to learn machine learning or AI and thinking, won't it be cool if I could apply this to problems like climate change or public health?

So one of the things I'm really excited about is that we have a number of people that we've invited to be guest speakers. And these are people working in different areas like public health and monitoring for wildfires, where they're using AI as part of their solution. And these are people who have spent their entire careers working together in social good and AI applications. So I'm very excited to be able to feature them in spotlights throughout this course.

And so with that, I think many of us have felt this desire to take AI and go out and let's make the world a better place, let's use AI for good. I'm excited to have you in this specialization and as you go through it, I hope that you learn how to build AI projects, that process that Robert and I were talking about, how to evaluate and make an AI project successful. And hopefully, perhaps after taking the specialization, you'll be inspired to go use these algorithms to actually go tackle and make a dent, make a difference in some of these most important societal problems facing us today. So with that, let's jump into the specialization. Please go on to the next video.