 Then my name is Stacy Wood and I am the global business development manager for startups at Amazon Web Services. I would like to welcome you to today's Just Tech Connect program on artificial intelligence for startups. For the next hour, my colleagues and I will be answering your questions about artificial intelligence, also known as AI for entrepreneurs and startups. Artificial intelligence is now everywhere in our day-to-day lives, whether it's ordering household products, conducting internet searches, playing music, providing weather forecast, or many of the other purposes that AI can serve. Entrepreneurs all over the world are also embracing the benefits that AI can bring to their businesses by improving efficiency and making their products more user-friendly and accessible to consumers. Now I'm excited to introduce my colleagues at Amazon Web Services. Hassan Soof is the director of artificial intelligence. Joe Spezak is the head of artificial intelligence partnerships and Ben Snively is a specialist solutions architect in data and analytics. Join our discussion by submitting your questions and comments through the chat role next to the video player or through Twitter at hashtag Just Tech Connect. Now before we start taking questions, I'd like to hear from our panelists how they would define artificial intelligence and how it can be useful for startups. Hassan, would you like to get us started? Hassan I'd be happy to. Good morning everyone. So my understanding of artificial intelligence is that we use data to learn from the machine. We make the machine learn from data and make decisions which otherwise would have to be done by the human. That is in general terms the definition in my opinion. There's multiple ways of doing this of course. You can pre-define the decisions by defining rules by someone sitting down and defining what is an answer to a specific question. You can define them by rules or today we use data to learn from so that we basically can adjust decisions according to real life and have basically a system learn the same way a human can learn. So that's I think what I would define artificial intelligence for myself. Thank you for that Hassan. With that being said, let's delve into the topic a bit more by taking some questions from our audience. First question, what are examples of startups or companies that are currently using artificial intelligence? Ben, would you like to take this one? Sure, absolutely. So a few startups to talk about. So in my role I focus on education, public safety, those sorts of organizations and one great example is a spot-shotter is using artificial intelligence in order to identify wherever there's gunshots within a city and then be able to react to that in minutes. Other startups are very easily adding intelligence as Hassan mentioned into their application. That might be understanding speech. That might be understanding video image recognition. So those sorts of technologies as well. Of course, wide range of areas from recommendation engines, financial trading, those sorts of things within the startup area as well. What are some online resources that we should know about in order to learn more about AI? Joe? Sure. I mean, I think you can always start with all the MOOCs out there. I think that's where a lot of people get their starts. Coursera has some great classes. So you know, Android class and machine learning is kind of the canonical starting point for many developers and then users of machine learning. Udacity has some great classes that would nanodegree now. I think all of the myriad of examples in open source. So you have the big trend over the last, I'd say, year or two here is to bring code into Jupyter notebooks or what we've previously called iPython notebooks. And so you can really now get fully worked examples of use cases with starting from data all the way through to predictions for many of these frameworks like MX, Ben, TensorFlow and Cafe and be able to apply your own data to solving different problems. So trolling GitHub and learning on some of the MOOCs on Udacity and Coursera is probably a great starting point. So the viewing group at the Information Resource Center in Kigali, Rwanda asks, how much capital would it take for an emerging business to build its own artificial intelligence systems? What are the skills required to build that kind of software? Hassan, I know you previously had your own AI company. So perhaps you can take this one. Of course. Of course. So today it's much easier to start up a company with artificial intelligence than it used to be. It used to be very expensive because the computing power which was available to you to build on top of that was not there, data was not there. You had to basically build a lot of things from scratch. So about, I mean, almost, I mean, about 18 years ago when I had my startup, it was a very capital-intensive endeavor to build something with artificial intelligence. Today you can relatively quickly with a few hundred thousand dollars start an application with which can utilize artificial intelligence. You can use these services which are out there and build on top of them. You can bootstrap the bootstrap capability which used to be not available. For example, when I had started my company in 1999, 2000, to do speech recognition and machine translation, we had struggled to gather the data which we can train our speech recognition on, which we can train our machine translation on. Today's resources are available due to artificial intelligence getting more mainstream, but also the world being ready to like share data. I mean, so the UN gives us data in multiple languages and it's now available on the internet. The European Union, same story, many, many, many content is already available in many different languages. Audio is available in acoustics and transcripts, too, of that acoustics, so you can basically ramp up a system very quickly. And again, you can bootstrap with existing systems. I mean, let's say, for example, you want to develop a natural language understanding system. You don't need to start it from scratch. Oftentimes, you can focus on things which other people already built, so other people built the platform. You can build basically your business intelligence on top of that, which is usually the better way of doing it, which saves you money, much, much money even more. So that would be my approach if I had to start a company now. Thanks, Asam. Lots of resources out there for folks looking to build businesses. Anyone else to add to that point? Yeah. Building on Asan, I think there's, I mean, even for Amazon, there's free tiers available for our AI APIs. Wouldn't recommend starting from scratch on things like face detection if you're looking to add that into your application. Those are kind of solved problems, and those are APIs that are largely available. You can use that, like I said, on a free tier and run thousands of images to test out your application and get it going. For most of the basic primitives in AI, you can actually leverage the APIs and start with. And we're also entering an era of being able to leverage small data in AI. I think there's enough pre-trained models out there. They're available in model zoos, like in Apache MXNet or TensorFlow, where you can take a pre-trained model out on a large dataset that costs someone maybe on the order of, say, tens of thousands of dollars on something like ImageNet or a large speech or a text corpus and be able to fine-tune that with a smaller dataset of your own and be able to get really great results. And that's a way to kind of start quickly instead of having to start from scratch or have to bootstrap a really large dataset, which is very expensive and takes a long time in some cases. One thing I would add on top of that would be skill set wise when it comes to bringing in intelligence into your solution or into your system. A lot of folks with application and engineering skills, you don't have to be a deep-dazed scientist to start bringing these intelligent systems into your application. So you could be using these higher-level services to add image recognition, image visual recognition. You could do speech-to-text and add conversations in your application without being a deep-dazed scientist. And I think throughout the stack all the way down to these lower-level libraries like MXNet and TensorFlow, all the way up to higher-level AI services have really enabled folks with different skill sets to add these intelligence into their system. Cool. Okay, we have another question from the audience. Stephen Rwanda asks, how can we solve problems in society, especially in Africa, with AI? So I think one of the critical things which AI might be able to help with is opening up content to anyone. I mean, this is, I think, this is important. Having AI help digest the huge amounts of data for anyone is probably a critical component. So that's, I mean, things like translation might be of relevance so that we can basically have all content in the world available to us or speech input or also things like, of course, I mean, now we're talking about, I mean, things like modeling of natural processes. I mean, there's many things which can be utilized across the world. I mean, that might be something which is of relevance. So depending on what component in Rwanda you want to basically cover, I mean, AI can always help digest a lot of data, so to say, and make them available to you. Abdul in Kandahar, Afghanistan asks, what are some ways that I can identify how to use AI in my new business? Joseph, do you want to take a crack at this one? Sure. I think it really depends on certainly the domain of the business. I think if it's something specific like maybe, you know, trying to do medical predictions or be, you know, a medical assistant or something along those lines and you're targeting something like computer vision, you know, there might be opportunities, obviously, to make predictions on diagnoses. You know, I think there's any time you're interacting with a human, you know, there's opportunities to take chatbots into the workflow and kind of add efficiency to the application, improve the user experience, you know, beyond just having, for example, a standard call hunter type of environment. So I think anytime there's any kind of basic primitive that you want to add intelligence to, whether it's being able to detect faces or be able to recognize people with images or adding an interactive chatbite interface to improve the user experience, those are opportunities for AI as well. I think there's, with, for example, like text-to-speech, you know, we've seen that's helped, you know, the visually impaired, you know, be able to listen to the books, as we've seen with, like, Washington Post start to integrate text-to-speech into all their articles, it allows you to personalize content as well and generally provide a better user experience. Thanks, Joe. So Mansour from Afghanistan asks, what are some examples of companies that have created physical manifestations of AI such as robots? Hassan, you want to take this one? So examples of that are drones, robots, in the environment of, let's say, in the warehouse environment, there is, I mean, there's startups which are building robots for the house, for the home use and for, like, interacting with children, utilizing basically, like, what, like a chatbot, but with the physical, with the physical appearance as well, which, so some startups are using that to support children with special needs, for example, they feel that they see that kids with children with special needs interact with robots in a much more calm way and they learn basically interaction can basically be trained with interaction on a, with a robot so that they're being trained to interact also with other people. Some of these, I mean, these startups are working on basically building the robot as an assistant, really, to these children with special needs to interact with a third party, with their parents, for example, because they feel that the robot is an extension to what they are, for example. Drones, I mean, you have heard about the drones which we are experimenting with to deliver packages into homes in which we are trying out in the UK and the US. I mean, there's a lot of, there's a lot of applications which have a physical appearance. And a couple more to, you know, just to kind of build on what Hasan was mentioning was, you know, in the physical world, you really see some sort of AI in a lot of artifacts, even if it's not in a robot. So, you know, just think of cars today, you know, cars have automatic braking systems, they have lane assistance, you know, those are forms of artificial intelligence, you know, it's performing an intelligent act on your behalf. You know, some other examples, NASA actually developed something called ROV, and that's a voice-activated Mars rover using services like Lex, which provides a conversational interface to a system. So, that's another manifest, you know, a physical object, you know, it's a Mars rover that they're controlling through voice and has intelligence built into it. So, I would just kind of mention on top of that, you know, you do see machine learning, artificial intelligence throughout many, many physical devices today as well. Thanks, guys. A question from MSI Abuja. What legal framework is used for AI across borders, given that the legal structures differ? Is any cyber governance possible? And so Ben, I know you've worked with the government in the AI space for quite some time. Do you have any insight into legal frameworks that guide AI? Yeah, so I don't know if I personally have hands-on experience with the legal aspect per se. I know, you know, there's definitely a consideration as you're building the system of any ramifications, especially in things like healthcare and that side of things. But I actually don't necessarily have anything to add on that, unfortunately. The viewing group in Lincoln Learning Center in Gardez, Afghanistan says, according to online sources, Amazon recently launched drone delivery in the UK, but it could not work during windy, rainy and unclear days. What are future plans for technology to solve this problem? What could be our expectation to see this technology in Afghanistan? I'm not sure which of you are able to speak on our drone delivery operations in the UK. I can't. I can maybe say a little bit generally speaking about, you know, using reinforcement learning, you know, to address that. Yeah, I probably can't comment specifically on our plans there. But I think the, you know, what we're seeing, you know, if you look at broader machine learning, you know, today a lot of the successes we've seen are narrow AI using supervised methods. So a lot of labeled data, in other words, I have XY pairs that, you know, that are known that I can train my model with. I think, you know, what we're seeing, and we've been kind of alluding to here with all the discussion of robotics and autonomous vehicles and systems is really the next wave of the reinforcement learning. That's really to, you know, for agents to act within an environment and actually learn and get better. And I think that's where we see, you know, reinforcement learning or RRL, as we call it, being applied. And I think drones are a great way to do that as they're interacting with their environment. They learn how to manage wind speeds and how to navigate, you know, within kind of changing environments. As to how we're actually applying that in our drones, that's not something I think we talked about. Okay. Thanks, Joe. Maybe I can add a couple things to that. So I think a big, an important area of research in today's time in artificial intelligence is how to deal with events, with things which we never had seen before. This is where reinforcement learning is critical. But reinforcement learning also reacts on things which are basically coming so that they learn for the next time to get better. So that's basically how all the AI then would learn while it is being applied. Now, a new, a very important use case of AI research is to be ready to react on situations which are not foreseen before, which I have never, never, never seen. And this is relevant for drones. This is for self-driving cars. This is for other use cases as well. I mean, these are the things which we are investing a lot of energy into, to, to be able to react on things which, which we're not seen before. Awesome. Okay. We have another question from the Lincoln Learning Center in Afghanistan. They ask, how relevant is artificial intelligence for school students? So what are some of the first subjects that people interested in AI should study in school? Math, I think that's number one. Math and, of course, computer science as well. I mean, if you have these things, you have a basis of on learning, machine learning in depth. You understand the theory. You can build new, new things, new approaches on how to do artificial intelligence. I think those are, in my view, the two most important things. Short, shortly after that, physics. I mean, physics will allow, teach you to apply the math in two situations to, to, to life, to what, what we see in, what we see in life. That's my personal opinion. I don't know my work is what they think. Yeah, I mean, I agree. I think coding, you know, I think we're seeing coding. I mean, I have a three year old. And, you know, I see like going into preschool, that literature school, even like basic concepts of coding are being taught. So I think, obviously, you know, learning a program language learning, you know, how to construct programs, you know, how logical structures work. I think math obviously is very important. I think in, if we're talking about AI in school, I think the biggest challenge that everyone has in AI is dealing with data. So, you know, having a little bit of data science, you know, having the experience to wrangle data and engineer data is actually going to be the most time consuming and challenging part. So having a little bit of experience there, I think is also key in addition. But yeah, learning Python or learning a language, and then being able to transform data and bring it into, into a model. Because, because honestly, the software tools these days are getting so good. And there's so much code available that, you know, I think really the, you know, if you're trying to solve a problem, you're going to spend most of the time trying to figure out what data you want to bring into your problem versus trying to, to code up an actual algorithm. Because algorithms are largely available out there in code that you, you can copy and paste and leverage. Yeah. And one thing that I would add on top of that is, you know, it's, you know, very, very true. Those are a lot of the concrete skills that folks need to learn. And if students get excited about those skills, they'll do well in AI. But just in general, getting, getting children really excited in STEM in general, and really in the science and technology area, you know, if you could get at a very young age, folks excited about STEM and science, technology, engineering, those sorts of things, I think that will be a catalyst to really help encourage kids. And what we're finding is in the AI space, we're seeing a lot of AI methods being applied in school at a very young age, like there's automatic reading applications to help teach kids learn to read, like, you know, Amazon Rapids, for an example, which uses poly to be able to teach kids how to read along and showing them how, how that technology really enables and getting them excited about STEM in general. Okay, thanks guys. A viewer in Getega Burundi notes that there is a fear that artificial intelligence displaces or replaces employees. How can we overcome or manage this fear, especially in countries where there is already a high rate of unemployment? Joe or Ben? Yeah, I could start. You know, one thing that I would mention is, you know, artificial intelligence and AI really helps enable folks. So it really empowers them to be able to do things more efficiently. You know, if you take a look at, you know, distribution, you know, Amazon distribution and how we have added robotics to really help folks in our distribution centers, but you know, really empower them to be able to do their job easier and faster and more effective. So, you know, I really see it as an enabler and making them be able to do their jobs very, very well. Yeah, I think I have to agree with Ben on this one. So I think, I mean, yes, there might be, there might be a shift of jobs from the one kind to another, but at the end of the day, AI and the internet is helping you to build something which used to be a costly endeavor to, and you can build that and distribute it all over the world. And I mean, the cost, I mean, today, today development, for example, doesn't have to be in, it doesn't have to be in Europe anymore, or it doesn't have to be in the US, or it doesn't, it can be all over the world. The internet is opening up basically the work which is related to AI can be done in any, in any place in the world. So I see it as well, like Ben, an enabler, rather than, rather than replacement of people or something like that. We are basically seeing another kind of industrial revolution, like when the machines came and helped us being able to do bigger things and so forth. We're seeing that with AI as well. Now we can go through more data, do more things at once, which we can do, which we were not able to do 10, 20 years ago, for example. This is, I mean, I feel that the opportunities are growing from year to year. I mean, not linearly, it's more exponential. Yeah, I want to add a couple things there. So I think, you know, to Son's point, I think AI is enabling things that we just couldn't do before. And I think, like for example, some of our customers like, like Wix, for example, who has a video platform, and they're able to comb through all the videos, the user-generated content videos to search for kind of not safe for work, you know, aspects. That's not something that they could frankly do at that volume of video before. You know, some millions of videos so if you can imagine hiring a number of people and having them sit there and watch videos and be able to flag videos, that's just not tractable as a problem. And the other trend that we see on the other end is human and loop AI. So I think that continues to trend. And as, you know, we don't want to, for example, you know, offload these, you know, these tasks and these workloads to AI completely because it doesn't make any sense, especially in things like medical imaging, when you know, you have a diagnosis, you know, you have maybe like, I just met with a doctor over at Stanford last week, you know, and this is in the radiology department, and they have a, you know, a stack, a huge stack of diagnoses that they go through every morning. So using AI to find the salient, you know, the top five or 10 right at the bathroom that are critical for that doctor to look at instead of combing through, say, hundreds of records and wasting hours and hours of time. I think it's this really, really nice use of AI and a time saver and a really great example of human loop. So a doctor is still making that diagnosis, but he's getting a lot of help from the AI to make it. Thanks for that. So moving on to the next question from the viewing group at Youth Network for Reform hosted by the embassy in Monrovia, Liberia. They ask, can AI be hacked? You know, it's so I mean, we've seen examples of AI, you know, sort of be hacked, you know, in the case of some of these is chatbots, you know, from Microsoft and Facebook. And so I mean, like anything, you know, that that learns from data, AI can be hacked, it can be if you feed it that data, you're going to get bad results. If you buy us your data, every data set has bias kind of inherently as the data sciences expert is procuring the data, you know, transforming the data and training the model, there's bias in that data. So you can, so in other words, yes, you can certainly hack an AI over them. Hassan, do you want to add? Yeah, sure. So I think this is this is another area of science which which we are very much interested in. How can you figure out how can you figure out from the data which is coming in? Is it good data or is it bad data? And how much reliability or how much do you rely on each set of data? And so this is a this is a component. The other thing is also what Joe earlier mentioned, human in the loop AI. I mean, chances are that if you have human in the loop AI, the system, the AI is able to learn more reliably. And you can filter out by using the machine and the human in combination in tandem. When this kind of situation is happening. But as Joe said, I mean, it's a it's, it's a technical system. Any technical system can be hacked the one way or the other. So we just need to make sure that a anything which we are developing in AI is ready to be figured this out as much as possible automatically. And have mechanisms in place to to circumvent that problem. We have another question from the viewing group at the Innovation Center in Pristina Kosovo. They ask, can AI be used for stock market prediction? If so, what companies do this? And which strategy is used such as SVM, Neutral Network or others? How can we get rich? Yeah, I mean, I'll start and everyone else could jump in. I mean, there's there's a number of companies and and they're not not necessarily the startups these days. A lot of the larger banks and people tell the robo advisors and cyber algorithms or cyber trading algorithms that are out there, you know, those are typically using an ensemble of predictive algorithms for stock trading. I mean, at the time, serious problems. So I would probably use something like an LSTM, the long short term memory architecture. So it's a recursive neural network that has has whole state has memory. So that's probably the algorithm. But as far as predictions, I mean, there's a number of companies out there, including the larger banks, hedge funds, pretty much everyone is using predictive algorithms to try and predict that where the stock market is going. Yeah, I think, I think the majority of the companies today are using that somewhere, the one way or the other. I mean, even if it's only to identify and enhance the human the human decision process. But in general, some kind of stock price prediction or portfolio analysis happens with using machine learning continuously. I think some of the companies are doing that since the mid 90s. So when I was teaching at the University of Aachen, I had a few master students working exactly in that area and building hidden Markov models at that time. On top of time series analysis to stock to predict where the stock where individual stocks are going or how much are certain stocks dependent on each other and which which stock is basically influencing which other stock and all of this. So it's a very it's actually an old area of research. It's not even new. But I think today's computers and data help us to potentially even do more these days. So I think that's why you see a lot of companies ramping even up their teams, their their decisions more and more towards using AI to analyze markets to make this to help make decisions and all of that. So online viewer sky asks, what is the possibility for governments to benefit from AI technology, such as in deep learning and neutral nets, especially in the areas of cybersecurity and managing government data? Ben, this seems like a good one for you. Yeah, absolutely. So, you know, we definitely see a lot of AI in the government space in the security area. Things like, you know, recognition of, you know, facial recognition is being adopted very, very quickly. Speech recognition in order to recognize not only what they're saying, but who people are in that area. What was the I apologize, what was the second part of that question? So, especially how are they using AI in the areas of cybersecurity and in data analytics or managing government data? Okay, yeah, thank you. So in cybersecurity, you know, cybersecurity is all about finding malicious patterns and being able to identify, you know, activity that is abnormal and at risk. And oftentimes in cybersecurity, you also have what Joe and Hassan was talking about human in a loop, being able to identify these patterns in order to help, you know, security centers and folks identify certain areas that they may need to look into. So many, many companies in the cybersecurity area are adopting machine learning within the government space in order to identify those signatures at the network level all the way up to, you know, malicious attacks for applications. I think this is more important today to use AI than it was ever before with the speeds of the networks being that fast with the data flowing so much faster today than it used to be with the amounts of connections which we have everywhere and so forth. So I think AI is critical to be able to go through all this. The era of rule-based approaches to cybersecurity is kind of over right. And I think, you know, in the past, we've had these, you know, virus, you know, software providers and they would just constantly update their rules and update their rules and oh, a new virus, it would be very, you know, backwards looking versus predicted an approach and being able to train a model and, you know, to the benchmark anomaly detection here is actually a critical aspect. So being able to figure out, you know, what is unusual and then be able to easily predict that and act quickly. That's not something that a human can do in a short period of time and there were milliseconds or faster, you know, to disseminate attacks. Gerardo from Peru is interested in more use cases of how AI is being applied in the healthcare industry for healthcare services. I think one of you had mentioned briefly about doctors using AI to go through records, diagnoses. Can Joe, maybe you, or Hassan, talk a little bit more about use cases of AI in healthcare? There's a bunch of really mundane things in healthcare that AI can help with. I think, you know, OCR, an Optical Character Recognition, we have a lot of kind of legacy documentation that's been built up over the years and, you know, being able to apply that to digitize and recognize all of that, all of those documents in order to make them searchable is something very basic and helpful in the healthcare space. You know, conversational bots, I think, are providing, you know, more timely interactions, you know, with patients. I think that's something that, you know, not having to wait on line for an hour to talk to somebody and be able to get information very quickly and move more towards an intent-based, you know, type of environment where I can ask very pointed questions, very detailed questions and actually start getting real answers instead of very basic questions. Those are probably two, obviously, medical imaging was a wide open field and something that a lot of people are trying to create systems around and gather data for. So I think the first time I used AI, or I was in touch with AI for healthcare is we were helping doctors interact in natural disaster zones with people who had a need for conversation across languages, which was not possible before. So Haitian, Creole people in Haiti did not necessarily speak English and people from the U.S. or from Europe came to support and the conversation was just not possible if not some human translator was there, but human translators were only a limited resource. So having the machine supporting with that was a very helpful use of machine learning and artificial intelligence in this context. Then as Joe said, I mean, at the end of the day, we can think of artificial intelligence being a way to build, like if you think about having an assistant, which is always with you, which knows where, how your health is looking, which is connected to you somehow and basically can figure out your heart rate and other information, if you think a little bit down into the future, that would be awesome that we can basically correlate how our health is with information, which is available to the doctors, for example, or which analysis with researchers, which was done before. And also things like, I mean, doctors have an issue that in different areas of research, there is sometimes so much research happening, so much, a doctor was telling me 6,000 publications a day happen in, I think it was radiology or it was cancer research and cancer research, 6,000 papers a day, which is doctor would need to be reading to be able to make informed decision and AI might be able to help consume that information, generate some kind of data, I mean, help the doctor to make decisions on whatever they need to make a decision on. So this is another use of AI in health care. The youth network for reform viewing group in Monrovia, Liberia asked, are there examples of AI helping the visually impaired businesses that are helping the visually impaired using AI, are any of you aware? I'll start with one really good example that I think is pretty exciting. So, you know, an organization in the UK, RNIB, they actually are using Polly to be able to deliver millions of book contents out to the folks with the blind. So they have this giant repository of books and they want to be able to be able to voice enable those for folks in the blind. It's a really large nonprofit in the UK, but you don't have to be a large organization to be able to do that sort of thing as well. You could be a small organization that is looking to be able to deliver results to folks with disabilities. You know, one really, another quick example that I'll just mention and then shift us on is, you know, state of Georgia, it's a state within the United States here, is using Alexa to be able to also deliver messaging and content out to their folks with disabilities. So, you know, with somebody with disabilities is able to sit in their house, ask about their driver's license, ask about all this source of information interacting with really public services within the system. Yeah, so my brother-in-law is visually impaired and he is very much in, I mean, in love with using Alexa for gathering information for other things. I mean, not that he cannot use a computer, but the computer is not available to you all the time. So, sometimes being able to communicate with Alexa and gathering information, not only about weather, but also about searching information in the internet, looking up Wikipedia for certain things and all of this is very good. But you can also consider using other technologies than just speech and text, text-to-speech as well, so speech recognition and text-to-speech. You can consider also thinking, you can think about applications where you can use recognition to identify certain information in a picture and getting it back to the visually impaired. This is, I mean, if you combine that with text-to-speech together, there is, I mean, there is a lot of applications you can think about. Saidi in Kandahar, Afghanistan asks, how can we use AI to automate experiences for our customers? Joe? There's some interesting, I think, trends around shopping that we're seeing. So, one of our customers Pinterest, as an example, they have their core competency as computer vision and being able to quickly snap photos and actually being relevant products or even that very product so that you can purchase it quickly. We do something similar at Amazon and be able to provide you the path to least resistance to things that you're interested in purchasing. I think that's probably one example I would say. That's a big one. I have to think about others. Yeah, that's probably, that obviously has big commercial impact. Nisar from the Lincoln Center in Kandahar, Afghanistan asks, do you have advice for choosing the best business partner when starting an AI-based business? What background or skill sets should they have? Hassan, what skills did you look for when choosing your partners for your business? This is, I mean, you can't compare when I had started with today. Today, for example, it depends on what kind of AI company you want to build. You might be interested in the people who understand the specific domain you care about. This is your more important resource or partner you need. For example, we were talking about a health-related AI. It's more important to have someone who understands what health is and where to get the data from and how to do business decisions on when the data is there. All of this is potentially more important than finding the math genius or the machine learning specialist and all of that because you can, instead of building everything from scratch, you can basically say, all right, I'm going to focus on the application level and build basically on something which exists. Let's say I can take Lex and I just need to define the business process, the business rules, the business decisions in some kind of we call it Lambda instance and then basically have the conversation between the user and the AI. I don't need to worry about how it's being recognized, how it is being understood and all of that. I just need to tell the AI what to understand. What do I care for whatever decision my AI company needs to do and then basically focus on that. Today's way I would find my partners is probably finding the domain specialists. The way I was doing it 20 years ago, I needed to find the machine learning specialists which today will help you anyway. Having a machine learning specialist even in a new endeavor will help you to make systems even better and all of this. I think this is lesser critical than having the person who understands the domain, understands the business, knows how to convert this domain knowledge and the business rules into something which then can be fed into a Lex, into a recognition and so forth and so forth. That's probably how I would start today. I think Nassan is hitting on an Uber theme that I've always had in engaging with startups. Over the past five or six years, I've seen startups that can survive for a while as kind of generalist AI platforms or services, but at some point they need to pick a domain or they need to have domain expertise in some vertical in order to survive and actually add value. They just get acquired through the talent. I think whether it's in healthcare, whether it's in financial services, whether it's in automotive, building an autonomous vehicle, I think having that domain expertise is where your competitive advantage comes. Because I've seen plenty of startups come and I'm going to go back to medical imaging because this is one I've seen a lot of kind of pretenders startups present to me. For example, my previous life in Intel where I was working closely with Intel Capital on investments and they would come and they would show great results overfitting on some dataset and say, look at us, we're getting 99% accuracy on these medical imaging, we're great. Then if you ask them what clinicians they're working with, what doctors, what is their background and they had really no background in the field. They didn't even know, for example, that if you're, you need to have FDA certification if you're not using human and loop AI and basic things about the space. I think to echo Christon, I think having domain expertise, picking your vertical, picking that domain and really getting experts in that field to complement your AI experts and AI usage is definitely critical for any startup. We have just under 10 minutes left and we have a lot more questions to try to get to. I'll ask that you keep your responses brief and that we just have one person answer each question for the remainder of the program if possible. The viewing group at the American Cultural Center in Algeria, they ask, what is the difference between AI and the Internet of Things? I would consider the Internet of Things or IoT as this network of devices, whether it's at the edge or connected, all connected to the cloud actually, but all producing data. Your phone is part of that IoT network. The sensors on your home, collecting your information, your smart meters as we have them in the value here, all those devices can consider IoT devices, even autonomous vehicles who are driving around our streets here in California. They're collecting data. You can consider those IoT devices. What is the difference between AI and IoT? Well, AI is really a tool that we're going to be applying or applying specifically machine learning to all that data in order to make predictions. In the smart meter examples, how do we predict energy usage in order to the best manage our grid? How do we take all that data from an autonomous vehicle? As we talked about before, learning about it not only is learning about things that are different in the environment, taking all that information, training models in order to better predict as vehicles drive through our streets. One is really, I would say, the infrastructure and all the devices, and the other is a tool that we're applying to make predictions about the data that's collected from that network of devices. Next question. A viewer named Mike asks, which industries might make the most positive change in society through adopting artificial intelligence and which industries are embracing AI now? Ben, are you back? No. Hasan? So, industries which are adopting AI quite a bit are the automotive industry. Anything where you communicate from machine to the human and the human to the machine as well as human to human, you see that adoption is happening quite a bit there. Also, where information is bigger than people who can basically deal with that information. I mean, news agencies are interested in AI. Of course, social media, I mean, they all have big teams of doing AI to figure out the interests of the people, commerce, of course. I mean, this is how Amazon all started with the AI that we need to understand the needs of the users and react on it. What else? Medical, I mean, definitely. It used to be that medicine, the health industry and the medical use of AI was being seen as evil and not liked very much. Today, it's exactly the opposite. The doctors are asking machine learning people, can you guys help us? There's so much data. We can't deal with that. So, there's that adopt. Actually, there's a lot of areas where adoption is taken place in areas which we're always ready for AI. Communication always was ready that AI needs to come in and help, but also areas which were resistant like medicine or something like that, which today, I mean, today's medicine wouldn't be available the way we have it today without AI at the end of the day. Habib Meemana is asking, is access to capital a significant barrier to using AI in a new business? What are some low-cost ways it can be implemented? I know you talked a lot about all the APIs and the data that's available, but are there any other ways that start-ups can lower their costs? I think the best way to lower the cost is utilize what is already available. Build something incremental on top of what exists already that saves you costs quite tremendously. I mean, this building basically the things from scratch is the most expensive thing and thousands of AI researchers built things out of what they did in research over the years into platforms. Utilize that. That will save you huge amounts. And then basically focus on gathering the data, which is relevant to what you need, gather that data, feed it in the right way to the AI platforms which you're talking about. Build basically the framework for decisions and then continue growing on that one. That basically will help you ramp up your application in very short time and then basically you gather data on the go. So Vanessa Rekundo on Twitter asks, what are some of the regulatory requirements to expect as one goes into this business? Could that be a challenge? A lot of the regulatory requirements that I face working with customers or work with customers with really relate to whatever area that domain is in. So in the healthcare, I think Joe and Hasan mentioned specific considerations where having human in the loop for certain diagnostics. So in the healthcare and the financial industry, folks like FINRA here in the U.S. processes, billions of messages every day in terms of identifying fraud. So really a general recommendation would be understanding the domain as was mentioned before and whatever requirements are done in that domain in order to fit those regulatory needs. Final question. Manusha in Kandy Sri Lanka asks, how will AI impact business operations and strategy in the next five years? Joe, do you want to take this final question? I think we're already seeing a lot of automation happen in marketing. I think in any kind of business, kind of critical task, we're seeing AI apply in HR systems, in payroll, in marketing. So I think that all of these areas that we consider business critical or business applications, I think there's some level of automation that's going to be permeating through them. So I'd say that kind of works. So before we conclude, I would like to ask each of our panelists, what would you say is the most important takeaway for our viewers? And please be brief in your response. And start with Hassan. Sure. So I think the most important thing which we talked about now a couple times in the conversation which we had is focus on domain, gathering the data, building basically systems not from scratch, but building something which is enriching today's state of the art is something which you guys should always focus on and consider doing. That way, you're enriching not only your things for yourself at the end of the day, you're helping with new ways, the humanity at the end of the day. Joe? Yeah, I think three quick things. I think one, understand what your AI value add is. In other words, where you're going to put your research scientists, don't try and solve problems that are already solved. So if you need face detection, don't go and find a large data set or create a large data set for faces and train a new algorithm for it. You can leverage something easily to do it. Second, find your data strategy. So how do you bootstrap your data, whether that's through taking your funding and applying it to human annotations or getting a large customer on board scraping the web, understand how you're going to apply your data and annotate it, and then quickly three, find that vertical that domain where you have an advantage. Just saying you're using AI on computer vision, for example, and providing an API like that. You're not going to move the naval. I think you need to find a vertical or somewhere where you have a competitive advantage and leverage that competitive advantage with domain expertise. And finally, Ben? Those are all great. The thing I would add is really just understanding the problem you're trying to solve. Don't try to start with just the data. Don't try to start with just data scientists. But what value add, what intelligence are you bringing into that particular domain and really understanding the problem so you could create shorter wins and really iterate on that very quickly. Thank you all. Unfortunately, we are at the end of the question and answer session for today's Tech Connect. Thank you so much to our wonderful panelists. You guys were awesome and everyone viewing today. You guys are also awesome. I want to give a special thanks to the hosts of the viewing groups around the globe. They include the Information Resource Center in Tunis, Tunisia, the American Corner in Tunis, the American Corner in Sefax, Tunisia, the American Cultural Center in Algeria, the Innovation Center in Kosovo, the American Corner in Tirana, Albania, Embassy Gabaron in Botswana, Embassy Abuja in Nigeria, Embassy New Delhi hosted at Nexus, Burama Incubator in Baku, Azerbaijan in partnership with the U.S. Embassy, Embassy Bujumbura Burundi, Youth Network for Reform at Embassy Monrovia, Liberia, Starlight Rwanda, Kigali Rwanda, Idea Centricity in Lahore, Pakistan, Lincoln Learning Centers in Kandahar, Mazar, Garda's, and Mi'amana in Afghanistan, the Jist Eye Hubs at Berry Tech in Beirut, Lebanon, and the Innovation Village in Kampala, Uganda, Embassy Kampala, Uganda, Embassy Maputo Mozambique, American Centers in Bouchi, Nigeria, and the American Center in Getega, Burundi. This wraps up our program for today. Please continue to send in questions in the chat space and on Twitter at hashtag JustTechConnect as our panelists will stick around for a few more minutes to answer them. I hope you enjoyed our discussion today.