1.1 What is Artificial Intelligence? - Video Tutorials & Practice Problems

<v ->So, Gradient Boost,</v> Support Vector Machine. I know. K-Nearest Neighbor. All right, is this what you were hoping to find out in this class? 'Cause if it was, not happening. 'Cause this isn't what you need to know? This is what a data scientist needs to know. This is what an AI expert needs to know. And guess what? If you have people talking to you, using all these words and trying to make you feel stupid, these are the people to run away from, because they're just trying to make you feel like you don't know what you're talking about, so you're gonna listen to anything that they say. That is not what you want. What you want to understand is, what do you do with artificial intelligence? You don't wanna understand all these techniques. You never need to know them. And you know how I can prove that to you? I've been working in AI for, jeez, 40 years. I don't know what half of these things are, 'cause I don't need to. That isn't my job. It's not your job either. So let's talk about what your job really is. So I'm gonna start by asking a really weird question. So what color is the sky? Now why am I asking that question? Well, I wanna make a point with it. So let's start by just saying, what if a four year old child asked you that question? Do you think you'd know the answer? I'm hoping. Yes. I'm hoping you would know that what she's asking you is, what's the name of that color I'm looking at? And I think most of us would be able to quickly say, "Hey, that's blue. "Kid, that's blue." You're all set. Now, suppose the same question was asked by a meteorologist, "What color is the sky?" Now, I don't know about you, but I would be a lot less confident answering that question. So, because what we know, even if we don't know the answer that the meteorologist is looking for, what we do know is that the meteorologist is saying, "Hey, how does the color of the sky help us predict storms?" So, Hm, same question. One we could confidently answer, the same exact question, not so confidence. And suppose what he said, it's an astrophysics professor asking us, "What color is the sky?" Well, I definitely don't know what he's talking about here. So what he's really asking is, why does it seem blue to us? And so, what have we got here? We've got different answers but the same question. Now, even though we might not have known the answer to the question each time, we definitely know it was different answers to the same question. And my question to you is, how did you know that these were different answers for the same question? And the reason is, 'cause you're intelligent. The reason is context. The reason is that you knew that who asked the question is just as important as what the question is. And the reason I took you through that is because I want you to think about what AI is. So if I asked you, what is AI? How would you describe it to me? Well, you might say, "It's the ability of a computer to perform tasks "that usually require the intelligence of a human being." And if you did say that, well, you'd be as smart as Wikipedia, because this has to be true, was in Wikipedia. If you know what the problem is with this definition it actually punched the hardest part, because it really is just giving you a definition of what artificial means. The hard part is, what is intelligence? That's really the hard part. So, intelligence is actually what you were exhibiting when you knew that there were three different answers to the exact same question, that's intelligence. And so, what we wanna figure out now is, how do we really define artificial intelligence? How do we really decide what thinking is? Because it's kind of interesting to say that artificial intelligence is about a computer exhibiting intelligence that is indistinguishable from a human being. This is actually called the Turing test, and it's named after a famous computer scientist, named Alan Turing. He devised this test back to 1950, and he actually had this very interesting device that could actually ask questions. If you remember the ELIZA program, it's something that maybe some people might know, basically it almost acted like a psychotherapist, whatever you would say, it would say, "Well, how do you feel about that? "And what kind of things does that bring up for you? "And why did you say that?" And we'll just ask sets of questions. And it wasn't intelligent at all, it didn't know anything. It just knew how to parrot and mimic conversation. But, back in 1950, they were very optimistic about how quickly computers would be able to think. And so they believed that something, they called "General Artificial Intelligence," which is basically the idea that a computer can do anything a person could do. They thought it would be achieved in about 20, 25 years, maybe by the mid '70s. Just footnote, that didn't happen. So what really happened is that by the time they got to the '70s, everybody was really tired of all these crazy people talking about AI. So what happened was that a series of what we call AI winters and AI summers. And so the Winters are when everybody kind of lost faith, nobody was investing at anymore. And then they would come up with a new way of thinking about it. And then you'd have a summer, where everybody was throwing money at it again, and everybody was very optimistic. And so the early systems really couldn't scale. They couldn't really do much of anything that was useful. But by the 1980s, some smaller problems started to be solved with things they called Expert Systems. So what they did is they kind of... They did what they call shrinking the domain. They made things so much smaller, what the problem was, that you could get a computer that actually solved it. But they found out Expert Systems had their limits too. And in the late 1980s, they had another AI winter. In the 1990s, suddenly things started to improve, that when they define the problems narrowly enough, now all of a sudden they started to be able to solve them. But here's the thing. This is a chart that you often see analysts show. I think Gartner shows something similar to that. What's happening is that, there's all sorts of expectations, and what happens is that people suddenly say to themselves, "Hey, we are disappointed. "We have not met any of those expectations." So something happens in technology. They have all sorts of inflated expectations. Then they hit what they call the Trough of Disillusionment. And then, they focus on coming back out and trying to figure out that, "Hey, it wasn't as good as we thought it was gonna be, "but we did figure out it was useful for something. And this is what you normally run into when you hear people talking about having another dollop of 5G Blockchain AI, or all sorts of stuff, where people think that they have all sorts of different things they throw together just because they're gonna raise money from a venture capitalist. We're not gonna talk about any of that stuff here. That stuff is the hype. We're gonna talk about the reality. Because practical applications actually show the value of AI. You identify a business problem, you understand what the value is of solving the problem, and usually, you need to have some kind of data. And you need to have good data, or at least good enough data. Has to be good enough that the AI is actually getting some signal out of it. So what we wanna think about is there's actually two types of artificial intelligence. So the first type is what we talked about before, General AI, or Artificial General Intelligence, that's what they were talking about in the early '50s. And there's a few crazy people that still talk about it. I think they're crazy. But one day, those crazy people turn out to be right. And what they are saying, is that a computer is someday going to be able to do just about anything a person can do. And we're nowhere near that in my opinion. And someday leaves a lot of room for how long that can take. So I'm pretty sure there'll be right, as long as they keep saying someday. So when they try and put a year on it, that it starts to be a little harder. Now, what we really wanna talk about is what the state of the art is today, which is called Narrow AI or Weak AI. And basically what it means, is that you can solve a problem that's in a small domain. So it's a single problem. So back in the 1990s, Beep Blue beat Gary Kasparov at chess. Watson beat Ken Jennings at Jeopardy! About 10 years ago. And so what's going on there is they picked a very narrow problem and they were able to solve it. They were able to do it better than a human being can. And that's what we really care about. What are the things that AI can do in marketing better than a human being can do it? That's what we really wanna know. That's what we're gonna spend all of our time talking about. We don't care about when AI is going to take over for human beings. I don't think any of us will be around then. It's not gonna worry about it. And so what you really need to think about is, how do you use AI to do better in your job today in marketing? Now, there's lots of difference AI that is in a solution. So if we just take one type of AI, we think about chatbots. Look at all the different types of AI techniques that are just involved in a chatbot. There's voice recognition. So they might need to understand someone speaking. There's speech-to-text. So they have to be able to take that speech, turn it into words, typed words, you can think about them, so that those words can then be analyzed by another form of AI called natural language processing. And so, NLP actually understands, we'll would do that in quotes, "understands" what it is that you're talking about. Doesn't really understand, but it recognizes patterns and knows how to respond to them. Then what it might do, is question answering, 'cause with a chatbot, you might be asking a question. Now, when you're asking a question, then we need to figure out how to get the answer to that question. Now, once the computer has the answer to the question, now we've got to reverse the whole process. Now, we have to use text generation. We have to take the answer, that's basically in typed form, and turn it into something that seems like an answer, because we may have found the answer in a paragraph on some webpage or in a book somewhere, but now we have to turn it around, and turn it into texts that sounds like the right linguistic answer to the exact question that you asked. And then the final step is we turn it back into speech, and it says it back to you. And so that's really an example of a very sophisticated type of AI. So I hope that this was helpful to you as your introduction to what AI is.