2.1 How is AI different from traditional software?
2: How AI Works
2.1 How is AI different from traditional software? - Video Tutorials & Practice Problems
Video duration:
7m
Play a video:
Video transcript
<v ->So the joke is if you know how it works, it's not AI.</v> Because traditional programs, we think we know how they work. They get instructions. There's a programmer who puts together the program and that software developer is basically doing something that gives instructions to the computer. So you can think of it as do what I say. So some omniscient being is ordering the computer, this is exactly how you do it. First do this, then do this. If this happens, do this. And so it's a set of instructions. They may have conditions associated with them, but you can kinda follow the logic of it. If someone explained it to you step-by-step, even though you're not a programmer, you would understand how it works. In fact, you might have find on the job sometimes you're the one telling the programmer how it needs to work. Well, first, this is what happens and then we have to do this and then you can do this calculation, then you check to see which one's the highest. You could even give instructions to a programmer that's step-by-step explain how a traditional software program oughta work. An AI model is very different from that. There are no instructions. Instead, there are examples. So you can think of it as being one big instruction that says, hey, I want you to deliver a certain result. So I want you to predict, for example, which of these pictures are actually stop signs, and I'm not gonna tell you how to do it. I'm not gonna say, hey, they're red. They have an octagonal shape. They have white letters. We're not gonna give you instructions for how to recognize them. We just are gonna give you lots of examples of stop signs. And then eventually, you should be able to recognize them. And so we're basically saying, deliver the result of recognizing a stop sign. And that's very different from a traditional program. And the truth is we don't know how the computer figured it out. What pattern did they figure out that said that's a stop sign and that isn't a stop sign? We don't really know. It's a black box. It's something that we don't understand what's inside. And so how does AI just deliver a result? So let's give an example. So in Photoshop, there are dozens of dials that you can turn until you like the way the photo looks. Now, I say you can do it, but maybe you can't do it. I know I can't do it. I don't know how to operate Photoshop. I can play around with it and put something on there and say do this, do that. But by the time I'm done, it usually looks worse 'cause I'm not that smart about Photoshop. You need to kind of be an expert in solving the problem to know what to do. So even though you can instruct Photoshop to say basically do what I say, you have to know what to do, so it takes a lot of expertise, but let's try to contrast that with something like Instagram filters where I don't really know why it looks better when I flip to this filter, but I think it does, so I'm gonna keep that one. That's closer to how AI works when it says deliver result. You see an awful lot of different things. And then at some point you decide, hey, I like that one. That's the one I'm going to do. That's the result I was looking for. And so you don't need a lot of expertise in order to be able to do that, but the computer program needs to be a lot smarter because it needs to know how to deliver lots of different kinds of results. And so one of the things that happens when you're using AI is you have very different characteristic of those AI models than the traditional programs. With an AI model, one of the reasons that they say that it's about learning is because the results actually get better the longer that the AI runs. So the more information you give it, the more data you give it, the more examples you give it of the right answers, the more you're going to see that the AI is improving, and that's very different from a traditional program. In fact, if you give a traditional program the same input and it gives you a different output than it did before, you know what that's called? A bug. It's actually bad. We wanna make sure that you get identical outcomes for the same input for a traditional program. It has to work that way because it's logically going step-by-step. But an AI model is different. If you give your AI model the same input that you gave it a month ago, if you've continued to train it and improve it, it might give you a better answer than it gave you a month ago because it's smarter now. It has you can say learned. And so this is more the way a person works. So if you think about it, if I gave you the same task to do 100 times in a row, by the 100th time, you'd probably would be a little more accurate. You might be a little faster at it. You might make fewer mistakes. You might not have to stop and start over again as many times because you're kind of in the groove. You've really learned how to do that task. AI is more like that, which is why they call it learning. Now the truth is AI doesn't really learn anything. It's not learning the way we are where it's drawing conclusions and generalizing things. It's really just developing a better algorithm that is more accurate, but it looks like learning to us and so that's why the moniker learning has really stuck, but AI keeps changing. The bar keeps moving. So I remember in 1989, I worked on a project that did the first linguistic search engine in 33 languages. So for example, if you searched for the word mouse, it also could find mice because it knew it was the same word. That was considered state-of-the-art AI back in 1989, but now no one cares. No one thinks that's AI. In fact, even in 1998 when Google introduced spelling correction, nobody really thought of that as AI. And so what happens with AI is that we tend as human beings not to be looking at some kind of technical definition of what it is, but really it's whatever surprised us lately, whatever a computer did that we didn't think it could do, that's AI. And so one of the things I want you to think about is it's a lot less important what we think of as being AI or not AI. It's a lot more important for us to choose a problem that we can use these techniques to solve because solving problems always has business value whether it's AI or not. Now, we're gonna emphasize all the techniques you can use in AI, but the truth is that what we really want you to focus on is picking a problem that is really valuable to solve and we'll help you use AI to do it whether or not people think that the AI is clever enough to be called AI today or two years from now. So that's really what your focus needs to be.