7.1 What's coming next in AI? - Video Tutorials & Practice Problems
Video duration:
9m
Play a video:
<v ->All right, so what is next in AI?</v> What's coming? What's coming down the pike in the future? We've talked a lot about what exists today, but where are we headed? So conversational marketing is one of the big trends that I foresee. So customers will be able to talk or just type in plain language to an app, to a website, to a chat bot, to a bot on your phone, and the responses can be personalized to the customer and to the situation. And so all of the parts of these things are kind of in place, but we just haven't seen it widely distributed yet. And so you can already see some signs of conversational AI. So if you wanna try typing into Google, "Presidents' heads in rocks, where they at?" You can actually get an answer that makes sense. And that's something that a human being would understand, but it seems pretty odd that a computer would understand that, but it does. And so really, if you think about how you usually you search, where you're trying to think of special keywords to type in, presidents and maybe heads are probably the only things that you can think of. I mean, I'm pretty sure the Mount Rushmore page doesn't have the word rocks in it, right? So, I mean, this is just not exactly the way that Mount Rushmore would talk about it. They're not gonna talk about presidents' heads in rocks, but yet Google figured out that's what you meant. Not only did it figure that out, but it follows up with more conversation. If you look at the questions under it, it suggests other questions that it also knows the answer to that are related to those questions. And so this is the beginning of conversational marketing, where people can ask questions and get answers and maybe be prompted with more questions. That's really the start of a conversation. So what could a conversational website do? Let's just use that as an example, 'cause here's the reality today for website visitors. They search for things in Google, but when they get to the company's website, well, the site search maybe doesn't work that well. And they don't use voice interfaces very much, and it's hard to type a lot on the phone, and call centers are kind of annoying, but they do work. Human chat is a little less annoying, but doesn't quite work as well, and chat bots stink on ice. We really don't like them, at least not the ones that are around today. What about if you're the website owner? What's the reality for you? Well, you have a really low conversion rate because visitors can't find what they're looking for, and so then they go back to Google to do a search and maybe they find you, or maybe they find your competitor. 'Cause your site search really isn't working, and call centers are really expensive for voice and not quite as expensive for chat, but still pretty expensive, and chatbots are expensive to develop, and our customers really still don't like them. So that's really what the reality is today. What is the dream of conversational marketing? Well, the dream is for a conversational website. For the visitor, it says, hey, I asked for what I want and it appears. I don't have to navigate through the website. It doesn't matter what words I use. I can just ask like I'm talking to a person. What about for the website owner? Says my visitors find what they're looking for. I don't need to use expensive call centers. I don't have to change the processes we already have. It uses the website content we already have now. And so this is really what the dream is of conversational marketing, and AI has all of the pieces today to start to put this together in the next few years. What other trends can we look at? Well, here's one also about websites where we're trying to detect problems with conversion. So what are the kinds of problems that pages might have? Right now, the way we do these things is with AB testing. So you make changes to the page and you AB test by showing version A to some people and version B to other people. Is your change working? Does it make things better? Well, what if you didn't need to AB test? Or what if you didn't need to wait so long for AB test? If you have pages that not too many people go to, they could be really important pages, 'cause maybe they're for very expensive products, but you just don't get a lot of traffic to them. Might take a really long time to AB test the page. Well, how can AI help with that? Well, they might be able to predict what the problems are and changes that you propose to the page, and the way they do that is they would look at all of the other pages on your site and look at the features, the characteristics of those features, that are correlated to different types of metrics. For example, like bounce rate. Now again, remember the caution that we gave you. Correlation doesn't necessarily mean causation, and so you, the marketer, have to really examine these correlations and say, does it really make sense that that's causing a problem? But if it is, then you could use this to have AI look at your page and make a prediction of what its bounce rate would be and tell you all the reasons that it did that. So maybe the page loads very slowly, or maybe it doesn't have responsive design. So it's not gonna work very well on phones or maybe the content is longer than it should be, or maybe it doesn't have very high search relevance. These are all things that could cause problems, and if you, as a marketer, says, yeah, I think those really could be causes of problems, then your AI model can give you some tips and techniques for you to fix that page even before you publish it without having to go through the trouble of AB testing it. So what can happen from today where you're doing AB testing is you can go to kind of a continuous optimization approach where AI is always collecting data about the effectiveness of your pages and your marketing and how your customers respond to it, and then maybe it can even automatically make some improvements. The other thing that we mentioned earlier in the course that is still an emerging trend is explainable AI. So remember we talked about how explainable AI can actually tell you what it is that the AI used to make its decision? So instead of it just giving a prediction with no explanation, now what you can do is you can use explainable AI so it can give you reasons, give you a rationale, for why it made the decision that it did. Now again, today, explainable AI isn't as effective as the garden variety type that we're used to, so that's why it's not in great usage yet, but that will change. Explainable AI will get better. Maybe it won't ever approach the black box AI, but it will get to the point where it's worth it in many, many situations, and remember, if you have ethical dilemmas now, you may really need this. But one of the good things about explainable AI is because it reveals its rationale, that might give you clues about how to make your model better, and so explainable AI might help improve your model even if the technology isn't quite as accurate as the non-explainable kind. Because it helps you improve the model, eventually you become more accurate than the other because you know what to do to make things better. We mentioned earlier how you need to work with your data scientists to choose the features that the AI ought to examine as it's putting together the models. And so that's true today, but at some point, you're not going to necessarily need to do that by hand. Someday, the machine can do that for us because things are getting faster and faster. So at some point, the machine might be able to look at all the possible features and be able to figure out what the salient ones are without a person having to do that for it. Because you remember, we talked about Moore's Law, where computer power is constantly accelerating. It'll continue to get faster. It'll continue to get cheaper. We're continuing to collect and analyze more and more and more data. That's gonna make AI better all by itself. And one of the biggest changes is actually happening right under our noses. One of the biggest changes is more and more software products are getting AI embedded into them, and so we've talked a lot about how you can solve problems with AI in this course, and a lot of those problems will need a data scientist. They will need you to do some specialized or maybe even custom work, and certainly some data-driven work, but what's really starting to happen is that the models are being embedded into products all over the place. And so just by focusing on which of your software products in your MarTech stack have AI in them and how you can take advantage of that, that's one of the biggest trends of all is that AI is going to be inside everything for you to take advantage of.