What you will learn?
The Future of Growth:
Demystifying Artificial Intelligence and Bots for Marketing
Alright. Up next, we’ve got Adelyn Zhou. Adelyn is the CMO of a company called TOPBOTS. And you know, this morning’s telly was talking about throwing AI in front of everything that you’ve got. Seems like it’s a bad idea if it’s not actually AI, but there is a pretty cool place for artificial intelligence and bots in growth marketing.
So for 15 minutes, we’re going to do this rapid fire session where Adeline is going to tell us what the future of it is and try to demystify it, right?
Okay, welcome. Come on.
Alright. Thank you. Can you guys hear me? Okay.
Well, I guess I guess Steffi and people have already given me an introduction how you guys all realize, you know, it’s sold you on AI is not the future and I’m here to tell you that it is or at least convince you to give it another perspective.
So, background. I’m at TOPBOTS. TOPBOTS is an advisory and education firm. We work with the large Fortune 500 companies to educate them on everything that’s happening with applied AI. You can tweet at me, follow me. If you’re at all interested in what the future has to hold then follow me on @adelynzhou on Twitter.
A little bit about myself. Even though I’m talking about AI and BOTS, my heart is, well, is with growth marketing. I was previously leading marketing at different companies including Nextdoor, Eventbrite and also at Amazon. But over the last couple years, I’ve just seen a huge trend with a wave of artificial intelligence and bots. And hey, I found that there is a huge opportunity to market using these new technologies and hence have kind of skewed my career towards this direction and hopefully by the end of this presentation, you too will be convinced on adding at least some AI into your marketings.
This. If you weren’t on the PC wave, you lost business. If you weren’t on internet, lost business. Social media, same story. And most recently, mobile. In all of those cases, you needed to be one of the first companies and brands to be on these platforms in order to get to disproportionate ROI. Yet the difference is by, shown by this chart, if human progress and technology has evolved over time on the linear part of this graph, we, right now, are that little stick figure on that corner. We are at a huge cusp, a change that is going to be ushered in by artificial intelligence.
What is AI?
So first, what is AI?
Automation. Automation has been around since the industrial revolution. It’s been here for over a hundred years. Artificial intelligence though, is only one small segment of that larger automation piece. And to help you understand what exactly is AI and what is it not, because there’s tons of hype right now about everyone claiming that they’re an AI startup, you’ll see how we’ll describe the differences between AI and what it’s not and you soon you’ll be able to recognize what is legit and what is fake.
So to do so, we at TOPBOTS created the machine intelligence continuum. At the basics, you have systems that act. A system that act is very simple. It’s very rule-based usually by decision tree. If you do a, then say b. If someone says c, then you answer with d. An example is with cruise control. If you’re using cruise control, you know that every time your speed falls below a certain level, your cruise control will automatically apply the accelerometer. That is a simple system. And at no point would you be driving your car putting on cruise control and say, “Whoo! Self-driving car!” That’s cruise control, right? You know, that is not AI. So that’s an example of a system that acts.
Next, you have a system that thinks. Tic-tac-toe. We’ve all played this, right? When we were back in the days. We’ll tic-tac-toe is a super super simple game. And as a result, both humans and computers can play this game with a hundred percent accuracy every single time.
I mean, if you remember back to those days when you’re like six years old, you knew that you needed to start with the X in a certain corner. And that if you started with a, you know, on the second step then you were bound to lose. And so as a human and a computer, you can map out with a hundred percent likelihood what a tic-tac-toe game will be.
And so in that case a tic-tac-toe game is really more like a system that acts. Whereas instead, on a chess game, because there are over 400 different opening moves, chess is too complicated to map out an exact decision tree. And in that case, you have more called boards that are tuned to a certain waiting.
And so this is what IBM did with Watson when they beat Kasparov in deep blue. They took every single chess board game that was possible and they waited on the likeness of that board to win. And so in that case that was more, it was a system that thinks because they were not a hundred percent sure, and you used probability and statistics.
So a system that thinks is one system that predicts. What I mean by that is you take what is known and you use that to define what is unknown. We use traditional probabilistic curves to help you determine what the future will be.
Systems that learn are the next phase in the machine intelligence continuum and this is actually where things actually become artificial intelligence. Systems that act, systems that think, actually aren’t AI. If you ask a lot of Watson programmers, you’re like, was deep blue an artificial intelligence machine? Most of them say no because that was based on probabilities and it wasn’t actually thinking and generating.
There’s a huge difference now with systems that learn. Systems that learn is what people are calling AI and that’s because you have this new type of technology called deep learning. The difference with deep learning and the old algorithms is that with more data, your performance increases. And so for example is, back last year Google took these games, Atari games and told a computer, “Hey, go play these games. Go win.”
In the old days, right, in chess, you teach it. This is a pawn, this is a rook and this is how you win and this is what each board is weighed at. Now in this one, Google did not tell the computer at all what, in this case, and break out, what a brick was, what a paddle was, or what a ball was. The only thing it told the computer was maximize your score. And so the computer without any direction learned how to play a game. Just like you when you were a kid, you learned without really knowing the rules. It did trial and error and over time, this over a 240 minutes, the computer program was able to become so intelligent that it even figured out that, that trick, do you guys remember in break out where you go into the corner and you make a tunnel and then the ball goes on the top and it just bounces back and forth, back and forth, and then you win. Well the computer figure that out after 240 minutes of gameplay without anyone teaching it. So it’s learning on its own.
Why now? Why are we talking about AI now?
AI has actually been around since the 1960s, but we’re talking about it now because we’re swimming in oceans of data. Every one of you as you’re using your mobile phones, which I’m glad not that many viewers are because you’re enthralled by my presentation, are giving different signals up into the cloud. Every time you tweet at me today or post something on Facebook, data is going up. And now sensors from your Lexus to your thermostats and your cars. All of that is collecting data.
And on top of that, there are now these large image sets such as this one image that has over 14 million images that are tagged. In imagenet, you have images that are so detailed in their tagging such as someone with a hot dog that has relish versus a hot dog with relish and ketchup and a hot dog with relish and ketchup and mustard. And computers are using this data to train their systems to understand and to be able to identify photos. I mean, you ever wonder, right? How does Facebook know that that’s your friend. It’s using data sets like that and the data sets of your friends to teach it to learn. This is a face. This is a nose. This is the eye and this is your friend John.
The second major change why we’re talking about AI is processing capabilities. This is a picture of a perceptron from the 1960s. It was huge! It was like 6 foot, 7 foot tall and big and boxy. And now this small device called a GPU which is very made popular by gamers, any gamers here? Yeah. So you guys recognize these right? Well, the gaming industry has pushed forward artificial intelligence because gamers are always chasing the latest graphical advancements, companies such as Nvidia have followed suit. And when Nvidia was initially creating these GPUs, they realized that these GPU machines can be trained to do artificial intelligence processing and as a result made this man as super super wealthy. Last year, Nvidia was the second highest grossing stock, faster growing than Amazon, Facebook, Apple, Google, any of those combined. So those, so if you invested in that, you were pretty lucky and it’s also made this man one of the richest as well.
Huh, it’s not progressing. Huh. Oh, okay. There you go.
So those are examples of systems that learn. So learning is happening when a computer system doesn’t have something programmed and it’s learning on its own.
The next state in machine intelligence continuum is what we call systems that create. And this is where it gets really fun.
Of course, we have a computer programmer who had a lot of time on his hands. He decided to feed his machine learning algorithm 14 million lines of romance novels. And then he gave a system this image and told it to caption it. This is what the computer created. “He was a shirtless man in the back of his mind and I let out a curse as he leaned over to kiss me on the shoulder. He wanted to strangle me considering the beautiful boy I’d become wearing his boxers.” That, my friend is a romantic snippet created completely by a computer that had indulged itself with 14 million lines of romance books.
This is an example from Sony. You guys Google this and you find on YouTube called Daddy’s car. It sounds eerily like the Beatles because it was a song that was created based on past Beatle lyrics and music. So if you have a chance, highly recommend you listen to that.
And this creepy looking image. What is this? Well this my friends is a photo that was created by Google’s dream net and sold for over $8,000 making it more successful as an artist than many humans.
And then finally what else can machine intelligence do? It can identify things like your friends as well as tell the difference between a muffin and a Chihuahua. Yeah. It sounds pretty good. Right? Well a machine can also do that now.
What else can it do? It can take sketches of a purse and fill it out and make it look real. It can do mathematical equations such as a man with glasses, take out a man without glasses, add a woman with glasses, and you get a woman with glasses. That’s not using Photoshop. So the difference between that is you’re not going in cutting out the glasses. You’re just telling a machine giving it photos and it miraculously understands. What is the face of a man? What is the face of woman? What are glasses and what are not? And it will create this spontaneously.
You can do things such as take a Monet photo, Monet painting and make it into a photograph. There were no photographs back when Monet was round, but you can now create in that likeliness.
You can swap zebras and horses and apples and oranges and spring to winter all using machine intelligence.
And this is one of my favorites. This is actually very recently from a university paper. You have these creepy looking faces, right? Like what are those? They kind of look like faces, their face except that they look a little strange and then on the right hand side you have much more beautiful faces. Those faces were all computer-generated. So if you’re a model looking at this, I was like, oh, I thought my career was safe, you know in the modeling industry. Nope. You can be replaced.
Just example what’s happening over time. Initially you have this creepy looking skeleton but as machine intelligence gets better, you have actually really hot girl. I’m pretty sure if she put her on Tinder, she’d get a lot of yes matches.
And so if you think about marketing and what can possibly happen is that what you had back in the days, the Dove commercials, you had to get a person of every single ethnicity to pose in your photos, and now you can use a single click of a button and generate an Asian, African American, a Latino, and miraculously change your marketing campaign from one country to another from one viewer to another. If you knew that the person looking at your ad was African-American, you can rapidly change that photo to match their perception.
So those are all examples of systems that create. Systems that create spontaneously without direction and creating new images, new videos, new sounds, and actually even marketing copy. Coca-Cola these days are now using some of these AI tools to help generate marketing copy. You have news reports that are completely generated by AI and you probably are reading them without even realizing. So it’s definitely here already.
And the last step is systems that adapt. Systems that adapt are what is known as general intelligence. General intelligence is what we as humans have. It’s ability to jump from conversation to conversation, from topic to topic and you can talk about the weather one minute and then asked about you know, how the game was. That is general intelligence and is also what’s made popular by movies such as her, ex machina, and of course, the Terminator.
Many times you might hear the word called the singularity and that is the point that people claim that these computers, these AI systems, will be smarter than us humans. And what will happen at that point, we actually don’t know. You have some people who are very pro such as, the people such as Mark Zuckerberg and Gary, Sergey Brin and the different teams at Google about AI.
And then you have definitely naysayers and even including Bill Gates and Elon Musk warning us about the time when AI can surpass humanity. So to that point, we don’t know. But until then, as marketers, we should use all these different things to reach your users.
So this is the machine intelligence continuum. We talked about systems that act, think, learn, create and adapt. And hopefully now you can tell the difference between what is AI, which is the learn, create and adapt part, and what is not, which is the act and think.
We, if also, if you’re interested, created entire landscapes of all the different companies that are playing, the AI companies that are playing this field, so we have this, tons and tons of companies who are cutting me out using AI to change every function, including sales, marketing, operations and customer support, which is some of the top top ones.
Bots, you might have heard about, being all the rage these days. You may have also read how they also suck. Well the reason they suck a lot of times is because they’re not AI. They are really simple systems that act, which you now understand is not AI. But that being said, tons and tons of brands are now jumping on this bandwagon to use bots as a way to connect with users. If users are on Messenger, bots are the way that you can connect with them.
So if you want to read more about the different brands, we’ve profiled over a hundred different companies using bots in creative ways to reach users.
And so with that, if you have any questions, feel free, and hope now you’re more enlightened about artificial intelligence. Thank you.
Adelyn Zhou, everybody. Great job.