AI Simplified: Sneakerhead Edition
AI Exposes Fakes
Today, we're talking about AI, and how it can help design streetwear, rap songs or even art.
What's dope is the guy who helped produce this track and this [clothing] line are both the same dude.
Robbie Barrat, 20 years old, is using artificial intelligence, deep learning, neural networks and to put it all in a box under the term GAN. That is the word of the day. What does it mean?
You know what? Let's get to that in a minute. First off, let's appreciate this hella dope line that he designed.
This ain't the only thing that he's done. He basically designed an algorithm as well, that helps him produce different types of designs.
One of the algorithms that he had put out on GitHub, well guess what? Someone used this algorithm to produce a piece of artwork that sold for over $400,000 on Christie's. But this isn't the only thing right?
Now politicians and even social media platforms are starting to ban deep fakes. So it's really becoming a topic of discussion.
I really want everyone out there even if you've never heard of AI, or don't understand it, don't understand what a GAN is, or neural networks. Don't let that throw you off the track. You've got to learn about this.
[Music generated by AI sings “I don’t know why”]
This AI may not know why it is doing what it's doing but it's based off of being fed, Kanye West lyrics. Really all of these networks have to learn from somewhere. Today I'm going to be giving you an easy way to understand the basics of a GAN.
Even if you don't know anything, even if you're not even into computers, I'm going to give you a really simple example, using everybody's favorite out there in the streetwear world, we're going to be using some sneakers. This I think, is the best way to explain it to anybody who still doesn't yet understand it.
Apologies out there to the experts, for oversimplification, but we must educate ourselves on this topic. If we want litigation, if we want to become entrepreneurs, if we want to be the next generation of technologists, and if we want that to start right here in the culture, then let's learn in a really simple way.
First off, sneakers are beloved. They have this sort of rarity to it so it creates a black market.
Now you have third parties like StockX, which in this case, is someone who is an expert and has seen so many authentic shoes that their ability to spot the fakes based off of micro details are better than anyone else. Both, the ways to authenticate the shoes, and Bootleggers’ techniques have gotten better over time.
It's kind of this, let's say, adversarial game against the Bootleggers and the “authenticators” to be able to spot the fakes. Once they’ve figures out that their fakes were spotted, then the fakes get better. And that's how we end up at a result like this.
In the beginning, the bootlegs are not really that good because they haven't yet been trained on how to make passable fakes. You really got to start at the ground level like nothing, right?
So, and just like these networks, you know, they are pretty new. This is sort of like the 8-bit era of what's happening right now. And it's advancing rapidly.
This concept was introduced by a former Google employee Ian Goodfellow in 2014, and kind of talked more out through 2016. These technologies have quickly gotten to a point where I would say, there's some videos out there that are really hard to tell what's real and what's fake. They get good by being trained. Training which involves producing thousands and thousands of designs.
Imagine a bootlegger being able to produce thousands of designs in seconds, to see what fools the sort of “authenticator”. StockX is what we'll call in this situation, the Discriminator, because they discriminate the details of the fake and then we'll also call the Bootleggers in this case, the Generators because they're generating the fakes.
Imagine they're competing against each other in an adversarial type of way or their “adversaries”.
They're sort of competing against each other in a game of cat and mouse and try to get better and better sort of like the bootlegging of money police and the fake money. They get better at making fake money, but then they get better at spotting them. That's how you end up with really incredible results.
These networks don't really know much about the context of things so it has to look at a lot of originals and that's fed by a person. So a person says, “Hey, this is a bunch of real data.”
Let's just say these are real Raf Simons collaborations. You have no fakes in here, so you have nothing to worry about study these. Also, here are thousands and thousands of additional examples. You may start with fewer examples but the whole point is, the more data it has, the better it gets.
In the beginning, the Bootleggers they're just going to throw out some trash and get snagged immediately, you know what I mean?
That's because they really haven't yet got good at producing the fake. When it looks at the real one, it may look at things like color or other reoccurring design details such as size and proportion. It may look at details of a brand’s logo which often clues to ensuring the sneaker is authentic. Keep in mind, this shoe in this as analogy is images, videos, audio, text, etc.
The Generator the bootlegger produces something like this and right off the top you I and Discriminator or StockX. They go “Hey, this is a fake” and you go “Why?” Well, first of all, this is a Nike. Second of all, it's got the wrong colors. It's, a white shoe. It's not even close to this dark shoe so it’s automatically rejected. This is then labeled as an unauthentic version of the Raf Simons collaboration with Adidas.
The Generator gets a little bit smarter and says, Well, what if it was an Adidas, right? And it was darker? Does it fool you? Okay. (No?) Still, in this case, the Discriminator is looking at a real shoe and going this is not the same sh*t. This is happening thousands and thousands of times per second. Imagine.
Okay, now it's getting a little bit more clued up during this training. It's like, “Alright, darker, maybe pops of color, Adidas…Does this work? No, this is still not it.” So even though these tricks may be getting better on the side of the Bootleggers or the Generators, it still is not generating something that is passable as authentic.
On top of that, the Discriminators are getting better at spotting face because it's like, “Oh, if you try to throw me something that's dark, or that is also Adidas branded, I know more details beyond that because I've seen your tricks before and I got better at it.” Boom!
You at home who may be an expert on streetwear and sneakers would know, there's no way that this YEEZY is an Adidas x Raf Simons even though you have the similar darker color and you have the Adidas branding. It's these types of details that it has to fine tune until it reaches this point right here.
Eventually, the Generator produces fakes that the Discriminator cannot tell the difference between the two. You'll see this cycle happening over and over. Where the fakes are starting to spill over into the real, and nobody can tell the difference. Sounds familiar right?
That's the goal with these GANs. It stands for Generative Adversarial Networks. So basically, the Generator is the Adversary to the Networks which is the Discriminator. They play a game against each other to get better and better.
They're obviously trying to do different things. The Discriminator is trying to label things as real or fake which is a different function from the Generator. The Generator is there just generate designs that eventually (or hopefully) become passable.
The reason why this is really important is coming up in politics. You see it coming up a lot and I want you to hear some audio, as well.
“Through the use of artificial intelligence, Coding Elite on YouTube has created a fake version of myself that can say or do anything. As these models improve, the potential for misuse is a scary thing indeed. In this new future, we must be vigilant and treat outrageous news with suspicion. I also advise subscribing to this channel, and leave your thoughts in the comments. Thank you. God bless you and God bless America” [AI Generated Deep Fake of President Donald Trump]
Basically, it's analyzing all of this information and able to interpret the nuance inside of these images until it gets better and better. Then you've now trained the network to do a seamless fake.
We've seen some laws go up when it comes to sort of these networks producing basically fake porn about celebrities or revenge. This has been outright banned.
You see social networks were now basically banning deep fakes because it could propagate basically fake news. The saying used to be “Believe, none of what you hear and half of what you see” and now you can't believe any of it, to be honest. It's a really interesting problem.
They're starting to create algorithms which could spot deep fakes because we feel that humans aren't gonna to be that good at that.
However, I don't want people to be outright afraid of this technology. Why? There’s a lot of amazing potential benefits, one that could save your life. It could possibly help us design better architecture and city structures. There's really no limit as to what this thing could possibly do.
It can even help design life-saving drugs. The reason for this, it can simulate thousands and thousands of experiments inside of the computer (with no real world consequences or limitations) and then propose the closest possible solution to solving the target problem.
If we go and we say “Artificial Intelligence, AI, it’s the Boogeyman! It's, bad.” The problem is, if we litigate against it, then we have an issue where we're not going to get those amazing designs which may help with efficiency in using our technologies to save energy.
There's really endless possibilities, anything that can be simulated, designed and improved on with different iterations and sketches, if you will. It's like prototyping thousands and thousands of possibilities automatically based off of something that you want, that are desirable. It really can change everything.
What if it could cure cancer? What if it could help us solve some of the world's biggest problems and that is where we stand right now. So there's this line in the sand. This is brand new. And the other thing is, is that this is happening in a few lines of code.
This really is something that you can learn if you put your mind to it, and you can use it to create all types of things. People have generated classical music. I think there's like a 24 hour death metal channel, which only plays death metal generated by AIs. Designs for clothing.
You are the decider of what these tools can help you design and I think Robbie put it well. I was trying to get a example of how this translates exactly, from a rough design into an actual line. And so I tweeted him and he responded. So thanks for that, Robbie. And I'm going to show you right here, I was trying to ask him about side by sides.
These are some of the crude designs that can come out of it. When I say crude, I mean, it's not full resolution designs. Then from there, designers can make improvements upon it.
He did make one very important fact, stand out to me even though this is something that I'm familiar with. I think this is something that needs to be said. There's a still a huge hand that is played by humans here. It's really about them deciding what is usable, what's not and what's improvable.
The point is that it's a tool that can offer you suggestions based off of any data that you decide, media, most importantly. That is what the opportunity is.
He says, “I'll probably post something side by side like this very soon, but please, the GAN doesn't do any designing. It's me making images with the GAN as a tool.” [Robbie Barrat]
I want you to quickly listen to this. This audio and video was generated by again and it's not real.
“We’re entering an era where our enemies can make it looks like anyone is saying anything at any point in time. Even if they would never say those things. So for instance they could have me say things like…I don’t know, Kill monger was right. Ben Carson is in a sunken place or how about this, simply, President Trump is a total and complete dipsh*t. Now, see, I would never say these things!” [AI Generated Deep Fake of President Barack Obama]
“At least not in a public address. But someone else would. Someone else like Jordan Peele. This is a dangerous time. Moving forward we need to be more vigilant with what we trust from the internet. It’s a time where we need to rely on trusted news sources.” [AI Generated Deep Fake of President Barack Obama]
This would fool you. If you were looking at this on Facebook, if you were looking this on Instagram. This would fool most people. Nobody’s going to go and check. This was done by Buzzfeed.
“It may sound basic, but how we move forward in this new era of information is going to be the difference between whether we survive or whether we become some kind of f*cked up dystopia. Thank you and stay woke b*tches!” "[AI Generated Deep Fake of President Barack Obama]
Stay woke! And that’s what we’re here about at The Digital Acid, is staying woke about what's going on.
Listen to this clip about Joe Rogan, this crazy!
“Ladies and Gentlemen, welcome to another episode of I Am Not Joe Rogan! No, seriously I am not. This is Joe FauxGAN speaking, an Artificial Intelligence created by Dessa. You guys know David Barstow? He's a New York Times reporter one of the most Pulitzer Prizes for journalism in history. He recently followed a crazy deep learning company called Dessa, as they created the world's first combination of realistic AI voice and video. Here's what they created…” [AI Generated Deep Fake of Joe Rogan]
“I’ve been thinking, and I realized, that it’s been almost ten years since the first episode of the podcast, that's f**king crazy! This show has become my life. That's why I decided to go out with a bang. So, on December 24, we’ll be doing our last episode, ever. That’s exactly ten years since the first episode came out. It’s been a good run and we’re all going out on top.” [AI Generated Deep Fake of Joe Rogan]
This is the issue that we're facing in 2020 and really, the whole decade ahead. It's only going to get better, exponentially better.
Already, in January of 2020 you have clothing, that is in part being generated (or at least suggestions to be completely accurate to Robbie's point). It's being used as a tool to design things in our everyday life.
The question is: How will you be a part of this revolution? The world needs designers. The world needs entrepreneurs and innovators. And my hope is that you'll be inspired by our show to go out there and just try it!
Sometimes it’s a few lines of code. Sometimes it's a YouTube tutorial. Sometimes it’s a bunch of YouTube tutorials. Sometimes it’s a message board. But you are the future of what will happen with these technologies.
Will all the opportunities be hoarded by the megacorps or will people within the culture have an option to win too?
That's what today's show is about.
GANs - Generative Adversarial Networks: a type of neural network, which is a type of AI.
Boom, there you go! Now run and tell that!
If you like what we discussed, and you want to share this with someone else, you can guarantee you’re recommending the best of street culture and futurism.
If you haven’t already, subscribe to our channel. We're always talking about future topics, now.
For more, make sure to subscribe to @thedigitalacid on YouTube, LinkedIn, Facebook, Twitter, and Instagram. www.TheDigitalAcid.com