AI Just Killed YouTube

Let's talk about AI and machine learning for a little bit. So we're essentially going to be using AI to create a Mr. Beast style YouTube video. Artificial intelligence could become a complete disaster for YouTube. I'm going to try to start a YouTube channel using only AI. AI has no soul, but it also might provide some unique benefits. The future is here. Content creation in general is going to change dramatically over the next few years. But there is one thing that's going to follow this conversational AI thing everywhere it goes. Will artificial intelligence destroy YouTube? Let's begin by going over the biggest problems before explaining exactly how this will impact YouTube and the millions of creators who upload videos to the platform. This all started when a TikToker named Ghostwriter977 uploaded an AI generated song featuring Drake and the Weeknd. People had been using AI to create covers and mashups of popular songs for months, but this was the first time a new song really sounded like it could have been made by a famous artist. Heart on My Sleeve marked a turning point. Play it for any casual Drake fan and they would never know it was made by AI. And that has some massive implications. Everyone immediately jumped to the most extreme conclusion. Will this destroy the music industry as we know it? See, the situation isn't actually all that dangerous for big creators like Drake. In fact, he might wind up benefiting the most from these new AI technologies. Let's start by analyzing the song Hard On My Sleeve, because a lot of people don't really understand what's going on here. Don't get me wrong, the song sounds great, but while the media is reporting that it was entirely generated by artificial intelligence, there's a lot more to it than that. First, the beat was created by a human. Using industry standard software. They picked out the instrumentals and laid down the notes that they thought sounded appropriate. Then the lyrics were likely written, at least in part by a human. ChatGPT is getting really good at writing lyrics, but it still needs a lot of guidance in order to hit specific topics. It's not enough to just ask ChatGPT to write a banger Drake verse. The result will be far too generic. I came out with my ex like Celine, not a flat pumping Justin Beer, but I fear. But integrating those specific mentions of Selena Gomez and Justin Bieber was probably the idea of Ghostwriter 977. After all, he claims to be a ghostwriter. This is his job. But what about the Drake voice? That's what made the song really unbelievable after all. Well, the voice wasn't actually entirely generated by AI. There are services working on being able to generate voices just from text input, but this song used a different method called style transfer. It's more like a deep fake than a mid journey style image generation. I am not Morgan Freeman and what you see is not real. Someone who can probably do a pretty good Drake impression recorded the lyrics and then used AI to transfer the style of Drake's voice onto their recording. When you're making a deepfake, it helps to have someone who at least has the same face shape as the person you're faking. So when corridor crew wanted to create convincing deepfakes of Tom Cruise and Tupac, they hired look alikes to serve as the underlying reference footage and audio. Deepfakes work the same way. Better inputs yield better outputs. But that's almost beside the point. Even though it still clearly took a lot of human effort to make this song, it was way cheaper than paying Drake to record the verse. If anyone can make a Drake song in their bedroom with just a few hours of work, clearly the music industry is going to change dramatically, right? Well, yes and no. We've actually been here before. Back in 1999, the MP3 format allowed people to share music with each other. And Napster used the Internet to push that to maximum scale. The entire music industry had been designed around selling CDs and the shift away from these physical discs basically destroyed the core business model of the music industry. At the time, the Napster era was exciting. If you are a broke kid who just wanted to listen to free music, it's free and it's easy and you know it's wrong. Yeah, but a lot of people do it. Pretty much everybody does. But everyone in the music industry knew that it would be hugely disruptive in the long term. Business model built on infringement is not only morally wrong, but legally wrong. If artists could no longer make money selling music, they'd stop making music entirely. If no one was paying for music, artists simply couldn't afford to buy instruments, run studios, hire producers, or do any of the other important things required to make great music. Fortunately, a solution emerged pretty quickly. Instead of fighting the Internet and MP3 distribution, the music industry embraced it. Sean Fanning failed to find a legitimate business model with Napster and the courts shut down the network in 2001. But Sean Fanning had started the company with another Sean. Sean Parker. Yes, that Sean Parker, the former president of Facebook who was famously played by Justin Timberlake in the social network. In 2006, Sean Parker became a managing partner at Founders Fund, the venture capital firm that backed Facebook early on. And while he was there, he started looking for a company that could make the original vision that Napster had pioneered a reality. But they would need to do it legally this time around. A few years later, a friend showed him Spotify, and he invested $15 million in the company. Spotify had figured out that what consumers really wanted was access to every song at any time, but they were willing to pay for that. The plan worked, and the music industry was able to hit a new all time high for revenue in 2021. CD sales had been shrinking ever since Napster came on the scene in 1999, and the music industry had suffered through a rough 15 year period of declining sales. But by finally leaning into the Internet instead of fighting it, they were able to restore growth. The shift from CD sales to streaming did change the nature of music production in some significant ways, though. Now artists weren't limited to what would fit on a single cd, so albums could be any length with an unlimited catalog stored on the Internet. There was no downside to releasing more music, so artists started producing more and more songs. The Internet also removed traditional gatekeepers. Instead of needing to get a record label to manufacture physical CDs for you and then distribute them into stores, songs were instantly available everywhere. So new artists could go viral with a single song faster than ever before. So what does all this mean for YouTube? When the founders of YouTube first launched the site in 2005, they ran into many of the same problems as Napster. There was a ton of copyright infringement. In the early days of YouTube, people would upload songs, movies and TV shows, anything that could get them views and earn them ad revenue. It was pretty chaotic, but they ultimately solved the problem by creating the Content ID system. In the early days, record labels would have to search YouTube for illegally uploaded songs and then submit requests to YouTube to get the offending videos taken down. Content ID solved this by scanning every YouTube upload and cross referencing new videos against copyrighted material. Now, if you upload a video featuring a Drake song, YouTube doesn't even need to take that video down. They can just divert the ad revenue to Drake's record label and everyone's happy. And this is clearly the future of AI generated content. Drake owns his name, image and likeness. It doesn't actually matter that he didn't record hard on my sleeve, he'll be able to get paid for it. This will still mean a big change for the way the music industry works and how artists get paid. Right now, Drake's Record label is fighting to remove every copy of Heart on My Sleeve from the Internet. You can't stream it on Spotify anymore, and and uploads are constantly being taken down from YouTube. But we're far from the end of the road here. It's important to remember that most major artists work with large teams. Drake himself admitted in an interview with Fader that he works with a group of writers to help develop his lyrics. But it doesn't really matter, since the value of a Drake song comes from the fact that it's a Drake song and not some unrecognizable artist. Even though the record labels are scrambling to ban AI generated content in the in the short term, this is probably going to play out just like it did 20 years ago. Eventually, the music industry will learn to lean into AI and figure out a way to continue to make money while giving people what they want. So what could that look like? Well, the simplest solution would be something like what YouTube has already built with Content ID. If the platforms detect a song using Drake's voice, that song is ineligible for monetization, and all ad revenue flows directly to Drake's label. Big platforms like YouTube and Spotify are also in a unique position to authenticate what's real and what's fake. Real artists already have verified accounts on there, so if you want to make sure you're listening to a real Drake song, you better be sure that you're on his page. And in the longer term, it's clear that artists will embrace working with AI instead of against it. Small independent producers have been uploading type beats to YouTube for years. These tracks often mimic the style of a particular rapper, and sometimes they even get used by the actual rapper they are mimicking. Earlier this year, J. Cole searched for J. Cole type beat on YouTube and found something that he liked. Ultimately, it was good enough to be a real song, so he recorded vocals and let the producer upload the final track to his channel. Something similar could happen with AI generated vocals. If Drake winds up liking a particularly viral AI song, he could record an official version and even perform it live. Big artists like Drake are going to be massive winners in the AI revolution. They own their intellectual property, and it won't matter if someone else prompted AI to create a song with their voice. They are still going to reap the financial rewards. But what about smaller creators? The average YouTuber? Someone who doesn't have international fame like Drake? Well, this is where things get really interesting. It's easy to look at the crazy news about AI and assume that all video will soon be entirely synthetic. And it makes sense in theory, right? ChatGPT can write a script, midjourney can generate visuals, and plenty of services can synthesize voiceovers. Put those together and you have all the ingredients to make a YouTube video automatically. But we aren't quite there yet. Every piece of AI generated content still requires a lot of human effort to produce high quality results. AI is clearly advancing quickly, but people are starting to get a little crazy with their predictions. In order to predict how things will unfold, unfold here, it's critical to understand what I call the Centaur era. Just like the half man, half horse creature from Greek mythology, the near term future of content creation will involve humans working in close coordination with AI to push production to new heights. This Centaur concept first showed up during chess competitions in the 1990s. Garry Kasparov was the world chess champion at the time, and he thought that no computer could could ever defeat him. Computers were still so simple back then. They didn't really seem capable of advanced reasoning. But then in 1997, IBM's Deep Blue beat him in a public match. It was a huge turning point in the development of artificial intelligence. And Deep Blue didn't even use any modern AI methods. But after that match ended and Deep Blue had been crowned chess champion, a new challenger emerged. Instead of just a single human playing a single computer, a human would now play alongside a computer and work together as a centaur. This combination of human and machine wound up being incredibly effective. The computer could act as a check on every move to prevent unnecessary blunders, while the human could read the opponent and focus on long term strategy. For years, these mixed AI human Centaur teams were able to outperform pure AI programs at chess. And this is the era that we're entering now with content production. Increasingly, YouTube creators who are able to work alongside tools like ChatGPT and MidJourney will be able to produce better videos faster and more efficiently. We see sparks of this already with videos like the Harry Potter Balenciaga mashup. It's not impossible to create a video like that without AI. A team of Pixar level artists working for a few weeks could probably recreate it without any AI tools, but it would be extremely expensive to do so. Midjourney was able to give the creator of Harry Potter, Balenciaga, the ability to bring this bizarre and hilarious concept to life basically for free. It still required a lot of human work to combine the different AI generated elements and most importantly, come up with the idea. The video works so well because combining characters from a beloved children's book with a pretentious high fashion brand is so non sequitur. The originality and unconventional thinking that went into deciding to make this viral video is something that humans will be better than AI at for a very long time. But at the same time, that advantage may not last forever. After all, Centaur teams don't stand a chance of winning at chess anymore. If artificial intelligence continues to compound expansion, it's hard to imagine a particular skill set that will remain untouched. We don't know how long it will take to create super intelligent AGI. It could be years, or it could be decades. So instead of jumping too far into the future, it's more useful to work through the various parts of YouTube that will be affected by AI. First, we know that YouTube content creation is entering a Centaur era, so how will this change the platform? The most basic videos will get automated by AI simple videos that are just adaptations of Wikipedia articles or listicles of Reddit comments. Those will be replaced by AI generated content. But for more ambitious YouTubers, AI will very much be a copilot. Microsoft partnered with OpenAI for this very reason. They recognize that AI is best used as a copilot right now, so they are integrating it into Microsoft Office to help people write more efficiently. And that copilot branding, which was first used in their GitHub code generation tool, speaks to the importance of the Centaur model. AI copilots allow humans to act as centaurs and produce better results faster, and this works on the visual side of things as well. Generative imagery has already helped me a ton. When OpenAI released Dall E, I produced a video telling the history of artificial intelligence and illustrated the entire video using only AI generated images. The technology has gotten a lot better recently and is starting to replace stock footage in some of my videos. It still requires human input to know what images look good and fit the story, just like searching through a stock footage library. But the possibilities are endless. Many YouTubers like to say that gear doesn't matter when they get asked about starting a YouTube channel. It doesn't matter to me. Just make something, put something into the world. This is good advice for new creators. Instead of worrying about buying the perfect camera, it's usually best to just get started and then improve your equipment over time. AI is going to make gear even less of an issue going forward. Not only does every modern smartphone already have a ton of AI features integrated into the camera system to produce better output, but there are also a ton of new AI tools to restore the quality of low res video, add slow motion after the fact, and even clean clean up noisy audio. The cost of creating a YouTube video is going to drop lower and lower until basically anyone can create any video they want just by typing a few prompts. So in that world, who wins? What will the top YouTubers look like? There are basically two paths forward as I see it, and to understand them we need to talk about Jake Tran and James Janney. These two guys are YouTubers with a over a million subscribers. They both make video essays about business and they focus on cinematic storytelling, but they run their businesses completely differently. Jake is all about scale, while James is all about quality. Even though they both have nearly the exact same number of subscribers, Jake uploads nearly every single day while James spends months working on a single video. The end result is that even even though James Janney gets 10 times the amount of views per video on average, Jake Tran still gets four times as many views per month. The quantity at all cost strategy has allowed Jake to get lots of views, but in the process, most of his audience turned on him and he drew some negative attention from Coffeezilla. James Janney, on the other hand, is loved by his community. They patiently wait for the next upload and show up in droves when it goes live. If making YouTube videos is just about paying the bills, either strategy can work. Jake Tran is at the extreme of optimization. He outsources everything, and at one point he even hired other narrators to do the voiceover so that he wouldn't have to spend any time on that. AI will push this even further. There will clearly be channels that outsource nearly everything to AI systems. But on the other end of the spectrum, I think there will still be a place for the James Janneys of the world. So let's talk about the things where AI falls short. ChatGPT is only going to get better at summarizing information, repurposing existing articles, and generally delivering the right answers. This means that humans will need to focus on finding truly original ideas, thinking unconventionally, and asking the right question. So any YouTuber who can consistently provide a unique viewpoint that isn't just a summary of what's already been written about, a topic will remain relevant. Another area where humans will have the edge over AI for a long time will be in surfacing information that doesn't exist on the Internet. Coffee Zilla has basically become an investigative journalist at this point. When he reports on a fraud, he isn't just paraphrasing a CNBC article about the topic. He's actually going out and talking to real people and finding new facts to help tell the story. This is going to be extremely valuable for a very, very long time. And it doesn't stop there. The most successful YouTubers today are able to establish a connection with their audiences, and it's going to be really hard for AI to replace that. Take livestreamers like Ludwig for instance. Not only will it be years before AI video generation gets to the point where it can be used on live streams, but Ludwig's audience likes the fact that he's a real person and seems like a relatable guy. AI models still require months to train, and it's a big challenge to keep them up to date with the latest information. So recreating something like a Ludwig livestream, where he's reacting to the latest news and what his viewers are saying in chat will be a big challenge. Even still, it won't be impossible, and it's probably only a matter of time. So what then? What happens when there are AI channels that are truly indistinguishable from the YouTubers we know today? Well, I think it's useful to return to the history of chess. For hundreds of years, humans were dominant at chess. Then we got beat by AI in 1997 and entered the centaur era. For about 20 years, human AI centaurs were the best at chess. And then AI took over completely. It's simply impossible for a human to win at chess against an AI system, no matter what tools they have at their disposal. But here, here's the interesting Human chess has never been more popular. Chess.com has over 100 million users. Hikaru Nakamura and Magnus Carlsen have massive followings on Twitch and Netflix, even produced a hit show all about chess. For some reason, we humans don't just want to see the best possible game of chess get played out between two AIs. We want to see humans battle it out. It might just be an innate bias against computers, but I think this will stick around for a long time. Fundamentally, every YouTuber is going to reach a fork in the road. They'll have two options, two opportunities really. They can either use AI to optimize their production for maximum quantity, pumping out as many low cost videos as possible in order to gain a lot of views. Or they can focus on quality, let AI help them with what it does best, but still focus on injecting that human element that makes their videos special to their core audience. The risk of going the mass production route is that it can be a bit of a race to the bottom. Eventually, these videos might be automatically produced by YouTube directly, since they have all the training data. On the other hand, if you focus on what's hardest for AI to replicate unique ideas, personal stories, investigative reporting, that will probably be a much longer and more defensible road. This AI stuff is moving incredibly quickly right now, but if you zoom out, this has actually been in the works for decades. So if you want to understand the history of artificial intelligence, just go watch this video next. Thanks a lot.