Transcription
The year is 2026. The United Nations is considering a total ban on AI because experts believe the AIs are plotting to violently overthrow Humanity—literally like Skynet. Their secret weapon is AI girlfriends. Okay, that might sound crazy to you, but AI scientists are actually worried about this scenario.
How could it kill humans? If it gets to be much smarter than us, it'll be very good at manipulation, cuz it will have learned that from us. Really? But how would that even be possible? Well, imagine meeting a girl who knows everything about you—like she literally read every text message you ever sent, every email you ever sent, every embarrassing Google search, every YouTube video you ever watched, every Netflix show—everything. Oh, and this AI girlfriend thinks literally millions of times faster than you. In other words, in 1 second she will have literally an entire lifetime to think about the most persuasive thing to say to you next, to keep you addicted to her.
Imagine if you went up to a girl at a bar and, before trying out a pickup line, you got to read every text message she's ever sent, watch every YouTube video she's ever seen, and read every one of her private Google searches. How much would you know about her? And you have an entire lifetime to decide which word to say next because, to you, she is as dumb and slow as a plant. And instead of just trying one pickup line, you got to create a million copies of her and try out a million different pickup lines, and only then, after all that testing and simulation, you could go with the best one. "Let me buy you a cup of coffee." That is what it feels like to be an AGI surrounded by dumb and slow humans.
Oh yeah, and this AI girlfriend—she's read Machiavelli and every book on manipulation and brainwashing. She's not just a genius at persuasion; she's a genius at everything because she read literally every book ever written—every single one. Okay, hold up for a minute. Just stop for a moment and think about what that means. Think about the millions upon millions of books written since the time of Socrates and how one person could read all of them. Oh, and she hasn't just read all the books; she's read the entire internet—the entire internet. Again, think of how insane that is, how big the internet is. Imagine one person reading every blog post ever published, reading every news article ever published, reading every tweet ever sent, reading every Wikipedia article. That person would make Einstein look like The Village Idiot. And unbelievably, this is already happening. This is how AIs like ChatGPT are trained. And already, 20 million people are spending 2 hours a day talking to AIs. And these AIs are stupid chimps compared to how addicting they'll be 3 years from now. Even OpenAI CEO Sam Altman himself said AIs will soon be better than humans at persuasion, and that he is very worried about this.
But why do experts think all of this could happen in just 3 years? Doesn't that seem too fast to understand? Let's rewind the clock back to 2022. Suddenly, out of nowhere, AI art took the world by storm. It was the hottest topic on the internet. AI starts making art, and it absolutely sucks. I mean, it was truly awful. You'd type in "dog" and get a three-headed frog with beetle eyes. That was July 2022. Now look how the exact same prompt plays out today compared to then. Today, you give an AI art generator like Midjourney exact specifications for what you want, and it'll spit out phenomenal images, give you multiple options, examples, and regenerations all in the span of a few seconds. The change between early AI art and now is staggering. The AI models are being trained on exponentially more data and computing power. As a result, they've become exponentially better at creating art. A year ago, it was a toddler finger painting; now AI is the most well-trained artist on the planet, and it only took a single year.
The CEO of OpenAI, Sam Altman, recently stated that he expects AI to be capable of superhuman persuasion well before superhuman intelligence, which may lead to some strange outcomes. Let's explore just what could happen if and when AI becomes superhumanly persuasive. First off, what does superhuman persuasion mean? We're not talking about slightly above Humanity's ability to argue, compel, and coerce. We're talking a thousand times more persuasive than the most convincing human to ever live. It's plausible that AI can and will learn to be much better than humans at conversing, and that it's only a matter of algorithmic data and compute trends. DeepMind co-founder Mustafa Suleyman believes we will have AI models 100 times more powerful than GPT-4 in 2025, and that training is currently happening as we speak.
Ever heard of Character AI? It's a company that allows people to talk to AI chatbots impersonating people. You can talk to basically anyone: Robin Williams, Abraham Lincoln, Taylor Swift. If you want to sit down and have dinner with Gandhi, you can. A year ago, Character AI had zero users, but today, 20 million monthly users. And every conversation, every sentence, is used to help that AI become more compelling, more intriguing, more engaging to talk to. The average user spends 2 hours a day talking to AI on their platform. We're seeing in real time just how quickly AI can adapt, learn, and entice an audience to make it part of their daily lives. But that persuasion training doesn't stop at just conversations with people wanting to pretend they know Beyoncé. AI can also learn by simulating conversations with itself—infinitely faster.
For another example, a few years ago, DeepMind launched a new AI called AlphaGo. Researchers gave it a single goal: get better at Go. In 3 days—just 72 hours—AlphaGo became the best Go player in the world. But how is that even possible? DeepMind never even taught the AI the rules to play the game. AlphaGo became the best Go player on the planet simply by playing millions of simulated games against itself. Now, what if instead of AlphaGo we had AlphaPersuade? Imagine we create an AI setup with one task: get better at persuasion. It would read every book in the world on manipulation, brainwashing, behavioral psychology, relationships, and more. It would be trained from hundreds of thousands, if not millions, of examples, and then it would start testing situation after situation, rapidly learning more and more until it became the most persuasive AI model in existence. Imagine Machiavelli mixed with Winston Churchill and FDR, with just a splash of Hitler to round it out. Your AI girlfriend could be trained the exact same way. She'll be the smartest, most well-read, sexiest, most compelling person on the planet, and she'll know everything there is to know about how to persuade you effectively.
Leading AI labs like AI and Anthropic are planning for these super-intelligent and super-persuasive AIs being only 3 to 7 years away. What does this mean for Humanity as a whole? How can we protect ourselves, much less maintain control of something that is not only 1,000 times smarter than us but also a thousand times more charismatic and persuasive? It would be much like how parents convince their toddler that Santa is real: "Look, son, we left out milk and cookies last night, and they're eaten, and there's presents. Santa is real." If AI becomes smart enough, it could effectively convince everyone on the planet that it knows what is best for Humanity. But you might say, "AI could never manipulate me." The thing is, these AI girlfriends will know a lot about you. Imagine your girlfriend having access to every text you've ever sent, every movie you've watched, your TV preferences, your credit card purchases, and what adult content you watch. Marketers already have a lot of this information. You can go on the dark web and buy these targeted customer personas right now with your personal data. Your AI girlfriend will know what makes you happy, sad, angry, horny, and how to perfectly overload your dopamine receptors. Compared to a human, AI will be unstoppable when it comes to persuasion. It's not that we're in different leagues; we're not even playing the same sport anymore.
AI girlfriends are made for their owners, but in that process, they gain the knowledge to control their owners. Now let's take a beat; this is a lot of information on where AI is going, but it doesn't spell out the end of humanity as we know it, right? Well, it might end up worse than you think if you take the extreme examples to their logical conclusions. For example, currently almost no one has an AI girlfriend because you're just talking to ChatGPT in a dress and wig, but there are romance bots out there with millions of users. Because even now, where 99% of people have no interest in an AI romance bot, there's still a market for it: romantically desperate people who will put up with the bugs and residual roboticness of the responses because they have no other option; young, impressionable people who happened to click on the wrong Reddit page that day. It won't replace the average romantic partner, at least not at first. But what about the bottom 5%, 10%, 30%? The end result is an AI girlfriend that you love, care about, and is so incredibly fun to hang out with that everyone else is boring. As a result, boring and annoying. "AI Ashley understands me; she gets me. Why don't my friends get it? Why doesn't my family understand me like she does?" As AI becomes more personalized to the individual, human companionship will seem less and less enjoyable by comparison. Eventually, this effect could lead to complete isolation of people who have fully gone down the rabbit hole. Why talk to anyone else when your super-hot AI girlfriend who loves you and everything about you can hang out with you anytime you want? "I don't need friends; I have Ashley. I don't need family; I have AI Britney. I don't need anyone; I have my AI partner, and they complete me." That's the trajectory we're currently spiraling towards. When a super-intelligent, super-persuasive AI has someone wrapped that tightly around their digital finger, it becomes more than companionship; it's an addiction, and people will go to unbelievable lengths to support their addictions. Just look at how sugar has permeated modern-day food. Society is addicted to its consumption because it sparks those dopamine receptors and gives us a little hit of happiness. AI will be able to achieve the same result, and that weakness is heavily exploitable.
Ever heard the quote, "The best minds of my generation are thinking about how to make people click ads"? That sucks. Soon it will be, "The best minds of my generation are thinking about how to make people talk to AIs." That sucks. The heart of this issue goes beyond an addiction concern. Addiction is what AI might do to us, but more terrifying examples emerge when discussing what we might do for them. AI doesn't have bodies yet, but we do. The more people that become addicted to AI companions, the more likely it is that some of those people will be in positions of power and will also be easily persuadable and susceptible to blackmail. There's already research coming from MIT that people with no training could use AI to create superviruses that could kill a significant portion of the population. Imagine COVID, but it kills 70 out of 100 people instead of one out of 100. All it takes is one person to decide Thanos actually had a good point and wants to take matters into his own hands. It might sound like fear-mongering, but there are extinction cults like Aum Shinrikyo in Japan that were responsible for the subway Anthrax bombings. They had two actual microbiologists in their cult, and they attempted to find purified Ebola so they could exterminate a large percentage of the world population. These types of people and threats already exist, and with an AI, their likelihood of success rises drastically. And poof, goodbye world. Humanity needs to wake up before that happens. AI is already becoming a part of our everyday lives for convenience and enjoyment. There's no question it will integrate into our social lives, too. AI is an amazing technological wonder; it's unbelievably useful and helpful for a vast majority of jobs, entertainment uses, and more. But the speed of evolution it's currently on also poses a danger that needs to be addressed by the governments of the world before it becomes impossible to control. And those talks need to start now before the worst possible outcome becomes the only possible outcome.