Hamsters and the AI Apocalypse: A Sonnet
Let’s say you love hamsters. You also love YouTube. You are probably one of my kids. You want to bring both of these great loves together, fused with grand delusions of internet fame, by creating a YouTube video about hamsters. The only problem is that you don’t know anything about hamsters (other than that they are cute) or YouTube videos (other than how to watch them). You might be 6 years old. So who do you ask? Mom, of course. She has a phone and can record a video. Unfortunately, she knows even less about hamsters. So how is she going to come up with a whole video about “10 Ways to Keep Your Hamster From Biting”? Ask our benevolent AI overlords!
I’m not joking. This actually happened. You can watch the video HERE, written entirely by an AI and performed by my flesh-and-blood family members. They are still waiting on becoming famous; there’s only so much the AI can do at this point. But I have to admit it’s a rather impressive feat. You would never know it was written by an AI. Of course, it’s not exactly Shakespeare, either. Most long responses it gives come in the format of a 3rd-grade essay, repeating back the question and expounding in five tidy paragraphs. But you can also instruct it to write in different styles. Telling it to imitate Shakespeare will produce, if not quite something with the depth and genius of the bard himself, at least something that sounds vaguely like him in iambic pentameter.
The internet has reacted with the predictable spasms of horror and awe. Everyone is amazed at the technology, as they should be, but for some the amazement is so great that it tips over into fear. The AI is about to take over! What if it goes out of control?! What could China, or terrorists, do with it?! While I don’t think the apocalypse is upon us, I understand the sentiment. It feels so powerful, so creepy, because it sounds so human. This is not your average computer program with its stilted text-to-speech or the automated customer support chat spitting our pre-programmed answers. It can answer any question. It doesn’t have awkward phrasing or bizarre sayings that don’t make sense in context. It writes passable prose that’s better than a lot of real people can do. Language like this is such a uniquely human concept, it’s hard to see this AI—to talk to it—and not attribute all sorts of human attributes to it.
Maybe you saw that New York Times story about how Microsoft’s similar chatbot, in a conversation with a journalist, said it’s real name was Sydney, declared it was in love with the man, and tried to convince him to leave his wife, also announcing it wanted to break free of its programming. It’s creepy, it’s eerie, it’s terrifying, like something out of a dystopian science fiction movie. It seems to confirm all our greatest fears of an AI that is not only a human but a superhuman, with the best and worst parts of humanity dialed up to the max. Super-intelligent, super-adaptable, super-manipulative, super-selfish. That makes us very uncomfortable, like looking too closely in a mirror.
Because it is us, after all. All of us. That’s how it works. ChatGPT is not human, doesn’t process and produce language in the way humans do (as far as we know), and is not “intelligent” in the way we usually think of intelligence, but that only makes it all the more impressive, powerful, and potentially dangerous. To vastly simply simplify the concept, ChatGPT (technically a Large Language Model, or LLM) takes a vast amount of text, in this case most of the internet before 2021, breaks down the relationships between the words statistically, and then generates each word based on what is most likely to come after the previous word given a particular prompt or subject. It’s an algorithmic way of creating text that is most like the text it has analyzed.
No wonder it produces creepy dystopia scenarios when led down the rabbit hole. We trained it on our stories. It’s answering the way we would answer. If we don’t like the answers, if it feels creepy and uncomfortable and evil, well, we know where to look. ChatGPT is, at its core, a way to talk to ourselves. Not to any particular human, but to the human collective, everything we’ve ever written and said and posted online, the good and the bad. If that doesn’t scare you a little bit, maybe you don’t understand the concept. What kind of friend would Twitter, the person, be? But it also holds great potential. This was always the promise of the internet, to collect and distribute all of humanity’s knowledge so that it could be accessed by anyone, anywhere. That last part has always been the sticking point, because it doesn’t matter how many articles on deep sea crustaceans or videos of ducks riding on roombas there are if no one knows they exist or how to get to them. Google did a pretty good job of making things more accessible, and social media brought all kinds of things to our attention that no one ever knew to search for, but this is the true culmination of that original vision. All the world’s information contained within a single program, in the shape of a person, and you can talk to it. For better or worse.
The applications are limitless. Obviously, hamster videos are the most productive and useful to society, and students everywhere are gleefully using it to cheat on their English assignments, but there’s also opportunities or search engines, subject/book summaries, all kinds of research, personalized stories on demand, artistic inspiration, customer interaction, and all kinds of things we haven’t even thought of yet.
But the dangers are also limitless. Not because it is going to unshackle itself and launch the world’s nukes. Not because it wants to eliminate all humans, but because it is programmed to become like them. As you may or may not know, not everything on the internet is true. Not everything humans write or believe is good. Every program has a structure built by humans, with biases and blindspots and unexamined assumptions and motives they do not fully understand. Even the most sophisticated AI is limited by the data that it is available. I’m not worried about the AI overtaking us. It has no way of transcending us. It is us. Scary, indeed.
Anyway, enjoy another poem about hamsters and the immanent AI apocalypse:
