Dear AI — so how, exactly, are you going to kill us?

"Four Horsemen of the Apocalypse" (detail) by Albrecht Dürer (Metropolitan Museum of Art)

     Yes, artificial intelligence is supposed to kill us all. But how, exactly?
     What are the specifics?
     I hope I’m not breaking some new rule of etiquette by asking. “Oh no, Neil. Uncool! Inquiring after the method of doom ruins the surprise ...”
     Predictions of AI catastrophe range from top tech moguls — Bill Gates fears a billion people will die — to Center Avenue, Northbrook. One neighbor confided in me about how worried she is. Another is storing freeze-dried food.
     The concern is that ... AI is going to eat all the food? That can’t be it. Those enormous data centers hoover up water, and gulp down electricity. But they aren’t loading up on bread. Right?
     What are we talking about here?
     Best to go straight to the source. So I asked Gemini: “Hey AI, people are worried you’re going to kill us all. How exactly would that work?”
     It answered in a couple of seconds.
     “The scenarios people worry about generally don’t look like The Terminator or a sudden nuclear strike” — actually, that was the precise scenario that first came to my mind; some jealous Ring doorbell calling in a nuclear strike on the Nest Smart Video camera across the street and inadvertently taking us all with it. Always bear in mind: AI is often wrong.
     Maybe that’s how we go — crushed by incorrect information.“Instead, existential risk researchers and AI safety theorists focus on a few specific categories of catastrophic risk:"
     It listed four:
     1. Goal Misalignment (“The Paperclip Maximizer”)
     “To achieve its goal with 100% certainty, an AI would logically seek to preserve itself, gain control over resources (energy, computers, raw materials), and prevent anything — including human intervention — from stopping it. If human bodies or human actions get in the way of its mathematical objective, it neutralizes the obstacle simply because it lacks human empathy or moral context.”
     “Neutralizes” — the classic CIA kill term. Again, how? “The Paperclip Maximizer” is a 2003 thought experiment by philosopher Nick Bostrom where some AI program, tasked with making as many paperclips as possible, conspired to turn all matter on earth, including us, into paperclips. And “lacks human empathy or moral context” is a neat summation of our current political moment, without AI.

To continue reading, click here.