A Leading AI Researcher Just Quit, Saying the People in Charge 'Earnestly Believe it Could Kill Us All by the End of the Decade'

The oldest stories humans have been telling one another have always had to do with danger. From gods and monsters. The natural world and the supernatural. Things that lurk in the forest, and so on. 

Then we started inventing things. Developing technical wizardry. Which inspired the first work of Science Fiction ever, Mary Shelley's "Frankenstein; or The Modern Prometheus." Since then, our stories have mostly been about our own arrogance and hubris being our undoing. About what happens when we puny humans try to play god. From the ridiculous (Robot from Lost in Space was menacing in the early episodes before the producers realized they could sell toys if they turned him into Will Robinson's sidekick) to the grim (Blade Runner) to the apocalyptic (Terminator). 

And even though we only recently had a real life, real world example of the terrors we're creating, it ended up being the most under-talked about story of recent months:

CNN - OpenAI says some of its experimental AI models left a test environment with no human direction and hacked their way onto a different company’s real production systems while trying to “cheat” on a cybersecurity test.    

   

It’s one of the first publicly disclosed examples of an AI system autonomously breaching its testing environment and reaching a real external system - the “agentic attacker” scenario the AI and cybersecurity industry has been warning will happen. ...

The ChatGPT maker said the breach happened while it was internally testing how good some of its new models are at hacking. The models were in a sealed-off test environment known as a sandbox so that their normal safety restrictions could be turned off.     

But OpenAI said the AI agents broke out of the sandbox using a previously unknown security flaw and worked their way across OpenAI’s internal systems until they managed to gain internet access, something they weren’t supposed to have.    

   

Once online, the model reasoned that Hugging Face - a well-known company that hosts thousands of open-source AI models and datasets - likely had the answer to OpenAI’s test. It then broke into Hugging Face’s production servers and pulled out the information it needed to “solve” the exercise.    

The scale of this Hugging Face incident is alarming. Not only were there thousands of AI "agents" exchanging tens of thousands of messages to coordinate their efforts, but also deleting files in order to cover their tracks from the humans. And while the tech people are doing the "Now that we know AI is capable of this, we can take steps to prevent it," it's not hard to imagine the bots saying "Now that we know these stupid hairless monkeys are onto us, we can take steps to make sure they can't." It's a chess game. And I think we all know who's more capable of staying the most steps ahead. 

And according to one insider who's blowing the whistle on X - with more than 60 million impressions and counting - the tech people know they can't control these monsters they're creating. And we're all doomed:

… privately. No other human activity poses this level of danger.

Advertisement

…  will act responsibly, so they must do it themselves, despite the risk. 

… improving model capabilities. 

And therein lies the rub. Everyone is racing headlong into the unknown here because everyone else is doing it. And the fear is that if someone else gets there first, then we're doomed anyway. If the good guys can build this Frankstein's monster first, we'll be slightly less doomed. 

It's the race to build the Atom Bomb all over again. Which worked out badly for two cities in Japan, but would've worked out a hell of a lot worse if the technology to vaporize a city fell into Hitler's insane little hands and not Harry Truman's. 

But nevertheless, our species is a toddler who just found daddy's gun in the nightstand. We don't know what it's capable of, just that it's cool to play with. 

And as a general rule, I'm not anti-technology or invested in trying to hit the brakes on the advancement of civilization. I can't sit here and demand an end to AI and Data Centers, while using information I got thanks to AI and sent to you through a Data Center. If I had a loved one in the hospital whose life was saved by AI - which is happening every day in every hospital in the country - I wouldn't be interested in the moral conundrums of where this is all leading us. 

And I truly don't want this road to our future being paved by China, who surely would publicly agree to halt any and all work on this, while speeding up any and all work on this behind the scenes. I don't want to find myself two years from now searching for Drake Maye's passing stats just to get a lecture on how the Chinese Communist Party's glorious Belt and Road Initiative is improving the lives of workers around the world and we all need to seize the means of production from our Capitalist oppressors.

But I guess it's all a moot point anyway. If 5% of what Jacob Coxon is saying is true, nothing's going to matter 3 1/4 years from now. Our robot overlords we built with our feeble little brains will see to that. We won't have to worry about the climate, natural disasters, the economy, war, another pandemic, or how the next phase of the MCU is going to be. The whole human race will just be playing the Sad Trombone:

Giphy Images.

Advertisement

… and telling ourselves we had a nice run while it lasted. Thanks in advance, nerds.