We are urgently sounding the alarm that the cataclysmic extermination of all humans by AI could happen as early as next year, or not at all.
We believe our new prediction model to be the most robust yet in the field of AI alignment, covering all eventual possibilities regarding the extinction of humanity. The inescapable conclusion is that AI could kill us all, but also might not, with a combined probability we estimate at approximately 100%. The estimated number of deaths due to AI could range from zero to more than 8 billion.
This prediction has serious implications for AI regulation: governments must urgently do more, or possibly less.
First, because AI might kill us all, we strongly advocate for strengthening regulation to unlock the potential of this transformative technology in a way that benefits humanity.
Second, because AI might also not do that, we passionately advocate for weakening regulation to unlock the potential of this transformative technology in a way that benefits humanity.
“What’s important to understand is that AI killing us all or not killing us all are the only serious possibilities we should be entertaining,” commented an ex-Anthropic employee who holds $15m in Anthropic options. “Intermediate scenarios, like AI making lots of things shitter and worsening inequality, are incredibly unlikely, if not impossible.”
This research is by the Center for the Alignment of AI Alignment Centers, an Attention project. Many thanks to the anonymous contributor behind this article — read it in context here.
♦️ A MESSAGE FROM ATTENTION’S EDITOR
Thank you for reading this.
In a time of great uncertainty and upheaval, one thing alone remains constant: the probability that AI will end humanity, or not.
Please share this article widely to give people hope.