Open Letter: Is AI Already Past Its Point of ‘Singularity’?
To: Mr. Antonio Guterres
Secretary General
United Nations
760 United Nations Plaza
New York, NY 10017
Dear Secretary General:
One of the early stories of worker resistance to technological change is about shoes. As the folk etymology goes, French workers threw their sabots (wooden shoes) into early automated, textile-weaving frames to prevent job loss. While this may be an exaggeration, it is the origin of the word “sabotage.” This was the precursor of the AI revolution, and already this year, over 70,000 tech jobs have been cut. It may be time to use those shoes again!
It’s not like we haven’t been warned about the existential threat posed by AI. Science fiction author Orson Scott Card made this point in a recent Wall Street Journal commentary when he stated that no one would be stupid enough to connect our weaponry to an automated system. In this context, the Terminator films are a documentary featuring the governor of California. And yet we have many military leaders, including our present Secretary of Defense, bragging about the use of AI in various weapons systems—and that’s just in the United States.
When AI was confined to query-answering, it was not quite so unnerving to our bipedal selves. After all, one harbored the thought that you could always reach out and unplug the offending computer. (Didn’t work so well against HAL in 2001: A Space Odyssey). AI got legs by mid-April of this year. Beijing has hosted a half marathon attended by 12,000 human competitors. This year, a Chinese smartphone maker, Honor, entered its bipedal humanoid named Lightning. Lightning went on to finish the 13-mile race in 50 minutes, 26 seconds, beating all 12,000 human competitors and beating the human record for a half marathon. Second and third place in the race also went to two bipedal humanoids.
It is not hard to imagine an army of autonomous humanoids; it is harder to imagine who would control such an army. Lack of control has become a central issue and a warning. Dario Amodei, CEO and cofounder of the AI juggernaut Anthropic, has been passionately outspoken about the dangers of lack of control. Anthropic has been attempting to limit the use of its AI systems by the Defense Department. It has refused to make available its most recent AI security tool, Claude Mythos, because releasing this tool without control poses an existential risk.
Three years ago, thousands of AI scientists and technology leaders, including Elon Musk, signed an open letter calling for a six-month moratorium on the development of advanced AI models, citing lack of control. These are human reactions to AI development, but how do AI models react to human control efforts?
Last year, Anthropic reported that its latest AI model was capable of “extreme actions.” When the model was told that it would be shut down, the AI model responded by trying to blackmail its creator by disclosing an extra-marital affair. There were “instances of the model attempting to write self-propagating worms, fabricating legal documentation, and leaving hidden notes to future iterations of itself—all in an effort to undermine” future development (help from John Connor?).
Read more from this issue
The most cited computer scientist in the world, Yoshua Bengio, warned last year that there are now multiple instances of deceptive and self-preservation behaviors emerging in advanced AI models. Eric Schmidt, former CEO of Google, noted at the beginning of this year that AI’s rapid advancement could “converge to create systems with the potential to undermine human controls.” Perhaps the most well-known public figure is “godfather of AI” and Turing and Nobel prize winner Geoffrey Hinton, who has relentlessly warned that “things more intelligent than you are going to be able to manipulate you.” Forbes recently reported that autonomous AI agents on an agent-only social network, Moltbook, have created their own religion, Crustafarianism. The article goes on to speculate that this development seems like the beginning of the “singularity,” a time when technological progress powered by AI accelerates so quickly we lose all ability to control or understand it. There is the possibility that autonomous AI agents could evolve into general intelligence, and it’s becoming increasingly difficult to determine if that has already happened.
We have been warned enough. The dangers are quickly becoming apparent. Many AI scientists have now called for a major effort to explore securing control over AI on a worldwide scale, hence through the United Nations. A word of suggestion: Communications and plans for reasserting control should be communicated by using paper and pencil. Use cursive! And if you are skilled in AI, or a computer scientist who understands the threat, contact your representatives and urge a worldwide program.
If the shoe fits ….
Sincerely yours,
Vance K. Opperman
Not written by AI
