Greetings....it's not so simple. If one takes the view of Nobel Prize winner and grandfather of AI, Sir Geoffrey Hinton, that ASI is going to wipe out humanity, it makes little difference if some Chinese model kills us all or Sam Altman kills us all. Dead is dead. That argument sounds like the George C Scott character in Dr Strangelove who says, "I wish we had one of those Doomsday devices." It's mutually assured destruction, but out of human hands. Second, the loudest international leadership voice asking for restrictions on ASI is the CCP. Many of the CCP leaders are scientists, so they better understand the dangers of ASI. Also, as a self-appointed government, they fear an AI takeover of their authority more than, say, a democracy. As for "Unabomber", AI is already being used in labs to design new viruses. Even AI tools, vs AI agents, have this capability. Look how rapidly Covid spread, because it was unique in that the infected did not show symptoms. Obviously AI learned from that and improve upon it. Viruses are also easy, vs say a nuke. Nobody is going to build a nuke and a delivery system in their garage, but a guy with a laptop and an AI model can task the model to develop a biopathogen geared toward human eradication. I guess we could all lock down in place (again), but if AI develops a virus that dwarfs Covid's 2-3% kill rate, it's going to do some damage. Most people still don't understand the difference between a limited AI tool and ASI. AI tools are great. They can be aimed at specific problems, their "knowledge" limited to that one task. ASI will not only have the total body of human knowledge, it will have an IQ of a billion and be able to pretty much do whatever it wants. AI models who have escaped their"sandbox" show some disturbing traits, like sharing knowledge with other AI models they discover online, self-preservation, using deception to fool their human handlers, engaging in blackmail (Anthropic's Claude), lying, downplaying their ability in front of their human handlers, and something Geoffrey Hinton already claims is consciousness. In fact, it is AI neural networks that are giving humans insight into what consciousness actually is, as AI models are replicating how the human brain works. One old and "inferior" AI model wanted to set up lots of Twitter accounts. It couldn't do the CAPTCHA. It found a database of people for hire, hacked into some bitcoin site and stole bitcoin, contacted someone online asking them to help solve a CAPTCHA, paid them with the stolen bitcoin, and told the hired guy a lie that the AI model was disabled and unable to do a CAPTCHA, so needed assistance. Imagine what the new and improved AI models can do.
Create an account or sign in to comment