As a postscript to yesterday’s post, I have been pondering the sudden resurgence of panic among the AI giants. The president belittles all warnings regarding AI as just another hoax, assuring us, as he did last week on a surprise public call to NVidia CEO Jensen Huang that: “It’s a scam. The robots are not going to take over the world. It’s not going to happen.”
Despite us knowing so little about how bad things can really go, I am quite sure that the prospect of robots taking over is not what keeps Dario Amodei up at night. I think we can be sure that with all the endless financial resources at their disposal, the AI companies are all conducting their own war games to test the worst scenarios they can conjure up. Since they are under no obligation to report such tests and publicizing the danger would only bring it closer to fruition, I assume that they know a lot more than we possibly could about AI’s potential harm. The kinds of scenarios that Amodei and many others in the know are gaming out probably involve the synthesis of toxic compounds and viruses. Among the more frightening possibilities is the synthesis of “mirror life”, which only two years ago was thought to be a decade or two away. Mirror life refers to organisms built of molecules whose structure is reversed, the mirror image of those naturally occurring molecules that make up life as we know it today. The concern is that a mirror image virus would go completely undetected by both plant and animal immune systems, which have evolved over millions years to react to naturally occurring molecules, and thus lead to a mass extinction event.
The obvious question “Why would anybody do that?” seems less and less relevant with each passing year. The unbridled nihilism the world eats for breakfast these days makes it very easy to imagine apocalypse “for the lulz”.
The odd nature of the current threat puts us in a game theoretic stalemate which was not the case during arms races of the 20th century. Building a deadlier bomb not only required an idea, it required rare materials, technology and skills that were generally beyond reach except for a tiny handful of state players. But now we approach a point where the idea itself is all one needs to cause great harm and such ideas are only one prompt away.
Walking back a technology will prove too difficult for the current batch of tech and state leaders, all of whom stand to lose big financially and reputationally if the AI race was even to simply slow down. The best example I know of in which a society colectively turned its back on a dangerous tool of power and control is already 400 years old and highly imperfect at that (Perrin 1979). Maybe that’s my next installment on topics I’m underqualified to write about.