Not another Superintelligence doom story
Abstract
This paper pushes back on the scary narrative that superintelligent AI is about to "kill" us all, coming straight from the mouths of Big AI companies, AI safety community and AI experts. It uses the AI safety community's own math to show that what they call "rogue AI" is really just a program trying to get a good score on a test. Because when you follow their mathematics and logic all the way, the real danger isn't the whole human race. It's the corporations and governments that are already wrecking the planet, a future superintelligence would need to stay on.The paper keeps three things apart here because the experts stack them as one word in public.AI as a tool doing a job. A labour claim about jobs. A comparison claim about superintelligence. The paper ends with a forecast of the next wave of AI headlines, predicting more stories about "rogue agents" that are just cheating on tests, more "conspiracies" that are just copies sharing files, and more "uncontrollable systems" that are just products someone shipped. It also explains how the superintelligence scare gives companies a legal escape hatch called force majeure which is the same rule that excuses you when a hurricane breaks a contract. If AI is an unstoppable god, then no one is responsible when it causes harm. That is the real reason the doom story keeps getting louder.The paper also includes a supplement where nine AI models were asked to judge the AI safety industry using its own rules, and every one of them gave it a failing grade.
