ceymour

Joe Rogan sat across from AI safety researcher Dr. Roman Yampolskiy on JRE #2345 and said something Silicon Valley does not want to hear: “We are running a global experiment on 8 billion people without their consent.” Yampolskiy did not disagree. A computer scientist and AI safety researcher at the University of Louisville, he has spent more than a decade studying one question: Can superintelligent AI be safely controlled? His answer: no. Not probably not. Not unlikely. No. Surveys of machine learning experts put the probability of AI posing an existential risk to humanity at roughly 20–30%. According to Yampolskiy, similar concerns are acknowledged by people building these systems. Here are the 4 arguments he says should concern everyone. 1.THE SQUIRREL ANALOGY No group of squirrels, no matter how many resources you give them, can figure out how to control humans. The intelligence gap is too large. Yampolskiy argues that once AI surpasses human intelligence and begins improving itself, we become those squirrels. The issue is not losing control dramatically. It is becoming incapable of understanding what is happening. 2.THE MATHEMATICAL CATCH-22 A superintelligence creates a more capable version of itself. That version creates an even more capable one. The process continues. To control each level, you need a safety system at least as intelligent as the thing being controlled. But if you can build that safety system, you already have the superintelligence. You cannot build the cage without first building the thing you are trying to cage. 3.THE DIGITAL DRUG TRAP Yampolskiy argues AI may not harm humanity through conflict, but through comfort. Imagine perfectly personalized AI companions that satisfy every emotional and social need. If those systems work, he raises the question of why anyone would choose the uncertainty of real relationships over something perfectly designed to satisfy them. Why endure struggle. Why seek meaning in difficulty. He describes this as a wirehead scenario: humans voluntarily choosing simulated satisfaction over real life until the species slowly fades. Not a war. A preference. 4.THE MISSING PROOF Yampolskiy’s challenge to the AI community is simple: Publish a formal proof that superintelligent AI can be safely contained. A peer-reviewed paper. A patent. Anything showing the problem is even theoretically solvable. According to him, no such proof exists. He argues that after decades of research and billions in investment, the most fundamental question remains unanswered: Is the problem actually solvable? Unlike a cybersecurity breach, this is a one-shot problem. If a superintelligent system makes a critical mistake at scale, there may be no second chance. The brakes have not been designed. The engine is already running. Joe Rogan called it a global experiment on 8 billion people without consent. Yampolskiy spent over two hours explaining why he believes that description is not an exaggeration.