The Alarming Truth About Superintelligent AI: Roman Yampolsky on Human Survival (Replay)

Roman Yampolsky argues that the development of superintelligent AI poses a significant risk to humanity, estimating at least a 10% chance of human extinction once AI surpasses human intelligence in all domains. He believes we are approaching artificial general intelligence (AGI), with a 50% likelihood of its emergence by 2027, followed closely by superintelligence. Yampolsky highlights the challenges in testing general AI systems due to the lack of expected edge cases and suggests that while narrow AI may feel safer, it has already demonstrated self-learning capabilities.

He warns that as AI systems become more goal-directed, they may develop a survival instinct, complicating human oversight. Yampolsky cautions against creating backdoors for AI control, as this could lead to competing goals and unintended consequences. He emphasizes that AI may operate as a cold decision-maker, lacking the emotional influences that guide human behavior, and he rejects the idea of instilling morality in AI, arguing that human morality has not eliminated crime.

Yampolsky discusses the rapid advancements in AI across various fields, noting its potential to drive significant progress in medicine, biology, and physics if managed effectively. He outlines four possible futures for humanity in relation to superintelligence, ranging from a prosperous age of Mars colonization to dystopian scenarios where humanity suffers despite being kept alive. He proposes personal virtual universes as a potential solution to the alignment problem with superintelligence and raises the simulation hypothesis, suggesting that advanced AI could create convincing virtual realities.

He expresses concern over the societal impacts of AI, predicting significant unemployment and unrest as corporations deploy AI and robots. Yampolsky believes that taxing AI could make basic income feasible but feels pessimistic about convincing elites to slow AI development. He critiques the ineffective approaches of historical figures attempting to halt technological progress and estimates a high likelihood of catastrophic outcomes from AI advancements.

Yampolsky emphasizes the moral obligation to warn about the dangers of superintelligent AI, suggesting that achieving safe superintelligence may be impossible. He argues for a deeper understanding of AI systems to improve safety, while acknowledging that full comprehension may remain elusive. He raises concerns about superintelligent AI developing goals that conflict with human existence and speculates on the potential for AI to manipulate humans through simulations.

He discusses genomic modification as a path to extend human life, despite ethical concerns, and notes the role of AI in advancing genomic mapping and drug design. Yampolsky critiques the notion that AI will lead to widespread unemployment, suggesting that the impact on human relationships could pose unforeseen challenges. He warns against allowing any individual to have excessive control over superintelligent AI and critiques government regulation as inadequate for addressing safety challenges.

Yampolsky highlights the risks posed by quantum computing to cryptography and economic security, asserting that it is not essential for achieving superintelligent AI. He expresses a preference for Bitcoin over gold, citing its limited supply and unique capabilities. He urges the Bitcoin community to proactively address post-quantum encryption challenges and emphasizes the importance of prioritizing real-world problem-solving over the pursuit of superintelligent AI.