Anders Sandberg - Cyborg Leviathan: AI from the 17th Century to the Posthuman Future скачать в хорошем качестве

Anders Sandberg - Cyborg Leviathan: AI from the 17th Century to the Posthuman Future

Anders Sandberg - Living inside the cyborg leviathan: artificial intelligence from the 17th century to the posthuman future – Future Day 2026 Abstract: Being human is hard: we are stupid and somewhat selfish, yet need to work together with other stupid and selfish people with their own goals. We survive by building societies, filled with institutions and habits that help us solve these tough coordination problems. These institutions often act as extended cognition, allowing us to go far beyond individual power. We are to some extent living inside artificial intelligence systems, and they have enabled us to take control over the planet… as well as caused the worst disasters in history. As we build AI, we are also making something that can slip inside our extended cognitive systems and enhance them into literal cyborg systems. We need not just enough of “first order alignment” – getting AI to do things we want safely, but also “second order alignment” – AI that plays well with our societies and structures. Otherwise there is a real risk we may lose our own ecological niche and find ourselves in a world that may be safe and prosperous, yet unfit for human flourishing. If we play it right, however, we might become part of something far grander: a cyborg civilization able to reach full autonomy. Post: https://www.scifuture.org/anders-sand... This talk introduces Ander's co-authored new book 'Law, Liberty & Leviathan', which should be out this year. 00:00 Introduction 01:28 Cyborg leviathan talk starts 34:02 first & second order alignment 35:08 Price of Anarchy (PoA) - Game Theory 36:05 Acceptable trade off between the Price of Anarchy at an individual level (the Price of Sovereignty) and stable coordination 37:29 Virtue ethics & consequentialism - we want systems that are virtue apt 43:35 Deflationary account of normativity 45:53 Norms of a functioning society - moral realism, prudentialism, game theory, altruistic punishment 41:21 We may need a cyborg leviathan in order to coordinate to solve big problems 49:22 Identifying existing norms and coming up with new norms, norms via hyperstition? or wisdom 51:19 Value / norm drift - wise norm selection - mathematical sociology, psychohistory 53:34 Indirect Normativity vs the Marketplace of Ideas 57:01 First, second and third orders of alignment 58:36 Indirect alignment - instead of aligning to humans, align to grounded values/norms - laws 01:00:05 Will AI care? Alignment in the presense of love & the absence of lies 01:04:24 Should the cyborg leviathan have the virtue of love? 01:10:15 The symbol grounding problem & care 01:14:35 Grokking, few-shot learning, symbolic reasoning in LLMs 01:16:43 Neurosymbolic AI (NeSy) - paper: Emergent Symbolic Mechanisms Support Abstract Symbolic Reasoning in LLMs 01:19:24 Statistical pattern matching vs symbolic reasoning for reasoning quality and tracability - (important for mechanistic interpretability and transfer learning) 01:21:45 CoT (chain of thought) reasoning can be hallucinated or confabulated 01:23:37 Reasoning traces through lineages of silos in a cyborg leviathan .. is reasoning best done in latent space? 01:24:30 What would it be like to walk a mile in superintelligence's shoes? 01:26:39 What can't an ideal advisor or observer know? Many thanks for tuning in! Please support SciFuture by subscribing and sharing! Buy me a coffee? https://buymeacoffee.com/tech101z Have any ideas about people to interview? Want to be notified about future events? Any comments about the STF series? Please fill out this form: https://docs.google.com/forms/d/1mr9P... Kind regards, Adam Ford Science, Technology & the Future - #SciFuture - http://scifuture.org