Uploads from Ontology Explained: Philosophy and AI

Currently Playing: Is AI Safety Just Alchemy? A Philosopher Reads “If Anyone Builds It,” Ch. 11

Is AI safety a science or an alchemy? In Chapter 11 of If Anyone Builds It, Everyone Dies, Eliezer Yudkowsky and Nate Soares argue that the people racing toward superintelligence don't understand what they're doing, and that you shouldn't bet human existence on alchemists. If Chapter 10 says we're up against the MonStars, Chapter 11 says all we've got is the Looney Tunes. (Unless we can get MJ to work on alignment?) In this video, I reconstruct the chapter's governing argument, walk through it section by section, and give my take as an analytic philosopher. IN THIS VIDEO • The governing argument, premise by premise • The alchemist's tournament: lead into gold, with your whole hometown on the line • Why Elon Musk and Yann LeCun get singled out • The rocket-safety vignette • Superalignment: can AI help align AI? The weak vs. strong versions • My response MY TAKE The chapter makes two good points. We need to be able to trust the competence of people working on hard problems that affect all of us. And using AI to solve alignment poses a real dilemma. But the case against the field rests on two people and some dated quotes. The vignettes spend six pages on basic points. And the "where's the outcry?" argument is heads-I-win, tails-you-lose. I also look at new interpretability work (Cammarata's "Neural Networks Are Already Clean") and explain why I still think AI doom is unlikely. I do agree with the authors on consent: if lab leaders like Dario Amodei really believe the risk is that high, it's hard to see how they justify pressing on. And yes, we find common ground on not handing Elon control of important things. This is part of my chapter-by-chapter series on If Anyone Builds It, Everyone Dies. Only three chapters left! Is AI safety alchemy or science? Tell me in the comments. BOOK If Anyone Builds It, Everyone Dies: Why Superhuman AI Would Kill Us All by Eliezer Yudkowsky & Nate Soares (2025) REFERENCES AP on the Musk/Tucker Carlson interview: https://apnews.com/article/elon-musk-tucker-carlson-ai-twitter-chatgpt-24119e28f10e495cf45494318d509096 Cammarata, "Neural Networks Are Already Clean": https://already-clean.goodfire.com/ Ontology Explained: an analytic philosopher and ontologist on AI and philosophy. Subscribe for more. #AISafety #Philosophy #ArtificialIntelligence