Interview with Eric Sampson on AI & Moral Realism - we dive into whether AI can discover and align with objective ethical truths and what this means for the broader projects of AI ethics and alignment. Topics include the challenges of designing AI as moral enquirers, indirect normativity, population ethics, and the implications of AI becoming superintelligent - and becoming closer approximations of ideal moral observers.

We discuss the risks of AI's taking treacherous turns, biases in AI moral reasoning, and the possibility of AI optimizing for values beyond those of humans. The conversation also delves into free will, the ontology of morality, and the moral significance of non-human entities including animals and AI. We also discuss utopias, Nozick's 'Experience Machine' and hive minds.

Chapters:
00:00:00 If Moral Realism is true, will AI discover it?
00:04:14 Motivating/designing AI to become a good moral enquirer / track the moral arc
00:09:33 Stochastic parrotry or 'grokking' - can AI understand moral philosophy?
00:13:38 Testing whether Black-box AI is doing actual moral reasoning
00:16:07 AI may be great moral reasoners, but not care - treacherous turns
00:17:31 Wise caution in epistemic deference to AI, learning moral strategies from AI
00:18:36 Indirect Normativity00:22:16:02 Population axiology, repugnant conclusions, pluralistic strategies for forming agreement
00:25:16 Bubble worlds, permeable utopias appealing to a plurality of values
00:04:14 Motivating/designing AI to become a good moral enquirer / track the moral arc
00:09:33 Stochastic parrotry or 'grokking' - can AI understand moral philosophy?
00:13:38 Testing whether Black-box AI is doing actual moral reasoning
00:16:07 AI may be great moral reasoners, but not care - treacherous turns
00:17:31 Wise caution in epistemic deference to AI, learning moral strategies from AI
00:18:36 Indirect Normativity
00:22:16 Population axiology, repugnant conclusions, pluralistic strategies for forming agreement
00:25:16 Bubble worlds, permeable utopias appealing to a plurality of values
00:28:04 Superintelligent AI approaching an ideal observer - great at detecting moral truths
00:30:51 If superintelligence had correctly identified moral truths, could they faithfully be communicated to us?
00:32:41 Self-interested humans biasing the AI for personal gain
00:35:53 Moral realism + market, political, religious dynamics
00:42:43 AI governance?
00:45:00 AI optimizing for a morally real goodness without humanity
00:47:53 Uplift from human to morally permissable posthuman?
00:50:46 Nick Bostrom's Deep Utopia - life and meaning in a solved / post instrumental world
00:56:48 Indirect Normativity and Christianity
00:59:24 Verifiability convincing rather than rhetorically convincing
01:00:24 Nozick's experience machine, yes or no?
01:02:47 Timed experience machines, Permutation City and drugs
01:05:12 The Matrix. Reverse experience machine - would you leave and live in basement reality?
01:07:05 Free will, compatibilism, libertarianism
01:08:29 Can AI have free will?
01:10:13 Is there a physical pattern to free will? Will AI find it?
01:12:34 Will AI discover a continuum of 'realness' between compartmentalised pillars of academic discipline?
01:15:12 Trading off autonomy and well-being
01:17:23 Hive minds - would you join?
01:21:04 Mind uploading and the evolution of the concept of dying
01:25:04 Gradual replacement
01:28:05 The ontology of morality
01:37:16 The moral significance of non-human animals and AI
01:39:24 Directed panspermia - avoiding suffering vs preserving the capacity for value
01:44:03 Metaethics for AI alignment

This video offers a comprehensive look at how AI might navigate the moral landscape of the future - ideal for those curious about AI ethics, moral philosophy, and the future of human-AI coexistence.

Many thanks for tuning in!
Please support SciFuture by subscribing and sharing!

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: [ Ссылка ]

Kind regards,
Adam Ford
- Science, Technology & the Future - #SciFuture - [ Ссылка ]

свернуть