Roman Yampolskiy - AI: Unexplainable, Unpredictable, Uncontrollable

Описание: In this presentation, Dr. Roman V. Yampolskiy provides a rigorous examination of the fundamental limitations of Artificial Intelligence, arguing that as systems approach and surpass human-level intelligence, they become inherently unexplainable, unpredictable, and uncontrollable. He illustrates how the black box nature of deep learning prevents full audits of decision-making, while concepts like computational irreducibility suggest we cannot forecast the actions of a smarter agent without running it – often until it is too late for safety. He asserts that there is currently no evidence or mathematical proof to guarantee that a superintelligent system can be safely contained or aligned with human values.


00:00 Introduction
00:49 Presentation
32:57 Q&A Starts - AI Safety via Control vs Motivation / Value Alignment
35:23 Indirect Alignment
37:34 The Person Effecting view
38:28 Indirect Normativity / CEV
40:09 Winter Intelligence Oxford 2012
40:31 What aspect of human-ness will remain after merging with AI?
41:48 Will AI ignore ethics constraints?
42:36 AI 2027
43:21 OpenClaw Moltbook
44:08 Linking AI and biblical prophecy
46:03 Omohundro's Basic AI Drives & Bostroms Instrumental Convergence
46:36 Goal Content Preservation
47:56 What are most poeple concerned about that may not be the biggest concern?
48:52 Can AI align to the better angels of human nature?
49:22 The landscape of possible minds - will AI appear at a random location, or are there convergences/attractor basins?
50:18 Random drift in mind design space, or convergences to attractors?
50:49 Will zombie AI be interested in aquiring sentience?
52:22 Will the "AI bubble" pop?
53:03 Non-existential risks of AI
53:51 General sentiement
54:22 Do we need buy in from most people to get AI safety research done?
55:19 Elevator speach to convince global leaders of AI risk
56:03 Market/geopolitical dynamics effecting AI governance coordination
56:58 Hugo de Garis' Species Dominance Debate
57:55 Would superintelligence stay a worthy successor?
58:28 How have you changed your mind? (i.e. about 99.9% doom)
59:32 Singleton vs quorum superintelligence
01:00:35 Extended mind hypothesis
01:01:29 Do you classify intrinsic motivation as a form of control?
01:02:38 Will AI want to traverse the moral arc of the universe?
01:04:02 Cosmic Hosts, Cosmic Collectives & Simulators - will AI only play nice when it meets another AI/civ/Simulator at or beyond it's own capability?
01:06:44 Another AI winter?
01:07:16 PauseAI - is there clear criteria for un-pausing? why not just stop?
01:08:45 Can narrow AI help solve the AI alignment problem?
01:09:33 Rice theorum
01:10:16 CoT - is chain of thought a reliable representation of what AI is thinking?
01:11:49 Is radical interpretability possible?
01:14:01 Debates
01:17:10 Don't ban AI, ban superintelligence
01:18:27 Is there any place in the space of possible minds where SI keeps humans around?
01:18:58 What would a guarantee of safe superintelligence look like?
01:20:00 Would superintelligence nessecitate epistemic humility?

See: https://www.scifuture.org/roman-yampo...


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