[Problem profile] “Risks from power-seeking AI systems” by Cody Fenwick, Zershaaneh Qureshi

80,000 Hours - Narrations • July 01, 2025 • Solo Episode

View Original Episode

Guests

No guests identified for this episode.

Description

The future of AI is difficult to predict. But while AI systems could have substantial positive effects, there's a growing consensus about the dangers of AI. Narrated by AI. --- Outline: (01:42) Summary (02:34) Our overall view (05:16) Why are risks from power-seeking AI a pressing world problem? (06:45) 1. Humans will likely build advanced AI systems with long-term goals (11:23) 2. AIs with long-term goals may be inclined to seek power and aim to disempower humanity (12:26) We don't know how to reliably control the behaviour of AI systems (15:56) There's good reason to think that AIs may seek power to pursue their own goals (19:40) Advanced AI systems seeking power might be motivated to disempower humanity (22:43) 3. These power-seeking AI systems could successfully disempower humanity and cause an existential catastrophe (23:16) The path to disempowerment (27:56) Why this would be an existential catastrophe (29:35) How likely is an existential catastrophe from power-seeking AI? (32:17) 4. People might create power-seeking AI systems without enough safeguards, despite the risks (32:59) People may think AI systems are safe, when they in fact are not (36:13) People may dismiss the risks or feel incentivised to downplay them (38:59) 5. Work on this problem is neglected and tractable (40:30) Technical safety approaches (46:19) Governance and policy approaches (48:32) What are the arguments against working on this problem? (48:51) Maybe advanced AI systems wont pursue their own goals; theyll just be tools controlled by humans. (51:07) Even if AI systems develop their own goals, they might not seek power to achieve them. (54:49) If this argument is right, why arent all capable humans dangerously power-seeking? (57:11) Maybe we wont build AIs that are smarter than humans, so we dont have to worry about them taking over. (58:32) We might solve these problems by default anyway when trying to make AI systems useful. (01:00:54) Powerful AI systems of the future will be so different that work today isnt useful. (01:02:56) The problem might be extremely difficult to solve. (01:03:38) Couldnt we just unplug an AI thats pursuing dangerous goals? (01:05:30) Couldnt we just sandbox any potentially dangerous AI until we know its safe? (01:07:15) A truly intelligent system would know not to do harmful things. (01:08:33) How you can help (01:10:47) Want one-on-one advice on pursuing this path? (01:11:22) Learn more (01:14:08) Acknowledgements (01:14:25) Notes and references The original text contained 84 footnotes which were omitted from this narration. --- First published: July 17th, 2025 Source: https://80000hours.org/problem-profiles/risks-from-power-seeking-ai --- Narrated by TYPE III AUDIO . --- Images from the article: Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts , or another podcast app.

Audio