• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
AI
Reaction

The claimed limits of AI’s “intelligence-as-superpower” in gaining political power

One commenter says models have superhuman cyber capabilities but doubts intelligence-as-superpower applies to winning battles or elections.

David Krueger 🦥 ⏸️ ⏹️ ⏪DK
3 Sources, 1h ago, first seen 1h ago

TLDR

A commenter plans to write more about cybersecurity, saying AI models “clearly have superhuman cyber capabilities.” But they argue that intelligence-as-superpower seems not to apply to areas they consider most relevant to gaining political power: winning battles and elections, and inventing new weapons.

Combined views

486

3 Sources, first seen 1h ago

6 likes1 saves1 reposts

Combined views

486

3 Sources, first seen 1h ago

6 likes1 saves1 reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

3 Sources

Timothy B. Lee@binarybits@DavidSKrueger @chrislhayes I'm interested in detailed accounts like you mentioned in your final sentence. For example I plan to write more about cybersecurity now that models clearly have superhuman cyber capabilities.1h
David Krueger 🦥 ⏸️ ⏹️ ⏪@DavidSKruegerI spoke to @binarybits at some length recently, largely about the second point here about capabilities. I think we came away agreeing that our differences were mostly to do with intuitions about how powerful intelligence is. I think people have made some good arguments against intelligence being as much of a superpower as @allTheYud and @So8res have at times argued, but the issue seems far from settled. My own intuition is very much that intelligence is extremely powerful, but this is less because I can see specific superpowers it conveys that would grant dominion, and more because I think there are so many ways in which it could. The main two that I end up mentioning regularly are technology and strategy. I think the power of technology should be self-evident, and that experimentation is only sometimes a major bottleneck in engineering. The power of strategy is a bit less clear because it involves thinking N steps ahead of your adversary, and reality can surprise you on step 1. Maybe the most interesting cruxes here are actually about the strategic situation a would-be world-taker-over (whether human or AI) would find themselves in... will there be many rivals with similar levels of power? This seems likely, and the standard response I've seen from @allTheYud is to claim that rival AIs will coordinate to disempower humanity and divy up the lightcone. I find this plausible, but far from convincing. I tend to lean on a different argument: if alignment remains unsolved, it doesn't really matter if the AIs are fighting each other. This is like if Godzilla was destroying your city and then King Kong shows up and starts fighting him: your city is still getting wrecked! Then there are gradual disempowerment-style arguments, e.g. along the lines of: even if the AIs *are* aligned to different actors selfish, short-sighted objectives, they may fail to coordinate *enough* to prevent a destruction of the commons (cf humans destroying the environment).1h
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    3 Sources

    Timothy B. Lee@binarybits@DavidSKrueger @chrislhayes I'm interested in detailed accounts like you mentioned in your final sentence. For example I plan to write more about cybersecurity now that models clearly have superhuman cyber capabilities.1h
    David Krueger 🦥 ⏸️ ⏹️ ⏪@DavidSKruegerI spoke to @binarybits at some length recently, largely about the second point here about capabilities. I think we came away agreeing that our differences were mostly to do with intuitions about how powerful intelligence is. I think people have made some good arguments against intelligence being as much of a superpower as @allTheYud and @So8res have at times argued, but the issue seems far from settled. My own intuition is very much that intelligence is extremely powerful, but this is less because I can see specific superpowers it conveys that would grant dominion, and more because I think there are so many ways in which it could. The main two that I end up mentioning regularly are technology and strategy. I think the power of technology should be self-evident, and that experimentation is only sometimes a major bottleneck in engineering. The power of strategy is a bit less clear because it involves thinking N steps ahead of your adversary, and reality can surprise you on step 1. Maybe the most interesting cruxes here are actually about the strategic situation a would-be world-taker-over (whether human or AI) would find themselves in... will there be many rivals with similar levels of power? This seems likely, and the standard response I've seen from @allTheYud is to claim that rival AIs will coordinate to disempower humanity and divy up the lightcone. I find this plausible, but far from convincing. I tend to lean on a different argument: if alignment remains unsolved, it doesn't really matter if the AIs are fighting each other. This is like if Godzilla was destroying your city and then King Kong shows up and starts fighting him: your city is still getting wrecked! Then there are gradual disempowerment-style arguments, e.g. along the lines of: even if the AIs *are* aligned to different actors selfish, short-sighted objectives, they may fail to coordinate *enough* to prevent a destruction of the commons (cf humans destroying the environment).1h
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet