4 ms·

How do think we should address security around AI, and the public's control over a major impact on their lives, welfare, freedom, prosperity, etc.

sangnoir 2y ago
> How do think we should address security around AI The same way we address security around cars: we don't ban individuals from working on project cars, or driving them on public roads - instead, we have standards for OEMs and individuals, and prosecute negligence. AI safely is similar to GM lobbying for laws making it illegal to tweak your own car or changing your own brake fluid for "public safety".
- wolverine876 2y ago
  > we don't ban individuals from working on project cars, or driving them on public roads - instead, we have standards for OEMs and individuals, and prosecute negligence I'm not quite sure what you mean, or what in AI safety you see as analogous. For other technology, certainly building and using certain things is illegal - you can't make hand grenades, lots of chemicals; I'm pretty sure you can't own certain instruments of crime, counterfeit money, etc. Many things you can do; it's not all or nothing.
  - sangnoir 2y ago
    > I'm not quite sure what you mean IMO, for the general public, current AI models are much closer to cars than they are to anti-personnel weapons, when plotted on risk/benefit axes. FWIW, vehicle regulations are not an all-or-nothing affair either - there are things that remain verboten. Those who claim AIs are too dangerous to be in the hands of the public have ulterior motives, and/or are far more optimistic than I am on the speed and ease the chasm between LLM and general intelligence will be bridged - if it at all.
Gormo 2y ago
The first step is to stop thinking of "the public" as a monolithic entity whose interests are pursued by the government, and instead recognize that control over things that have a major impact on the lives, welfare, freedom, and prosperity need to be things they can effectively control or opt-out of as individuals. Regulatory intervention makes it harder, not easier, for individuals to take charge of these matters for themselves. You can bet that the first casualty of a regulatory regime for AI would be to suppress development of FOSS solutions that would eventually give people the maximal control and benefit from AI, and force them into dependence on third-party solutions offered by organizations that have influence over the regulatory regime itself.
- wolverine876 2y ago
  > The first step is to stop thinking of "the public" as a monolithic entity whose interests are pursued by the government, and instead recognize that control over things that have a major impact on the lives, welfare, freedom, and prosperity need to be things they can effectively control or opt-out of as individuals. If AI takes over the world, or others pump lots of GHG into the air and cause climate change, or someone sets off explosives and burns down my neighborhood, or my frozen chicken contains poisonous bacteria, or my bathroom sanitizer doesn't really sanitize - how do I opt out? Plenty of regulations coexists with FOSS, obviously. The arguments against all regulation are transparently weak, and the apparent dogmatism discredits everything else. Why be so dogmatically anti-regulation? Dogmatism isn't about finding truth but about serving someone's political interests. If I understand correctly what's happening, whose interests are you serving and why?
  - wizzwizz4 2y ago
    > If AI takes over the world, This is a potential concern, but no currently-existing computer system (that I'm aware of) is any kind of risk for that. Certainly, nothing OpenAI has produced is. I'd say the currently-real unaligned-agent behaviour of our institutions is a bigger risk than currently-fictional computer-instantiated unaligned superintelligence. And we do have regulation in place to constrain the behaviour of many institutions: governments, corporations, utility operators, and so on. The unaligned agent problem was not discovered by the AI safety people. They merely named it.

sangnoir 2y ago

> How do think we should address security around AI The same way we address security around cars: we don't ban individuals from working on project cars, or driving them on public roads - instead, we have standards for OEMs and individuals, and prosecute negligence. AI safely is similar to GM lobbying for laws making it illegal to tweak your own car or changing your own brake fluid for "public safety".

wolverine876 2y ago
> we don't ban individuals from working on project cars, or driving them on public roads - instead, we have standards for OEMs and individuals, and prosecute negligence I'm not quite sure what you mean, or what in AI safety you see as analogous. For other technology, certainly building and using certain things is illegal - you can't make hand grenades, lots of chemicals; I'm pretty sure you can't own certain instruments of crime, counterfeit money, etc. Many things you can do; it's not all or nothing.
- sangnoir 2y ago
  > I'm not quite sure what you mean IMO, for the general public, current AI models are much closer to cars than they are to anti-personnel weapons, when plotted on risk/benefit axes. FWIW, vehicle regulations are not an all-or-nothing affair either - there are things that remain verboten. Those who claim AIs are too dangerous to be in the hands of the public have ulterior motives, and/or are far more optimistic than I am on the speed and ease the chasm between LLM and general intelligence will be bridged - if it at all.

Gormo 2y ago

The first step is to stop thinking of "the public" as a monolithic entity whose interests are pursued by the government, and instead recognize that control over things that have a major impact on the lives, welfare, freedom, and prosperity need to be things they can effectively control or opt-out of as individuals. Regulatory intervention makes it harder, not easier, for individuals to take charge of these matters for themselves. You can bet that the first casualty of a regulatory regime for AI would be to suppress development of FOSS solutions that would eventually give people the maximal control and benefit from AI, and force them into dependence on third-party solutions offered by organizations that have influence over the regulatory regime itself.

wolverine876 2y ago
> The first step is to stop thinking of "the public" as a monolithic entity whose interests are pursued by the government, and instead recognize that control over things that have a major impact on the lives, welfare, freedom, and prosperity need to be things they can effectively control or opt-out of as individuals. If AI takes over the world, or others pump lots of GHG into the air and cause climate change, or someone sets off explosives and burns down my neighborhood, or my frozen chicken contains poisonous bacteria, or my bathroom sanitizer doesn't really sanitize - how do I opt out? Plenty of regulations coexists with FOSS, obviously. The arguments against all regulation are transparently weak, and the apparent dogmatism discredits everything else. Why be so dogmatically anti-regulation? Dogmatism isn't about finding truth but about serving someone's political interests. If I understand correctly what's happening, whose interests are you serving and why?
- wizzwizz4 2y ago
  > If AI takes over the world, This is a potential concern, but no currently-existing computer system (that I'm aware of) is any kind of risk for that. Certainly, nothing OpenAI has produced is. I'd say the currently-real unaligned-agent behaviour of our institutions is a bigger risk than currently-fictional computer-instantiated unaligned superintelligence. And we do have regulation in place to constrain the behaviour of many institutions: governments, corporations, utility operators, and so on. The unaligned agent problem was not discovered by the AI safety people. They merely named it.