That some people apparently can't see this strat with AI "safety" laws is genuinely bizarre.
I don't think people give Meta enough credit for their roll in that. Microsoft/OpenAI got cozy and thought they had a huge moat made of $100K hardware to protect their $10B investment. Then Meta shadow drops Llama and within weeks regular folks were running GPT3 level LLMs on consumer hardware. That was such a huge win for opensource and consumers.
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".
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.
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.
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.
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?
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.
In other disciplines, this is known as the "principal-agent problem" [1] or "fiduciary risk" [2].
Like you, I'm far more worried about the behavior of human institutions controlled by third parties with their own agendas than I am about the behavior of software that implements stochastic algorithms.
The risk in both cases comes from people surrendering their own agency and autonomy to external parties, and we already live in a word where too many people put blind faith in dysfunctional institutions which are demonstrably "unaligned" with the actual values and interests of those they pretend to act on behalf of. Turning to those very institutions to stop people from relying on "unaligned" software seems absolutely insane.
----
[1]: https://en.wikipedia.org/wiki/Principal%E2%80%93agent_proble...
That strongly depends on the details. In particular, regulation that compel clear disclosure of information to individuals.
Imagine how impossible it would be to "take charge for yourself" if none of the food in the grocery store had any ingredient/nutrition/allergen information, and bottles of pills didn't tell you the active ingredients and amounts.
I can't think of many such things. Do you have any examples?
> There's not really a way for an individual to opt out of a corporation using AI in a way that negatively effects them.
Refrain from doing business with that corporation? If they are transgressing against your person or property, that's where the law comes into play.