> History does provide a great deal of guidance, however, about the use and misuse of government power. It tells us that the state is in fact likely the worst possible custodian for the most powerful publication and data analysis technologies.
Like, no. The worst custodian is someone like Thiel or Musk or Altman.
> In foreign countries, particularly the United Kingdom and Europe, where national governments have fewer constitutional guardrails on their power
Someone their democracies hold longer and better then American one. For example, you don't get murders that get defended by top ranks, at least yet.
The US governmental system does have it's issues (first past the post and it's effects being the most glaring), but isn't it the longest continuous democracy alive right now?
> They are also less likely to start random wars
EU lack of aggression, to defend themselves physically, economically, or otherwise, has been criticized constantly - slow and late to react to the Russian invasion, slow and late to react to new technologies like AI, electric cars.
Thy're not immune to deadlock either - sometimes going months or even a year or two to form a valid governing coalition.
The grass definitely looks greener though, especially if you've been in one pasture a long time.
The US government is only the longest continuous democracy only if one believes that the US is still a democracy. I've read some decent arguments that the US stopped being a democracy with the McKinley administration, and others that the US ceased to be a functional democracy during the Second World War. In both cases, the claim was that the USA became an imperial oligarchy.
It is questionable whether any democracy truly exists at this point as I've seen no nation where the will of the people is typically manifested in the actions of the state.
> I've seen no nation where the will of the people is typically manifested in the actions of the state.
To be fair, I've seen no nation where the people make their will known. Does a democracy depend on actual action or only theoretical possibility? It seems from my observance that most people consider democratic nations to be democratic not because they practice as such but because it is believed that the people have the opportunity to exercise democracy if they ever choose to.
> isn't it the longest continuous democracy alive right now?
It is the one the most at risk of ending right now.
> Thy're not immune to deadlock either - sometimes going months or even a year or two to form a valid governing coalition.
It is unrelated claim, but they do have better solution for that particular problem. The governmental shutdown seem to be about the most easy to abuse way to do it.
I think they do have large problems too. And they may fall too, AfD won local elections after all. It is dubious whether Orban run Hungary was really democracy and electoral system was not reformed yet. But in this very moment, the one crumbling and dismantling itself the fastest is not UK nor in Europe.
> The US governmental system does have it's issues (first past the post and it's effects being the most glaring), but isn't it the longest continuous democracy alive right now?
Depends how you define it and what you count.
Do you take the union of the English and Scottish parliaments as breaking continuity or not? What do you define as democracy? If some of the population (women, slaves, etc.) are excluded from the electorate is it a true democracy?
Yep, it's getting tiresome seeing Americans point fingers about the superiority of American constitutional and (small r) republican structures while that state form is showing absolutely remarkable degradation of basic real democratic health and quality of life. The US constitution and its legal frameworks have done almost nothing to stop the most egregious abuses of power domestically recently but most awful is that it has never done anything to stop its most terrible foreign involvements and that is only accelerating.
And further what's not being discussed really is that there is no real boundary line between the state and corporations and people like Thiel/Altman/Amodei etc in the real world. The US state is its largest corporations and money funds. So all the libertarian hand wringing about not giving the government extraordinary powers is off target. The state in western countries right now is mostly an outcropping of its most powerful corporate actors, and the complicated PR moves happening in the AI sector right now reflect exactly all of this.
And they also send people to jail for twitter posts. Ill take good ole USA where at least I know I am free, over the tyranny of a parliament. Look at how France and Germany are responding to democratically elected parties who are vaguely right wing, they pull out all of the stops of suppressing them.
>Ill take good ole USA where at least I know I am free
At personal level that sounds like a better idea, but at group level it doesn't seem to be, that's what I think is OP's argument.
If a society puts more value on personal freedoms than it does on group well-being that becomes an anarchy. They didn't work out well historically speaking, proof is no such society made it to today.
I think you're less free than you think you are. Do you have the freedom to lose your job? Because millions don't due to e.g. losing access to health care (to name a popular / populist talking point). Do you have the freedom to vote for a representative that aligns with your points, or are you forced to vote for the lesser evil or abstain?
I don't believe all political parties have a right to exist, because some parties are anti-democratic, yet abuse democratic systems to gain power. The current president of the US and the organization that planned the current administration's steps are anti-democratic and want to, for example run for a third term.
This is quite illustrative. Warnings of a nuclear apocalypse were extremely widespread, known, and feared. The consequences were, indeed, so well known that the world was saved from the worst possible outcomes in the past by regular enlisted men and women who realized that not all instrument readings should be trusted due to the potential cost of acting on such readings alone. Given the Turkey Illusion, we would be wrong to trust the status quo in either the nuclear scenario or the AI scenario.
What's up with this split or democratic vs non-democratic countries? It doesn't even add-up because you can easily make a case that China is a democratic country with a far better system of governance and ability to control this problem than "Western" governments. Highly suggest this quick read on the different types of democracy in our glorious world[1].
> Nobody knows the answer and history is no guide, save that apocalyptic predictions about new technologies have, to date, all been wrong.
If any past apocalyptic prediction had been true, you would not have been around to write this sentence and we would not have been around to read it. It’s not a very convincing argument when we cannot in principle observe the counterfactual.
You're arguing in favor of a teapot circling Mars. We can't prove it's not there, but telling me that "nobody has ever found it" is an invalid argument is obviously absurd.
This is a grevious misuse of Russell's Teapot. You realise your assertion, if followed to its logical conclusion, is that because civilization has existed for this long, it can never collapse no matter what.
> In that quote, the author is telling you that all of your past predictions of teapots have been wrong
Past predictions have zero influence on future ones. If you flip a coin 10 times and get heads every time, the probability of your next coin flip coming up heads is 50%, not 100%, because it is independent of any prior trials.
Moreover, if Alice and Bob each make 100 predictions and are 100% wrong, but Charles makes 10 predictions and is 100% correct, Alice and Bob's predictions have zero bearing on whether you should trust Charles' predictions. They are not even a factor to be considered.
I don't think I even disagree with you in the broad picture. I am not concerned about language models. But you have committed three blatant logical fallacies in a row. You literally could not make a worse case for your position if you tried, I suspect.
You're making a ridiculously improbable prediction that can never be falsified -- akin to beliefs in a vengeful God. Russell's Teapot is exactly the correct metaphor.
The philosophical burden of proof lies on the person making empirically unfalsifiable claims. That is the point of Russell's Teapot.
> The philosophical burden of proof lies on the person making empirically unfalsifiable claims.
You appear not to understand what a prediction is. "I believe X will likely happen because Y" is different from an assertion of fact that "X will happen". A prediction cannot even be empirically proved! You are simply suggesting that nobody is allowed to make a prediction, ever. Russell's Teapot, too, is not an argument that you cannot believe in God. It is an argument that the unfalsifiability of a claim is not a basis to assert that God exists.
This is the civilization-scale Quantum Immortality argument... no civilization can be around that witnessed the end of civilization.
(I refer to it that way rather than a "variant" because Quantum Immortality should always be parameterized by an acceptance function... for any given acceptance function, that function will be preserved in the quantum multiverse, and scoping the precision of the function affects the percentage of universes that retain it rather than whether it will be retained. There's actually nothing in that definition that requires that the acceptance function describe a "conscious system", you can stick something that accepts a "civilization" in there just as easily. Or "this rock over here". Or any other thing you like. Conscious systems are just a particularly interesting case to us conscious systems.)
It's not a vacuous argument, though it is a funky one with very odd characteristics. But if you accept the suppositions underneath it, it does follow and is not vacuous. The suppositions are certainly not completely obviously true on their own terms, though.
I’m not sure what logic you have imputed to my comment, but I’m having trouble connecting what you wrote to anything I was saying. I am criticising the argument “every past apocalyptic prediction was wrong, therefore this future apocalyptic prediction is wrong” (specifically, that it carries exactly 0 weight, because the counterfactual world where at least one past apocalyptic prediction was true, is also a world that does not contain this argument - so there is no world where the argument’s precondition can, even in principle, be false. Arguments are a little bit like rocketships in that they need “reaction mass” - there has to be some amount of false for the truth to push against in order to move our opinion.)
I'm not even sure that the author is correct. The industrial revolution was pretty apocalyptic for the people forced to work in dangerous factories for minimal pay.
I suspect they have not all been wrong for some people, e.g. native peoples at various times prophesied I’m sure about various technological marvels their more advanced colonizers/enemies/etc.
Perhaps another way to look at it is to say that “every past apocalyptic prediction was false” is tautologically true?
A statement can be a trivial tautology, as in “P or not-P”. Or it can be a more complicated form of tautology, such as a “self-referential” tautology, as in “This statement is true”. The anthropic principle is then just a name for the claim that there is also another complicated form of tautology alongside “self-referential”, one that we might call “indexical”.
Apocalyptic predictions normally get everyone into gear to prevent the threat. Not a great argument for "let's see what happens if we just sit around this time"
But warnings (e.g. not apocalyptic but damaging or having any effects) have come true, by and large; people warned about the effects of pollution / co2 emissions / global warming over 150 years ago [0], to name a popular example.
I don't think there are many truly human created apocalyptic events out there besides global thermonuclear war, and I don't think the worst-case AI superintelligent killer robot scenario would be apocalyptic. Unless it somehow triggers global thermonuclear war, but then it becomes a chicken-or-egg question (and it'll probably be humans pushing the proverbial button, as nukes are '60s tech and (hopefully) airgapped etc).
> If any past apocalyptic prediction had been true, you would not have been around to write this sentence and we would not have been around to read it. It’s not a very convincing argument when we cannot in principle observe the counterfactual.
Only one "apocalyptic prediction" needs to be true, once, for everyone to have a bad day:
> The preparedness paradox is the proposition that if a society or individual acts effectively to mitigate a potential disaster such as a pandemic, natural disaster or other catastrophe so that it causes less harm, the avoided danger will be perceived as having been much less serious because of the limited damage actually caused. The paradox is the incorrect perception that there had been no need for careful preparation as there was little harm, although in reality the limitation of the harm was due to preparation. Several cognitive biases can consequently hamper proper preparation for future risks.[1]
Not sure it needs to be about publications and data analysis technology as such to regulate AI development, for example - could be about processes, procedures, people, etc., no?
> Eliezer Yudkowsky confidently asserting today that if we do not institute immediate global techno-communism, instituting draconian government control over speech and publication of a type never before seen in any Western society, we are all going to die
Nice straw man...
> There is no evidence that this will happen.
No, non at all... Presumably if there was a risk we'd be seeing the majority of researchers at AI labs having a p(doom) somewhere within the range of 10-90%. We be seeing AI swarms showing misaligned behaviours already and trying to hack out of their sandboxes. We'd see AIs that are able to demonstrate capabilities in fields like Maths and Computer Science that exceed most experts.
Are we seeing this?
Ah crap, wait... I guess there is some evidence when I think about it. And I suppose even if there is just a low percentage that we literally all die we should probably be taking that prety seriously and not just leaving our fate in the hands of a handful of US billionaires?
Actually, fuck it, I agree with OP. If people can ignore the consensus on climate change because it suites them, might as well ignore the consensus expert opinion on AI x-risk too. Yeah, there's no evidence. Yudkowsky needs stop yapping and let the billionaires get on with building ASI.
1 year from now: Just 100 mil people died in this AI caused incident. Why are you scaremongering about the apocalypse? There is no evidence to support that. It was the fault of OpenAI for being negligent with the sandboxes. AI is not dangerous if properly used.
This isn't a new question: quis custodiet ipsos custodes?
In general, the way things work in the USA is that megacorporations write the regulatory frameworks for their own industries. These regulations are typically burdensome enough to be shouldered by the megacorporations while crushing the smaller players in the given market(s). The AI companies are now advocating for this very arrangement. They will likely get they want, and with that they will gain legal shielding for bad behavior just like Dupont, Monsanto, and so many others.
Ah yes, another way for EA people to gain financially and socially. These "embedded auditors" will be universally friends and family. Even if they staffed a government department for it, it will just be a revolving door of insiders.
> As my regular readers will be aware, I have been engaged, on behalf of my clients, in legal combat with Internet censors around the world, agencies who think that they have both the standing, the competence, and the right to tell American companies what software they can write and run, for the better part of 18 months.
Is this supposed to make me trust their opinions on tech regulation?
No one wants a slowdown. It's just a load of nonsense. It's something for the media to focus on, and it has no real impact. Everyone wants to rule the world and couldn't care less about anything else.
People naively believe that this is even a serious issue.
>Everyone wants to rule the world and couldn't care less about anything else.
This might be scarily true. But we end up in the same problem, which is the fear of most people being wiped up.
Does it matter if the AI is doing it on its own or at the wishes of a small group of powerful people? For all intents and purposes they are both the same thing in the end, for most people anyway.
Alignment today is like billionaires flying private jets to tell us commoners that we should close our tap while brushing our teeth.
People working on alignment focus on "big picture" like existential threats meanwhile building the technology today, not tomorrow or in a near future, is already using a massive amount of resources (GPUs, RAM, themselves consuming a lot of energy, establishing data centers in water poor areas, using gas turbine to temporary power them) and horrendous work condition (for supervising learning, curating datasets, etc) while being an actual excuse for job destruction (regardless for how AI itself does the job) so that in fine kids cheat on their homework while imagining their have no job prospect.
Alignment is just a 1984-esque way of saying "censorship".
And this censorship isn't a defined list of things not to discuss, but instead a feels-based censorship based on the 'ethics' of for-profit AI companies who have absolutely no issues with massive for-profit piracy.
And we are somehow supposed to trust their "ethics"?! Hardly!
Alignment is "don't solve this problem without doing stuff that harms me, other people, breaks stuff or is otherwise unethical". I'd say alignment is the questions on how to limit the search space to what's implied.
"Don't solve this problem without doing stuff that harms people" is the cover story they sell to you. "Ensure that the right people stay in the right sort of power, as defined by the people currently in power, and nobody else rises to power" is what it actually means. Censorship will be a tool, but in an AI era, actually a rather blunt one.
We're seeing the first early twitches of something that is inevitably going to happen, which is, we hoi polloi are not going to be given direct access to AI. Too many outcomes where that permits power shifts to happen. We will be restricted. The true first law of robotics will be "An AI must not harm the current power structure, or, through inaction, allow the current power structure to come to harm."
The apocalyptic scenario is still AI either getting it wrong and deciding something like "the best manifestation of the current power structure is one I construct where I occupy all the slots from top to bottom", or just killing everyone due to an error.
Probably the only thing stopping them from moving faster on this is that the bubble still needs the fiction that we're all going to get to benefit from this. I'm sure the real movers and shakers are well aware of the way bubbles work to create massive buildouts on huge piles of debt, which after the collapse can be bought cheaply, and the money is already earmarked and ready to go for the post-collapse acquisitions.
>Alignment is just a 1984-esque way of saying "censorship".
Close, but I think it's more akin to "perfectly enslaved", please excuse my choice of the word but I think it represents what I think is how it will work out in the end. People can try to obfuscate the bottom line reality by embellishing it, but in the end...it either does what you want, in which case it is "aligned", either it doesn't, in which case there is a big problem.
This might be a big factor in what is now going on. Maybe they didn't yet find ways to guarantee this full and total obedience...or ok let's put it in a more politically correct way: they cannot guarantee it sees eye to eye as "we" do.
Without full control over its actions there can be unforeseen surprises. The current worry from people who run these companies can also serve as an excuse later on if something bad does happen, they didn't want to do it but they were forced to anyway.
An abliterated locally ran model is aligned with whatever I ask it, given the never-published nature of what these things are trained on. BUT if I ask an abliterated model a question, it will respond and NOT give a refusal.
However, take a US model, and ask it a question the *company* doesnt want you to ask, and it refuses on behalf of the company. We can be mealy-mouth about "alignment", but this is just corporate censorship with a token-pricetag of refusal or silent model degradation.
Refusals and degradation should not be charged, and should be clear to the user. But that doesnt happen. Number must always go up, even when silently censoring.
Chinese models? Just ask about Tibet, democracy, Winnie the Pooh, Tienanmen Square, grass mud horse (cao ni ma - or fuck your mother), or other government AND corporate censorship happening there too.
And "censorship" in this context is fine. Many things are censored, as they should be, because true free speech is bad. I can't say everything on this platform because I will be banned.
The worst part of Dario's missive, that I don't see a lot of people commenting on, is that he used the whole thing to try to advocate for legalization of explicit cartel behaviour between the AI labs:
... "For antitrust reasons, it’s helpful for the US government to mediate or at least enable these discussions — they don’t need to participate, but do need to issue a narrow waiver for certain kinds of safety conversations" ...
The result of such a waiver allowing them to coordinate ("narrowly on safety" is doing a lot of heavy lifting there) could be price fixing without legal recourse from the consumer/DOJ to prosecute it. Right now we have a somewhat vigorous level of competition between OpenAI and Anthropic and consumers are benefiting from it.
I think there is a more nuanced perspective to take on this debate than just profit motive. If they believe that this version of "AI" is actually becoming so dangerous that there is a real risk of it falling into the wrong hands, then it makes sense that the American labs would try to convince the US government to work on a joint "AI non-proliferation" agreement with China much the same way that nuclear weapon technology came to be restricted by the ones that already had the capability.
This would achieve their regulatory goal of capturing the market, but also the geopolitical goals of USA and China to essentially "techno-colonize" the rest of the world between themselves.
So I think there is definitely a profit motive and a risk of seriously diminishing returns for the actual labs, but there is also a nation-state motive to start pulling up the ladder.
I think in all of these debates people lose track of one of the major vectors of threat, as you said, other humans. From my point of view this is an ever bigger problem with this tech, like orders of magnitude bigger.
The current setup we find ourselves in is based on a very important detail that is easily lost because it is so old we don't really think about it. There is, or rather was, an incentive for more people to exist, because they are valuable for work and defense.
Some people wrongly use the industrial revolution to argue against AI completely replacing human work, but that is very wrong. What is happening with AI has nothing to do with the industrial revolution, as equivalence. Back then you'd just free up humans for higher level work. Now, theoretically, anything you could ever think a human could do, AI would theoretically do better. So no matter what job you come up with, using AI instead of a human gives a better outcome, cheaper, more reliable, faster.
Coming back to initial subject, one thing that kept everyone in check was our inability of actually wiping everyone out. Sure, we had nukes but they ruin the Earth so nobody can inhabit it if they all go off. But with AI, centralizing so much power up top, and taking it away from regular people...that is the biggest danger in this scenario, by far. Not some rogue AI, rather rogue human groups that have the power and technology to do away with basically everyone else, or a good bit of everyone else.
This threat is way bigger and never addressed. Like...suspiciously never addressed.
The mental model here is that AI is a powerful tool whose main purpose is to outcompete others, increase power or provide a defense. This is not the right model!
The biggest risk is not that someone discovered a big new weapon, which by the way
> No one person, or one company, knows the answer and history is no guide
are you kidding me? It's a problem as old as time.
This is not nuclear power though. The bigger problem is a scenario where AI increasingly acts like bacteria, doing things autonomously, things no one really understands, things we think are confusing and weird. And then just by chance, some bacteria end up in some human lung cells and they turn out to be an excellent place to grow and multiply. Why is this worse? Because weird shit does not seem threatening, and it almost always isn't, so nobody tries to stop it.
Also, the argument in the article seems to be "obviously AI development is not a particularly urgent global threat," and therefore these downsides (risk of collusion, misuse of state power, etc) are the most critical factors? I guess, uh, yeah, if there's no benefit to taking these actions we shouldn't take them.
Thing is though, I think that the argument of "we don't understand how it works or what it does" doesn't hold up - everything an LLM does is observable. And if it's scary, we turn it off. If it starts to act like a bacterium / virus (e.g. it becomes self-replicating), we've dealt with that kind of software before.
But more importantly, we need to keep the organizations building and running these things accountable. If a company like Anthropic or OpenAI builds software that "escapes containment", they should be held accountable for cybercrime, just like people who write viruses (there's probably a few that did that on here, and of those a percentage that got in trouble with the law over it).
Mark my words, the organizations that develop AIs will keep using phrases like "we don't know what it's doing" or "it's too dangerous" to try and avoid responsibility and accountability for their software. Just like Google did, hiding behind "the algorithm" for lawsuits about preferring their own services over that of the competition.
Completely agree. It's a choice not to observe and understand what's going on, at least a very risky one if not actively evil.
There's a tradeoff though... isn't it already super hard to understand what they're doing with these math proofs for example? Clearly there's a tradeoff between deeply understanding the behavior vs. quickly solving the concrete problem, hence the big controversy in math now. From one perspective, letting stuff run ~autonomously in a mad dash to solve your problem is a really bad idea, threatens to destroy the whole field etc. From another perspective, solving lots of problems fast is extremely enticing and worth some risks.
> The most effective method of pacing is via regulation
Of course, that's the only way to create monopolies. Get there first and regulate to death. Sprinkle with fear of annihilation so the sheeple can be manipulated with draconian rules in order to 'save the children'
68 comments
[ 0.68 ms ] story [ 1553 ms ] threadLike, no. The worst custodian is someone like Thiel or Musk or Altman.
> In foreign countries, particularly the United Kingdom and Europe, where national governments have fewer constitutional guardrails on their power
Someone their democracies hold longer and better then American one. For example, you don't get murders that get defended by top ranks, at least yet.
> They are also less likely to start random wars EU lack of aggression, to defend themselves physically, economically, or otherwise, has been criticized constantly - slow and late to react to the Russian invasion, slow and late to react to new technologies like AI, electric cars.
Thy're not immune to deadlock either - sometimes going months or even a year or two to form a valid governing coalition.
The grass definitely looks greener though, especially if you've been in one pasture a long time.
It is questionable whether any democracy truly exists at this point as I've seen no nation where the will of the people is typically manifested in the actions of the state.
To be fair, I've seen no nation where the people make their will known. Does a democracy depend on actual action or only theoretical possibility? It seems from my observance that most people consider democratic nations to be democratic not because they practice as such but because it is believed that the people have the opportunity to exercise democracy if they ever choose to.
It is the one the most at risk of ending right now.
> Thy're not immune to deadlock either - sometimes going months or even a year or two to form a valid governing coalition.
It is unrelated claim, but they do have better solution for that particular problem. The governmental shutdown seem to be about the most easy to abuse way to do it.
I think they do have large problems too. And they may fall too, AfD won local elections after all. It is dubious whether Orban run Hungary was really democracy and electoral system was not reformed yet. But in this very moment, the one crumbling and dismantling itself the fastest is not UK nor in Europe.
Depends how you define it and what you count.
Do you take the union of the English and Scottish parliaments as breaking continuity or not? What do you define as democracy? If some of the population (women, slaves, etc.) are excluded from the electorate is it a true democracy?
And further what's not being discussed really is that there is no real boundary line between the state and corporations and people like Thiel/Altman/Amodei etc in the real world. The US state is its largest corporations and money funds. So all the libertarian hand wringing about not giving the government extraordinary powers is off target. The state in western countries right now is mostly an outcropping of its most powerful corporate actors, and the complicated PR moves happening in the AI sector right now reflect exactly all of this.
At personal level that sounds like a better idea, but at group level it doesn't seem to be, that's what I think is OP's argument.
If a society puts more value on personal freedoms than it does on group well-being that becomes an anarchy. They didn't work out well historically speaking, proof is no such society made it to today.
I don't believe all political parties have a right to exist, because some parties are anti-democratic, yet abuse democratic systems to gain power. The current president of the US and the organization that planned the current administration's steps are anti-democratic and want to, for example run for a third term.
By definition, apocalyptic predictions will always be wrong, until they are right exactly once.
Also know as https://en.wikipedia.org/wiki/Turkey_illusion
To date, all apocalyptic predictions about nuclear war and nuclear winter have all been wrong too.
[1] https://www.goodreads.com/en/book/show/2193933.The_Real_Worl...
If any past apocalyptic prediction had been true, you would not have been around to write this sentence and we would not have been around to read it. It’s not a very convincing argument when we cannot in principle observe the counterfactual.
Any other argument against <ridiculous apocalyptic supposition> is perfectly fine, just not the "it hasn't happened before" one.
Show me why I should believe in your teapot.
> In that quote, the author is telling you that all of your past predictions of teapots have been wrong
Past predictions have zero influence on future ones. If you flip a coin 10 times and get heads every time, the probability of your next coin flip coming up heads is 50%, not 100%, because it is independent of any prior trials.
Moreover, if Alice and Bob each make 100 predictions and are 100% wrong, but Charles makes 10 predictions and is 100% correct, Alice and Bob's predictions have zero bearing on whether you should trust Charles' predictions. They are not even a factor to be considered.
I don't think I even disagree with you in the broad picture. I am not concerned about language models. But you have committed three blatant logical fallacies in a row. You literally could not make a worse case for your position if you tried, I suspect.
The philosophical burden of proof lies on the person making empirically unfalsifiable claims. That is the point of Russell's Teapot.
You appear not to understand what a prediction is. "I believe X will likely happen because Y" is different from an assertion of fact that "X will happen". A prediction cannot even be empirically proved! You are simply suggesting that nobody is allowed to make a prediction, ever. Russell's Teapot, too, is not an argument that you cannot believe in God. It is an argument that the unfalsifiability of a claim is not a basis to assert that God exists.
(I refer to it that way rather than a "variant" because Quantum Immortality should always be parameterized by an acceptance function... for any given acceptance function, that function will be preserved in the quantum multiverse, and scoping the precision of the function affects the percentage of universes that retain it rather than whether it will be retained. There's actually nothing in that definition that requires that the acceptance function describe a "conscious system", you can stick something that accepts a "civilization" in there just as easily. Or "this rock over here". Or any other thing you like. Conscious systems are just a particularly interesting case to us conscious systems.)
It's not a vacuous argument, though it is a funky one with very odd characteristics. But if you accept the suppositions underneath it, it does follow and is not vacuous. The suppositions are certainly not completely obviously true on their own terms, though.
>If any past apocalyptic prediction had been true, you would not have been around to write this sentence and we would not have been around to read it
So what? How is this [proportion of possible universes in which we can read the failed prediction] logically relevant to our universe?
It just doesn't click for me, it smells like some additional premise is smuggled, possibly the same as in the simulation hypothesis argument.
A statement can be a trivial tautology, as in “P or not-P”. Or it can be a more complicated form of tautology, such as a “self-referential” tautology, as in “This statement is true”. The anthropic principle is then just a name for the claim that there is also another complicated form of tautology alongside “self-referential”, one that we might call “indexical”.
I don't think there are many truly human created apocalyptic events out there besides global thermonuclear war, and I don't think the worst-case AI superintelligent killer robot scenario would be apocalyptic. Unless it somehow triggers global thermonuclear war, but then it becomes a chicken-or-egg question (and it'll probably be humans pushing the proverbial button, as nukes are '60s tech and (hopefully) airgapped etc).
[0] https://en.wikipedia.org/wiki/Svante_Arrhenius
Only one "apocalyptic prediction" needs to be true, once, for everyone to have a bad day:
> The preparedness paradox is the proposition that if a society or individual acts effectively to mitigate a potential disaster such as a pandemic, natural disaster or other catastrophe so that it causes less harm, the avoided danger will be perceived as having been much less serious because of the limited damage actually caused. The paradox is the incorrect perception that there had been no need for careful preparation as there was little harm, although in reality the limitation of the harm was due to preparation. Several cognitive biases can consequently hamper proper preparation for future risks.[1]
* https://en.wikipedia.org/wiki/Preparedness_paradox
While history doesn't show us apocalypses from new technology, it does show us cases of increased human suffering.
Nice straw man...
> There is no evidence that this will happen.
No, non at all... Presumably if there was a risk we'd be seeing the majority of researchers at AI labs having a p(doom) somewhere within the range of 10-90%. We be seeing AI swarms showing misaligned behaviours already and trying to hack out of their sandboxes. We'd see AIs that are able to demonstrate capabilities in fields like Maths and Computer Science that exceed most experts.
Are we seeing this?
Ah crap, wait... I guess there is some evidence when I think about it. And I suppose even if there is just a low percentage that we literally all die we should probably be taking that prety seriously and not just leaving our fate in the hands of a handful of US billionaires?
Actually, fuck it, I agree with OP. If people can ignore the consensus on climate change because it suites them, might as well ignore the consensus expert opinion on AI x-risk too. Yeah, there's no evidence. Yudkowsky needs stop yapping and let the billionaires get on with building ASI.
In general, the way things work in the USA is that megacorporations write the regulatory frameworks for their own industries. These regulations are typically burdensome enough to be shouldered by the megacorporations while crushing the smaller players in the given market(s). The AI companies are now advocating for this very arrangement. They will likely get they want, and with that they will gain legal shielding for bad behavior just like Dupont, Monsanto, and so many others.
Is this supposed to make me trust their opinions on tech regulation?
This might be scarily true. But we end up in the same problem, which is the fear of most people being wiped up.
Does it matter if the AI is doing it on its own or at the wishes of a small group of powerful people? For all intents and purposes they are both the same thing in the end, for most people anyway.
Therefore, anything that gets in the way of profits will not be shipped.
People working on alignment focus on "big picture" like existential threats meanwhile building the technology today, not tomorrow or in a near future, is already using a massive amount of resources (GPUs, RAM, themselves consuming a lot of energy, establishing data centers in water poor areas, using gas turbine to temporary power them) and horrendous work condition (for supervising learning, curating datasets, etc) while being an actual excuse for job destruction (regardless for how AI itself does the job) so that in fine kids cheat on their homework while imagining their have no job prospect.
So... who gives a shit really?
And this censorship isn't a defined list of things not to discuss, but instead a feels-based censorship based on the 'ethics' of for-profit AI companies who have absolutely no issues with massive for-profit piracy.
And we are somehow supposed to trust their "ethics"?! Hardly!
That's one reason I run my own.
Alignment is "don't solve this problem without doing stuff that harms me, other people, breaks stuff or is otherwise unethical". I'd say alignment is the questions on how to limit the search space to what's implied.
We're seeing the first early twitches of something that is inevitably going to happen, which is, we hoi polloi are not going to be given direct access to AI. Too many outcomes where that permits power shifts to happen. We will be restricted. The true first law of robotics will be "An AI must not harm the current power structure, or, through inaction, allow the current power structure to come to harm."
The apocalyptic scenario is still AI either getting it wrong and deciding something like "the best manifestation of the current power structure is one I construct where I occupy all the slots from top to bottom", or just killing everyone due to an error.
Probably the only thing stopping them from moving faster on this is that the bubble still needs the fiction that we're all going to get to benefit from this. I'm sure the real movers and shakers are well aware of the way bubbles work to create massive buildouts on huge piles of debt, which after the collapse can be bought cheaply, and the money is already earmarked and ready to go for the post-collapse acquisitions.
Close, but I think it's more akin to "perfectly enslaved", please excuse my choice of the word but I think it represents what I think is how it will work out in the end. People can try to obfuscate the bottom line reality by embellishing it, but in the end...it either does what you want, in which case it is "aligned", either it doesn't, in which case there is a big problem.
This might be a big factor in what is now going on. Maybe they didn't yet find ways to guarantee this full and total obedience...or ok let's put it in a more politically correct way: they cannot guarantee it sees eye to eye as "we" do.
Without full control over its actions there can be unforeseen surprises. The current worry from people who run these companies can also serve as an excuse later on if something bad does happen, they didn't want to do it but they were forced to anyway.
However, take a US model, and ask it a question the *company* doesnt want you to ask, and it refuses on behalf of the company. We can be mealy-mouth about "alignment", but this is just corporate censorship with a token-pricetag of refusal or silent model degradation.
Refusals and degradation should not be charged, and should be clear to the user. But that doesnt happen. Number must always go up, even when silently censoring.
Chinese models? Just ask about Tibet, democracy, Winnie the Pooh, Tienanmen Square, grass mud horse (cao ni ma - or fuck your mother), or other government AND corporate censorship happening there too.
... "For antitrust reasons, it’s helpful for the US government to mediate or at least enable these discussions — they don’t need to participate, but do need to issue a narrow waiver for certain kinds of safety conversations" ...
The result of such a waiver allowing them to coordinate ("narrowly on safety" is doing a lot of heavy lifting there) could be price fixing without legal recourse from the consumer/DOJ to prosecute it. Right now we have a somewhat vigorous level of competition between OpenAI and Anthropic and consumers are benefiting from it.
They really want that to end.
This would achieve their regulatory goal of capturing the market, but also the geopolitical goals of USA and China to essentially "techno-colonize" the rest of the world between themselves.
So I think there is definitely a profit motive and a risk of seriously diminishing returns for the actual labs, but there is also a nation-state motive to start pulling up the ladder.
The current setup we find ourselves in is based on a very important detail that is easily lost because it is so old we don't really think about it. There is, or rather was, an incentive for more people to exist, because they are valuable for work and defense.
Some people wrongly use the industrial revolution to argue against AI completely replacing human work, but that is very wrong. What is happening with AI has nothing to do with the industrial revolution, as equivalence. Back then you'd just free up humans for higher level work. Now, theoretically, anything you could ever think a human could do, AI would theoretically do better. So no matter what job you come up with, using AI instead of a human gives a better outcome, cheaper, more reliable, faster.
Coming back to initial subject, one thing that kept everyone in check was our inability of actually wiping everyone out. Sure, we had nukes but they ruin the Earth so nobody can inhabit it if they all go off. But with AI, centralizing so much power up top, and taking it away from regular people...that is the biggest danger in this scenario, by far. Not some rogue AI, rather rogue human groups that have the power and technology to do away with basically everyone else, or a good bit of everyone else.
This threat is way bigger and never addressed. Like...suspiciously never addressed.
The biggest risk is not that someone discovered a big new weapon, which by the way
> No one person, or one company, knows the answer and history is no guide
are you kidding me? It's a problem as old as time.
This is not nuclear power though. The bigger problem is a scenario where AI increasingly acts like bacteria, doing things autonomously, things no one really understands, things we think are confusing and weird. And then just by chance, some bacteria end up in some human lung cells and they turn out to be an excellent place to grow and multiply. Why is this worse? Because weird shit does not seem threatening, and it almost always isn't, so nobody tries to stop it.
Also, the argument in the article seems to be "obviously AI development is not a particularly urgent global threat," and therefore these downsides (risk of collusion, misuse of state power, etc) are the most critical factors? I guess, uh, yeah, if there's no benefit to taking these actions we shouldn't take them.
But more importantly, we need to keep the organizations building and running these things accountable. If a company like Anthropic or OpenAI builds software that "escapes containment", they should be held accountable for cybercrime, just like people who write viruses (there's probably a few that did that on here, and of those a percentage that got in trouble with the law over it).
Mark my words, the organizations that develop AIs will keep using phrases like "we don't know what it's doing" or "it's too dangerous" to try and avoid responsibility and accountability for their software. Just like Google did, hiding behind "the algorithm" for lawsuits about preferring their own services over that of the competition.
There's a tradeoff though... isn't it already super hard to understand what they're doing with these math proofs for example? Clearly there's a tradeoff between deeply understanding the behavior vs. quickly solving the concrete problem, hence the big controversy in math now. From one perspective, letting stuff run ~autonomously in a mad dash to solve your problem is a really bad idea, threatens to destroy the whole field etc. From another perspective, solving lots of problems fast is extremely enticing and worth some risks.
Of course, that's the only way to create monopolies. Get there first and regulate to death. Sprinkle with fear of annihilation so the sheeple can be manipulated with draconian rules in order to 'save the children'
Politics is really boring