AI regulations are crony capitalism
reason.com
reason.com
OpenAI has a head start, but the technology is very well understood and the only limitation to building new models is budget. Facebook is actively working to undercut them and commoditize the technology and it looks like they will succeed. OpenAI can't keep spending billions to stay ahead of the commoditization, so they need the government to pull up the drawbridge and give them that moat.
Altman himself doesn't care one iota about safety and he just pushed out the people that did. This is just a strategic play to him.
EU AI Act has the following provisions: document foundational models, disclose copyrighted sources, never use AI in critical/dangerous applications etc. Altman threw a fit and threatened to pull out of EU
News to me.
Or did you mean the quote from last year that was:
> If we can comply, we will, and if we can’t, we’ll cease operating… We will try. But there are technical limits to what’s possible.
> The law, Altman said, was “not inherently flawed,” but he went on to say that “the subtle details here really matter.”
https://time.com/6282325/sam-altman-openai-eu/
Because I find it extremely irritating that for the last year so many people have taken the worst possible interpretation of him saying "we won't break the law".
I wish this whole topic wasn't a massive popcorn drama fest; but I suppose that was always wishful thinking for a tech that, at it's friendliest, is going to be a 10x speed-run of the industrial revolution.
That was quite literally, translated from weaseltongue, "we will pull out of the EU because we don't like the law"
There's nothing in the law they can't comply with.
> so many people have taken the worst possible interpretation of him saying "we won't break the law".
People are smarter than you give them credit for. Altman is good at grandstanding about safety, regulation, oversight etc. What he lacks is delivering on that grandstanding.
"Oh the law demands the exact things I'm grandstanding about? Oops, we might pull out because somehow we can break the law".
Same goes for literally everything else. Open? In name only. Safety? Trust us our best people are on it, can't verify or audit their work though. Government regulation? Oh, we might not comply. And so on.
No, that's you that is.
You're choosing to see the worst possible version, and putting your words in his mouth, and then telling everyone else to ignore what he actually said and listen to you instead.
> "Oh the law demands the exact things I'm grandstanding about? Oops, we might pull out because somehow we can break the law".
"The law, Altman said, was “not inherently flawed,” but he went on to say that “the subtle details here really matter.”"
But, apparently, you don't want to accept that he might mean what he says and say what he means. The conspiracy version where he's an evil villain whose mask you can pull off in order to show your cleverness is so much more fun and exciting, perhaps?
Or I perhaps I putting words into your mouth in order to demonstrate why this is annoying? :D
(Or perhaps I'm still sore about the time someone did that exact thing to me, on this site, repeatedly, with regards to Palestine).
> Open? In name only.
Published research, published weights for some of their models, and…
> Safety? Trust us our best people are on it, can't verify or audit their work though.
… and published a 100 page document about that work explaining with examples why it would be a bad idea to publish the weights of the better models.
(But those who know what's up about the superalignment thing… are all (paraphrased) like "it's days like this I wish I wasn't under an NDA", so that will be interesting to watch when it actually gets published and is not just wishful thinking on the basis of people making the worst-possible takes like a horror film keeping the monster off-screen).
> Government regulation? Oh, we might not comply.
They literally said the exact opposite of that. In my quote. Which I gave you. Which you chose to interpret in the worst possible light, despite (or perhaps, given human psychology, because) I indicated how annoying this is.
You know what? For all I know every evil that's been attributed to the guy is 100% true. But none of the public statements you're relying on actually demonstrate that. Perhaps someone will break an NDA and we'll see something deserving of all this drama, but the evidence isn't there yet.
So far, everything is extremely straightforward and simple, it doesn't need 5D chess or mind-games or whatever, just a bunch of nerds being nerds and speaking like nerds with nerdy caveats and nerdy caution and nerdy excitement — you know, like this site.
It's not a conspiracy theory. It's the world we live in.
> Published research, published weights for some of their models, and…
And what? Should I remind you of their original stated goal vs. "some of the weights" and "we argue we can't release weights for more powerful models"?
> published a 100 page document about that work explaining with examples why it would be a bad idea to publish the weights of the better models.
It's bad, so we must blindly trust "Open"AI that the work they do is correct, and proper, and...
> Perhaps someone will break an NDA and we'll see something deserving of all this drama, but the evidence isn't there yet
The proof, as I said, is in the pudding. That is, in the doing. There's a lot of grandstanding on a lot of issues, and very little to show for it. Oh, and a lot of weaseling out.
It's the world you live in.
> And what
Follow the ellipsis.
> Should I remind you of their original stated goal vs. "some of the weights" and "we argue we can't release weights for more powerful models"?
"""This decision, as well as our discussion of it, is an experiment: while we are not sure that it is the right decision today, we believe that the AI community will eventually need to tackle the issue of publication norms in a thoughtful way in certain research areas. Other disciplines such as biotechnology and cybersecurity have long had active debates about responsible publication in cases with clear misuse potential, and we hope that our experiment will serve as a case study for more nuanced discussions of model and code release decisions in the AI community.
We are aware that some researchers have the technical capacity to reproduce and open source our results. We believe our release strategy limits the initial set of organizations who may choose to do this, and gives the AI community more time to have a discussion about the implications of such systems."""
- https://openai.com/index/better-language-models/
So again, you're putting words into their mouths, not taking them at face value.
> It's bad, so we must blindly trust "Open"AI that the work they do is correct, and proper, and...
And apparently either fail to spot how I was using an ellipsis, or ironically make the mistake you thought I was making.
But also, in your case, not bother to read the research.
> The proof, as I said, is in the pudding. That is, in the doing. There's a lot of grandstanding on a lot of issues, and very little to show for it. Oh, and a lot of weaseling out.
You sure you're not looking in the mirror when you see grandstanding?
Where have you lived for the past... well, forever?
> while we are not sure that it is the right decision today, we believe that the AI community will eventually need
It's a lot of grandstanding about nothing, amounting to the same thing: "our only modus operandi is being as closed as possible for as long as possible"
> And apparently either fail to spot how I was using an ellipsis,
I can also put ellipses everywhere. You know that their "open research" is rarely reproducible? That they keep deprecating and removing the completely closed proprietary models they "open" to the public? That...
Follow the ellipsis
> You sure you're not looking in the mirror when you see grandstanding?
---
grandstanding /ˈɡran(d)standɪŋ/
noun
the action of behaving in a showy or ostentatious manner in an attempt to attract favourable attention from spectators or the media.
---
No, definitely not I. There's a certain "Open" company in the AI space that does exactly that.
AIs which just answer questions people pose seem to be mostly harmless. Some people might not like the answers, and some of the answers might be wrong. That's true of web search results now. That's just an extension of the social media controversy.
AIs which have authority over people are scary. "AI says No!" This is a corporate power problem, not an AI problem. If corporations or landlords have arbitrary discretionary power to affect people's lives, that's the problem. Not that they delegated that authority to an AI. The EU regulates decisions against individuals made by algorithms. The EU has already stopped Uber from "robo-firing", where the app back end fired underperforming drivers.
Few proposed AI laws make this distinction. That's because it would raise the issue of arbitrary exercise of power and collective individual rights. That might lead to political unrest, or even unions.
Corporations, meanwhile, are not supposed to wield disproportionate power over the laws and regulations. Separate issues.
Works well for the US financial sector despite claims to the contrary
At least the [elected] government is under the responsibility of its citizens, while corporations are only responsible to their shareholders.
That is precisely what happens in practice though. Referencing the thousands of people killed by police in the US every year, with little to no accountability (who is to hold them accountable?)
Our government in general and our law enforcement specifically are very much out of control in many ways, but that is not an inherent state of affairs.
The whole point of this article is about crony capitalism.
Nobody blames Google when the answers are wrong, because search doesn't give you answers. Search only returns results, and it's your job to evaluate those results and decide which is most likely. AI skips this and tries to give you the answer directly, which makes it responsible for its accuracy. But there is no signal for whether an answer is right or not, especially because it usually doesn't give you a source for its response. You have to roll the dice in a way you never have to for search.
This is a fundamental issue with AI, which will always exist until it stops pretending to know everything and instead work by finding the answer and being transparent about its "thought process".
Then those panels fall victim to exactly the same issue that AI is having today. When you hide your sources, you are assuming responsibility for accuracy. You better be right, because the user has no way to judge the quality of your answer.
Don't they?
https://en.wikipedia.org/wiki/Criticism_of_Google#Possible_m...
Right, that would be Stack Overflow and all the sites which use SEO to rank for "best".
Trouble is, everyone puts a different thing after that dash.
It's not that arbitrary exercise of corporate power isn't a thing, it totally is, but AI is every problem that humans can come up with, because AI is all about automating things and solving problems, and that includes as a subset all the bad stuff we do to each other both on purpose and also when we just fail to care about the side-effects of our actions.
Similar to how cannabis rolled out in USA. Now many places require single-use RFID from the sole provider. Get in and then form the laws to build your business.
For the rest of us, the relevant question is: are the costs worth the benefits?
> The risk [...] is that radically new products and approaches in the arena never get a chance to be developed and benefit consumers.
Totally. And the opposite risk is that these companies deploy technologies that cause massive harm to people, without adequate testing, because they're caught up in a race.
I think some regulations would be helpful on balance -- like reporting of large training runs, as in SB 1047 -- and some wouldn't -- like (hypothetically) requiring a license to train small models.
This is where capitalism breaks down, when it comes to technology that is powerful enough to erode liberty and undermine civilizations.
Or we could just make the oligopoly explicit like whatever it is you are proposing.
Second point: the market-based legal system will not devolve into a system of ineptitude only if the money used in the market system is sound and free from political control, which we currently do not have in the US.
Move away from shareholder value as the be-all-end-all of a company's existence.
Reducing the privacy of the big players to almost nothing on pain of financial repercussions that are actually worth a damn would go a long way. Boardroom talks about "exactly how shitty can we make UI to milk maximum revenue out of our users before they just drop X product" should not be a private discussion. After all, as the NSA says, if you have nothing to hide you have nothing to fear, right?
Put a greater emphasis on different parts of education. How computer technologies work, psychology and how marketing, advertising, and social media abuse it, economics and all the stupid Wall Street financial creations that lead to Great Depressions, politics and how people play those games. Increase the level of awareness and jadedness among the general population so bad actors have a harder time getting the benefit of the doubt.
Do away with forced arbitration entirely. Enforce actual standards of customer service. Prosecute dark patterns and heavy-handed SaaS terms and EULAs. Maximize right-to-repair and couple it with an extreme reduction in the protection time for IP on things like schematics and software. Treat negligence, noncompliance, and bad-faith compliance as equal crimes. Companies taken to court can contribute to the public good by generously covering the costs of both parties unless they can prove beyond a reasonable doubt that a lawsuit is frivolous. Financial repercussions should be proportionate to annual revenue and be actually punishing. Executives who sign off on decisions that are detrimental to the general public should actually face prison, as should shareholders that pressure those decisions.
The people who say this garbage would never let any other ideology weasel its way out of the effects of their system in practice, but for some bizarre reason, the giga brains at Reason are allowed to get a pass and say that this is not true capitalism.
The type of capture by regulation outlined in the article is the direct result of a political landscape that prioritizes corporate power over government power. There is nothing "crony" about it; it's just capitalism.
P.S. This is not pro or anti-capitalist; it's anti-shitty argument.
Right now, AI-induced unemployment just isn’t a problem.
So it’s extremely likely the advocates for regulation are AI corporations, rather than reams of unemployed workers.
It makes sense to say new buildings have to meet fire safety regulations that are based on learnings from past fires. AI is too new; everyone is either guessing or lobbying for personal benefit.
We should be vigilant, but I don’t for a second believe the “any regulation is better than nothing” rhetoric from those who stand to gain.
It's not. All the learnings we've had from ML (the mistakes, the failures etc.) are just as applicable to AI.
I'm not sure how you'd test this hypothesis, but it certainly looks like a tonne of the ongoing FAANG layoffs are based on assumptions about a decreased need for low-mid level programmers.
I therefore think it's a good idea to have a plan for AI induced unemployment before people take to the streets because they have no job, no hope of a job, and no money left to pay the bills.
I have no idea how far away economic mass unemployment is, and I'm not convinced the current indicators in specific fields (much as it affects those fields) is "it", but I have seen a lot of news stories about governments getting surprised by things everyone was warning them about and they ignored.
What, if, for example, we’ll have a mass information system that would start pushing something that is very beneficial for special interest communities (say community is furries and this increases their dating pool), but tends to result in higher level of depression and suicide in teenagers. And then we also find that overall the system reduces mortality, by providing high quality medical advice.
Surely shutting down the system would not be a good idea. And requiring private enterprise to change their messaging infringes on free speech. But at least requiring to properly disclose biases is likely a good idea.
If you mean it in a sense of "what about all the lost jobs?", then we should start seriously talking about UBI, and how to fund it by taxing automation.