Hacker News Reader

Comment by jeroenhd | Hacker News Reader

> 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.

Isn't that what any normal company does? Comply until it's impossible to operate legally? They can choose between ceasing to operate or getting arrested.

If the AI regulations are reasonable and AI companies can't operate, the technology just isn't ready for wide-spread adoption yet. See also: Tesla's Autopilot.

I have to wonder what the investors over at Microsoft are thinking right now. I doubt they'll be happy with threats of leaving lucrative markets from a company so deeply integrated into their product suite.

View on HN

> If the AI regulations are reasonable and AI companies can't operate, the technology just isn't ready for wide-spread adoption yet.

Conversely if the AI system is reasonable, but the regulations are making them illegal, then the blame is on the regulations.

There is a trade-off between potential benefits and potential harm, and it's not immediately clear to me, who is in the better position to find it: the AI companies or the legislators. The first are biased because they want to sell their products, but the second have worse understanding of the technology and are IMO on average too over-protective.

So it's subjective right? The balance.

Since a company is usually harmed by increased regulation their opinion is biased.

But there's no oppposite to this. Governments don't benefit if companies leave and in fact they lose out. This also includes the population (jobs etc) that means the opinion of the government isn't biased.

> Since a company is usually harmed by increased regulation their opinion is biased.

I know this is the prevailing point of view in business, but is it actually true? How many bank failures would have been avoided if US banking regulations hadn't been slackened so much?

When the government covers your loses and even let's you take bonuses for fucking up and causing a recission why wouldn't you think regulations slows you down.

how many guns would get sold if the government didn't created laws that protects the industry from liability.

or section 230 which shields websites from the rent seeking of the cable companies but allows them to rent seek advertisers.

Common sense is what protects gun manufacturers from liability. If you run someone over with your car, Ford is not liable either.

> How many bank failures would have been avoided if US banking regulations hadn't been slackened so much?

Most of the failures after 2000 would have been prevented. All of the 2008 financial crisis was caused by removing regulation.

Prior to Commodity Futures Modernization Act of 2000 [0], several states ruled that credit default swaps were insurance and thus regulated CDS as insurance. Some states ruled that CDS were gambling and regulated it as gambling. The CFMA basically said "credit default swaps are now a federal issue and it will be regulated at the federal level" and set up a poorly funded agency that had no teeth until after the bailouts in 2008.

During the 1920s, many banks got involved in both commercial banking (taking deposits and making loans) as well as investment banking (arranging IPOs). In 1933, the Glass-Steagal Act [1] said that banks may do one or the other, but cannot do both. In 1999, Glass-Steagal was later repealed by GLBA [2] as CitiCorp was doing both (due to mergers) while daring the regulators to do anything about it.

0 - https://en.wikipedia.org/wiki/Commodity_Futures_Modernizatio...

1 - https://en.wikipedia.org/wiki/1933_Banking_Act

2 - https://en.wikipedia.org/wiki/Gramm%E2%80%93Leach%E2%80%93Bl...

Regulatory moats can be pretty powerful - not sure regulations always harm all companies.

The regulation is meant to protect the EU citizens which couldn't care less about the financial incentives of a US company or their hype du jour. If hypothetically no AI can be created with respecting the citizens, what should the choice be, block or accept Skynet? (I said "hypothetically")

As an EU citizen I think it's about joining SkyNet or falling behind in productivity. I prefer joining SkyNet.

If, hypothetically, a SkyNet like AI actually existed, how much damage to humanity would you be ok with as long as productivity and quality of life went up went up for those who were spared? Let's assume that you and your loved ones are among those who have been guaranteed safety by this AI.

I am sorry if the tone comes across as snarky. Not my intention. Genuinely curious.

My comment was about if it's starting to exist, EU regulatation won't stop its power and influence.

If everyone uses ChatSkyNet for programming (or even ChatGPT-4), but EU is not allowed, EU will simply lose its competitiveness in creating products.

> My comment was about if it's starting to exist, EU regulatation won't stop its power and influence.

it... might?

it possibly existing in the future is not a good reason to lay down and just surrender control of world governance to ChatGPT today

Sadly, the EU has already lost that competitiveness through many other similar regulatory situations. The real risk is not competitiveness in creating products, that's a lost battle already :( The real risk is losing competitiveness in even being able to work for the American companies that are creating products.

> If hypothetically no AI can be created with respecting the citizens

The problem is you're assuming the people writing the regulations are perfect. It's not that bad AI should exist, it's that we're probably at least a decade away from regulators even having a hope of starting to understand what's going on.

Why should I assume anything? We are what we are and we do what we can with what we have - imperfect as it is. Inactivity is way worse. By the way understanding all things (AI or whatever) is definitely not a requirement for the legislators. I'm sure you remember they have hordes of specialists in their staff and those will understand the aspects at hand to sketch recommendations.

It's not legislators, although I disagree with your assessment there. The people actually doing assessments and inspections of companies do not have much expertise at all. Even for far simpler technical tasks than assessing AI.

Suppose a new generation of AIs results in a big economy boost. Then a country might be put in a position to choose between some risks to privacy etc and a risk of being left behind.

A similar situation would be a country that in 1900 decides whether or not to outlaw cars.

Nobody outlawed cars.

There were preciously little regulations in the 1900s though. Don't you think if cars were invented today they would face a whole lot of regulations?

Cars do face a ton of regulations today. You can’t sell a car if it doesn’t meet or exceed hundreds of regulations.

1900s also had child labor and black lung. Not a standard to aspire to.

In the UK we did mandate that a pedestrian must walk in front of the car with a red flag to warn of the approaching vehicle though for a while!

https://en.wikipedia.org/wiki/Red_flag_traffic_laws#Red_flag...

I am curious, do you think SkyNet, if it's really "SkyNet", can be blocked within the confines of the country in which it appears first?

And if it's not SkyNet but just an awesome tool like ChatGPT is today, do you think regulators will manage to take into consideration the advantages lost making this tool illegal while protecting us from a potential SkyNet?

> Conversely if the AI system is reasonable, but the regulations are making them illegal, then the blame is on the regulations.

have you read the EU's proposed approach to regulating AI?

https://digital-strategy.ec.europa.eu/en/policies/regulatory...

it seems to be an extremely reasonable approach to me, if anything a little too lax

the problem for OpenAI is its products can't easily be segregated into the different risk buckets...

and they want ChatGPT wired into the high risk areas as that's where the profit is

I think the proposal is reasonable provided ChatGPT is classified as moderate risk.

One thing I don't quite understand is how they are going to ensure that AI-generated text is always marked as such. When the text is generated in the chatbot, the fact that it is AI-generated is clear, but the user is free to copy-paste it anywhere. There is no good way to watermark plaintext.

Individuals will always be able to do this, but it might be enough if it's illegal and a bit of extra effort. You don't need to eliminate 100% of AI-generated spam, just make it risky and unprofitable enough that it's not a common practice, and have people call out those who seem to be doing it. I'm not sure it's possible, but it seems worth trying.

And if someone is operating a troll farm in another country, this won't help.

By proclaiming they'd leave if they can't comply, it might cause the EU regulators to "self-censor" the regulation such that it's "easy to comply".

but tbh, i dont see what the regulations would be. I am a fan of no regulation unless actual harm has been shown first.

> I am a fan of no regulation unless actual harm has been shown first.

I used to be, but some of the modern crop of CEOs has taken the wind out of that. They have done tremendous damage, very quickly, and the regulation has been unable to keep up, until severe damage has already been done.

Once the bullet has left the gun, it can't be stuffed back in.

Can you be more specific, with examples? Who did what?

You need an example of a bad thing happening because regulations didn't exist?

Uber and Airbnb?

Uber brought ride sharing and cracked the taxi market wide open - perhaps to the benefit of riders.

Airbnb is also a boon for travellers looking to stay cheap, and a boon for property owners to cash in on underutilized real estate.

The "harm" that has been cited to have been caused by them are only the incumbents that do not move with the market or times.

It's like a necker cube.

Glance at it and she looks like a renter or typical youth of a city with tourism but poor purchasing power/ social mobility, but stare a bit longer and relax your focus and she becomes an incumbent who doesn't move with the market or times

You don't live in a European city near the sea. Airbnb is a plague. We used to use a house sharing in summer, we did it for years, until Airbnb killed it. It was an exchange 1 to 1 in term of population, we now do x5 to x10 in summer.

Now that the harm is done (speculation on housing priced out teachers and low income workers, old, niche stores turned into tourist traps), we finally have regulation. A bit too late. But at least the mayor is adding 20k housing units (+20%), driving the price down and making speculators eat a fat loss.

Uber was regulated much faster so it's impact have mostly been positive, unless you've been raped by a predator as they used to not check their drivers.

You have to regulate ahead of time, even if it's soft and toothless in the beginning. Check the EU proposal to regulate AI, it seems really, really soft.

Asbestos. Phthalates. Fen-phen. Chloroflorocarbons. Leaded gasoline. All were legal because no harm had been shown... yet.

"Modern crop of CEOs"

All of those examples are primarily damage done pre 80s

Phthalates? Aren't those currently an issue, right now, in water? Aren't they currently being used in plastic? Or did I get the name wrong?

The name is right but the claims of clear badness are suspect.

https://unherd.com/2021/03/the-truth-about-spermageddon/

Was the potential harm well known and accepted before showing it existed? How?!

User privacy, cambridge analytica, facebook, and the impact those together had in several elections, most especially Trump 2016 and Brexit.

Cambridge Analytica had basically no effect on any election. People who actually understand online advertising consider it a joke.

The "organic" social stuff that helped Trump 2016 (leaked Democrat emails, etc) had nothing to do with privacy or ads.

Cambridge Analytica's sold their services literally as "election management agency" and promised to be able to influence a large number of undecided voters.

More than 87M people were profiled, and more than 3M people were targeted by Cambridge Analytica in 11 key states during the election to feed them the content that would swing their indecision towards Trump based on their psychological profile and preferences.

Trump paid for that, and Michigan and Pennsylvania and New Hampshire were all very close.

To boldly state that CA had no impact on US elections, Brexit and I don't remember which South American election when the people involved in it literally sold it and used previous results as evidence seems a bit of a stretch.

I might be wrong, but there's no clear way of knowing unless votes were not anonymous.

Imagine that someone says Trump hacked the voting computers by writing code on a piece of paper and slotting it into the machines.

The CA thing is the same level of stupid. The targeting would just be senselessly bad based on CA's methodology.

What was the actual, documented, damage on the "user privacy" issue though? The only effect I know is targeted ads which seems like a net benefit to me.

I can be, but, really, why should I?

It would take about twenty seconds with Google, to come up with hundreds of examples. Some, better than others.

They may seem obvious to you, but to a lot of us, it's not obvious at all. Vague notions of "damage" could refer to anything, like harming competition. Different value systems and all that.

Well, the original post was "Stop me if you can."

Basically, if there's no rule saying I can't do it, then I should do it, until rules are developed to stop it; regardless of the morality or danger.

I suspect that one way to make people more careful, is to make future rules and regulations retroactive.

For example, if I had developed a fintech product that was a legal ponzi scheme, or a downloader that is a legal way of accessing copyrighted material, because of loopholes, and those loopholes are closed, after the fact (happens all the time), then, if the new remedies can be applied to the "legal" implementations that were in effect before the closing of the loopholes, I suspect that it would make the "loopholer" a wee bit more cautious.

But that won't happen. There's way too many ways that it could go off the rails.

So people will keep on bending, folding, spindling, and mutilating, until they are stopped by a steel door, slamming down in front of them.

I strongly suspect that a lot of companies are set up to harvest as much as possible, before the door slams, so they can reap the benefits.

> if the new remedies can be applied to the "legal" implementations that were in effect before the closing of the loopholes, I suspect that it would make the "loopholer" a wee bit more cautious.

and there's moral implications about applying new rules retroactively. What you do today that is completely legal, should not be punished tomorrow even if it was decided in the future that it is illegal. Otherwise, there would be a chilling effect on innovation.

Exactly. I could see a government, in need of tax revenue, enacting a law that taxes some activity, then going after people that have been doing it for years.

Suddenly, we owe ten years of back taxes, with penalties.

It’s not feasible to do this.

But there are plenty if people –many on this very forum, I suspect– that know what they are doing is highly immoral, and will likely get shut down, but … MONEY … so they do it for as long as they can gt away with it.

> the regulation has been unable to keep up

it's because the gov't is not incentivized properly. I think they should be pressured to keep up. This isn't an argument to "pre-regulate", because incompetence at the gov't can be fixed!

> Once the bullet has left the gun, it can't be stuffed back in.

But once many people start shooting guns willy-nilly, you start regulating guns. But not before.

This is a normal negotiation tactic on the part of OpenAI. Businesses often say they will leave a jurisdiction or be forced to shut down if X regulation is put in place.

It's a tactic of all businesses. In the US they play states against each other to get higher and higher tax breaks for moving.

It's not a tactic. It's just reality. If all the expense of coping with regulation is too much, then it's not worth doing. It seems almost too facile for Sam Altmann to have said, but perhaps it's worth reminding people that regulatory compliance is an expensive business in and of itself, even if your product already fully complies.

The name for it is Boulwarism [0]. It is a "take it or leave it" attitude. You see it every day when presented with an "I agree" checkbox/button.

0 - https://en.wikipedia.org/wiki/Boulwarism

I don't see the difference from normal pricing, it's take it or leave it in nearly all retail contexts nowadays.

Businesses are not animals, it's immoral

This is a nonsensical statement. All markets, and thus all goods sold on a market are regulated. You can’t have markets without regulations.

This is just thinking about the wrong level of regulation. Not all goods need to have CE marking, or get audited regularly, or anything like that. I can go to someone who makes furniture in their shop and buy that furniture without them ever contacting the British Standards Institute.

You can't buy it if you can't agree about what money is. That's one of those pesky regulations.

As I say, the wrong level of regulation.

The regulations forbid competing currencies (sorta), but people create currencies for trade even when no government is doing so.

You're not making a point on the argument being made.

It's not about _not_ having regulations. It's about not introducing new regulations without demonstrating harm first.

The current crop of regulations, i argue, is sufficient, and no new ones introduced for AI.

    … no regulation unless actual harm has been shown first.

What?

It is the duty of politics to protect the citizens and environment - also proactively. From reckless CEOs, greedy shareholders and lawyers which claim that it wasn’t strictly forbidden to cause harm.

Politics isn’t doing always right. The Cookie-Directive tried to make misuse of Cookies illegal. Actually it made web browsing horrible and created further questionable businesses like “cookie cloud storage providers aggregating cookies of several websites”.

On the other hand this people require ABS and ESP in cars. Banned Teslas “Autopilot” and had smart ideas like type certification for aircraft’s. And then they forbid stuff which can cause cancer…that was smart! What comes next? Acting against monopolies and oligopolies like Microsoft and Apple?

Regulation without known harm is a case of we need to do something and this is something. The rules will be arbitrary and expensive to comply with and will most likely have no preventative effect on any negative effects.

vaccines are like the medicine analogy to regulation.

You have to first prove that it is not going to do more harm than good. And without knowing what sort of harm AI's do - it's all theoretical at the moment - any regulation is merely speculating a harm, and the cost of prevention might be high (or stifling).

ABS and ESP in cars came _after_ harm was shown. And so is emission controls, and crash-related safety measures and testing/certification.

> And then they forbid stuff which can cause cancer…that was smart!

which makes sense, as those stuff has been shown to cause cancer.

In other words, harm has to be shown, rather than speculated, before regulation is desirable.

We know certain possible high risk applications already (e.g., dosage recommendation for medication), so it is not all about figuring if something is risky or can cause harm.

The question then is more: regulate those within the particular area of use or have something that generally aims to regulate risky use.

How are you going to *proove* in advance that a certain economic or social measure is doing what you expect?

(Harm has already been shown. Its more nuanced than this.)

Also proactively, indeed, but usually rather reactive. Why bother with say cops on the street to check for weapons when civilians tend to not carry them there? Proactive is rather: seemingly proactive, like AI is seemingly intelligent. In reality 101, its reactive, but the quality of the reactivity matters (speed, thoroughness, accuracy, transparency, cost, etc). Some of these qualitative measures are at odds with each other.

> And then they forbid stuff which can cause cancer…that was smart!

Tobacco is regulated; not forbidden.

Why wait until harm occurs to regulate?

Why pay a cost to regulate, if it may not ever be needed?

Assuming there's reasonable probability that something could go wrong you don't wait. It's better to be proactive instead of reactive

As quoted in the article:

> He called for some regulation “between the traditional European approach and the traditional U.S. approach,” whatever that means.

And you phrase it:

> I am a fan of no regulation unless actual harm has been shown first.

This is the "traditional US approach". As has been shown far too often, by the time actual harm has been shown, it is far too late. Especially as large industries like to use their economic clout to pervert things with propaganda like "oh, the science isn't settled yet".

The "traditional US approach" is why cigarettes were sold for 100 years after they had been shown to cause cancer along with other major medical issues. And why cigarettes are still sold.

And another HN story from today discusses this exact issue: https://news.ycombinator.com/item?id=36067401

The European approach is "show us that it is safe before you put this stuff on the market". They use the 'Precautionary Principle'.

> The precautionary principle is the idea that when there is the chance of negative consequences from an industrial practice, that the burden of proof lies with the inventors/implementers of the process to prove that there are no negative (ecological and environmental) consequences of it. The principle is often cited by "technological conservatives" and/or environmental activists when there is a perceived lack of evidence showing that a technology is absolutely safe.

https://rationalwiki.org/wiki/Precautionary_principle

https://en.wikipedia.org/wiki/Precautionary_principle

Is AI "safe enough" to be sold/marketed to people? Many people say it isn't. The people who tend to get into executive positions at companies have a visible lack of concern about anything other than their personal wealth. So while AI might be "safe enough", the people who tend to get to the top will make even the safest things unsafe.

And yet, the empirical results is that the US is now wealthier than the EU (at least, in totality).

And not to mention that such regulations (on AI) might not be adhered to by countries such as China, which would make it all but moot - except it causes harm to western companies looking to profit.

A cautious approach is valid only if you have unlimited budget and you are not in a natural darwinian selection process of elimination.

The title of the article is undoubtedly a lie.

"any real regulation" does not match in any way what was said.

Yeah, sounds like a reasonable take (the title makes it a bit more... inflammatory?).

Of course all business should comply until they can't, at which point it's really not their problem anymore. I think it's perfectly fine to say "hey, we did our best", but if it still isn't enough, to simply not play the game.

This is an article by Gizmodo meant to generate clicks.

Unfortunately, HN has fallen for the tricks.

> If the AI regulations are reasonable and AI companies can't operate, the technology just isn't ready for wide-spread adoption yet. See also: Tesla's Autopilot.

It couldn’t be that the EU regulators are clueless could it?

And you don’t see a difference between the harm that ChatGPT could do compared to the harm of an AI controlling a self driving car?

Reply on news.ycombinator.com