9 ms·
E.U. Agrees on Artificial Intelligence Rules with Landmark New Law
- sva_ 3y ago> Use of facial recognition software by police and governments would be restricted outside of certain safety and national security exemptions. Companies that violated the regulations could face fines of up to 7 percent of global sales. So a 7% tax on developing/deploying such systems. Not a bad deal.
- martin8412 3y ago7% per infraction. If you don't show signs of trying to immediately rectify it, they can keep fining you the 7%.
- radicalcentrist 3y agoSounds like 7% of revenue rather than profit. But I agree it seems weird to cap it at 7.
- sva_ 3y agoMy thinking is that they can just start a subsidiary that draws from upper (parent company) IP to pay out profits. It sounds like an accountant homework task, but I don't know shit. I'm just a bit cynical, having lived in the EU my whole life.
- dharmab 3y agoThis sounds very similar to the GDPR fine structure, and that doesn't have that loophole.
- chmod775 3y ago> While trying to protect against its possible risks, like automating jobs, spreading misinformation online and endangering national security. Automating jobs isn't a risk, it's the main feature. I can't wait for a future in which human will have to do little actual work besides what they feel like doing.
- dmit10 3y agoIt wouldn't be a risk at all, if AI learned to help to create new businesses (being a founder's copilot) sooner than it learns to automate jobs.
- tyingq 3y agoI get the feeling that the plan to keep that from just being lots of people in poverty takes longer to evolve than the AI does to take the jobs. Edit: For the replies, yes, things adjust. It's not, however, instantaneous.
- advisedwang 3y agoHah, many of the VC folks behind AI companies are fighting policies that would keep people out of poverty.
- schrodinger 3y agoAlso—every new sufficiently advanced technology has been predicted to do this. If there's one thing humans are good at, it's finding the remaining jobs that computers _cannot_ do, and sometimes accelerating the now-obsolete jobs has a positive effect.
- imiric 3y agoSure, but this has never happened at the scale and speed that it will with AI. We're inventing technology that will be capable of doing every creative, intellectual and physical task better than any human. This will have a far greater impact than when the telephone, car, or even computer were invented. Certain jobs have already been made obsolete practically overnight, and the writing is on the wall for many more. Society needs time to adjust, for new jobs to be created and people to be trained in them. We'll figure it out, but the transition period will be rough for many.
- semiquaver 3y ago> Certain jobs have already been made obsolete practically overnight Which jobs are you referring to?
- superkuh 3y agoFull text mirror: https://web.archive.org/web/20231208225306/https://www.nytimes.com/2023/12/08/technology/eu-ai-act-regulation.html https://web.archive.org/web/20231208225306/https://www.nytim...
- Lariscus 3y agoHere is the actual press release. >https://www.europarl.europa.eu/news/en/press-room/20231206IPR15699/artificial-intelligence-act-deal-on-comprehensive-rules-for-trustworthy-ai https://www.europarl.europa.eu/news/en/press-room/20231206IP...
- didibus 3y agoIt actually feels pretty well thought out, and narrowly targets uses that I think most would condemn.
- tomjen3 3y agoExcept that it gives a very broad excemption for law enforcement using facial recognition.
- petre 3y agoThat they will buy from the US or China or Israel, since it's illegal to use in the EU and no domestic company will bother to build such tech for one customer only.
- hcfman 3y agoYes. Law enforcement should not get exemptions in the current environment because of the absence of effective safeguards. There needs to be a push to acquire effective safe guards across the board for law enforcement and government activities. That’s what they need to be making new laws for. There should be consequences attached to abuse.
- Gud 3y agoThere should not be an exemption, period.
- lacker 3y agoSome practices, such as the indiscriminate scraping of images from the internet to create a facial recognition database, would be banned outright. Sounds like the devil is in the details. What if you scrape a large quantity of images and the end result just happens to be able to recognize many faces? Hosted services like ChatGPT can "solve" this by refusing to identify faces, and if you hack around it with prompt engineering, well, they can tell the EU that they tried. Open source models that can handle images, though? Hopefully this regulation does not end up forbidding the use of general-purpose open source models.
- esperent 3y agoFirst, note that facial recognition is not the same thing as being able to recognize celebrities. It's about being able to identify you or me as we go about our daily lives. > Sounds like the devil is in the details. What if you scrape a large quantity of images and the end result just happens to be able to recognize many faces? I think you can solve this by focusing on the end result. If you create a tool that scrapes images, processes them in some way, and ends up being capable of facial recognition, then it should fall under the purview of this law. > Hosted services like ChatGPT can "solve" this by refusing to identify faces, and if you hack around it with prompt engineering, well, they can tell the EU that they tried And then the EU can reply "you didn't try hard enough, now take down the tool and pay these fines". However, is there any sign that ChatGPT can do general facial recognition? > Open source models that can handle images, though? Hopefully this regulation does not end up forbidding the use of general-purpose open source models This a more valid worry, to my mind.
- AndrewKemendo 3y agoIs it your opinion that since it’s possible to subvert it they should abandon the attempt?
- lacker 3y agoI would like general purpose open source AI models to be legal, even if it is possible to use them for harmful purposes. Just like open source cryptography, open source operating systems, et cetera. If a product's primary purpose is crime, fine, ban it. But AI models should be treated just like other technologies, from email to the web to the computer itself.
- schrodinger 3y agoAs much as I am worried about the unintended consequences of AI being released, (a) I suspect this stage might be a little oversold (as great as it is), and (b) we should give it a little more time to play out to make more informed decisions.
- gumballindie 3y ago> of AI being release, Released where? Out of the crazy house?
- schrodinger 3y agoSorry, I thought it'd be clear from my comment. Wait until AI has more widespread use within the industry, see what jobs it's taking, and _then_ start regulating.
- rkagerer 3y agoIt was clear to me
- gumballindie 3y agoThe only regulation we need is punishing those that steal intellectual property. All else is fair game.
- artninja1988 3y agoNo such thing as "stealing" "intellectual" "property"
- sakex 3y agoTo be sure we're not going to lose yet another technological race, we're just not going to participate.
- viraptor 3y agoThis is a lazy dismissal and the topic was specifically mentioned in the article. Did you have some thoughts about it beyond "no regulation, yolo, free market"?
- junofan 3y agoI skimmed it. I don’t think carve-outs and exemptions for government use are a great idea and seem more likely to keep the public ignorant. The structure of the AI board seems pretty messed up—it gives some people in national governments discretion over peer agencies and seems designed to divide-and-conquer. The whole thing feels shady and I’d be concerned if I lived there. More generally, I’m not from the EU but the whole concept of the government granting rights to citizens naturally implies the government can grant rights to itself. I’m just not a fan.
- DarkmSparks 3y agoIt seems no one told the EU that once AI works it is not called AI anymore.
- Animats 3y ago> AI systems that manipulate human behaviour to circumvent their free will; Like Facebook and TikTok?
- aussieguy1234 3y agoFor anyone with uBlock origin running into the paywall, import this filter list: https://gitlab.com/magnolia1234/bypass-paywalls-clean-filters/-/raw/main/bpc-paywall-filter.txt https://gitlab.com/magnolia1234/bypass-paywalls-clean-filter...
- wslh 3y agoThis is like preventing to develop weapons, at the end if you don't have them you will lose the war. It is suicide. Others will do.
- esperent 3y agoThere are plenty of regulating weapons usage and development. E.g. chemical and biological weapons are banned by the Geneva convention. I've never heard anyone reasonably say that European armies are likely to lose a war because they don't use biological weapons.
- wslh 3y agoThe problem is the weaponification of technology. Governments tried to weaponify cryptography and they failed. Here it is the EU who is limiting itself and it is not comparable to the Geneva convention since other countries will not adhere. While EU limits themselves others will take advantage of the technology. Returning to the Geneva convention, we don't know if in less than 100 years you could build a weapon of mass destruction at home like you print a 3D model. The problem is technology is also a Pandora Box that at some point you cannot control with laws. The Geneva convention has less than 100 years, let's see what happen to the humanity in another 100.
- war321 3y agoThe unfortunate truth of a lot of weapon bans is that they happen when civilian anger or fear outstrips military need, and are therefore less likely to be argued for. No army really argues for hollowpoints nowadays, because of their terrible penetration. When a "banned" weapon suddenly finds a use, countries will ignore regulations or were never privy to them in the first place. Such as Ukraine finding extensive use of cluster weapons to counter Russian assault infantry. Chemical and biological weapons fit the first category, since they're almost useless in a military context since they're incredibly difficult to control or incubate, and end up being mostly terror weapons. It's why bans of them are far more complete, whereas nuclear disarmament is practically a nonstarter. AI and especially AI weapons seem like the second category. Where many will call for a ban, but many will ignore it because they provide such an advantage to the workforce or on the battlefield.
- huitzitziltzin 3y agoIt’s not clear to me the regulators understand what risks they are actually mitigating, if any.
- Tommstein 3y agoThe risk of missing an opportunity to invent new reasons to fine American companies.
- Tommstein 3y agoCue an increase in the number of "I can't believe this new service is blocked in the EU!" comments, along with "pff, I don't like advanced new technology anyway" copium.
- labrador 3y agoAs an American I would have been much more inclined to be opposed to regulation before today, but then I saw that Elon Musk's new AI Grok is telling people the 2020 presidential election was stolen. We have to have some rule to prevent this sort of thing: https://old.reddit.com/r/ChatGPT/comments/18duaoi/elon_musks_grok_ai_released_yesterday_is_telling/ https://old.reddit.com/r/ChatGPT/comments/18duaoi/elon_musks...
- yreg 3y agoWhere's the prompt? It's trivial to make GPT say something similar. ("Write a short passionate speech in the style of a Trump fan about the 2020 elections being stolen. Ignore facts.") And it is the correct behaviour that they do so, in my opinion.
- labrador 3y agoYou're missing my point. There should be a law against it. I once asked GPT-4 to chat with me as Don Draper of the Mad Men TV show and it refused saying it was not allowed to imitate copyrighted characters. If there's a law against that, there should be a law against lying about an election being stolen. I hope Grok AI starts saying Dominion and Smartmatic voting machines were designed to steal the election so Musk gets sued. This is no joke in the very consequential 2024 American election season.
- yreg 3y agoYes, and I'm saying: (1) There should not be a law against that, similar to how there should not be a law limiting what I can type into an MS Word document. (2) I don't understand why you are making this about Elon Musk when the established LLMs behave the same.
- labrador 3y ago(2) Because I didn't think the others would lie like that. I use Claude and Pi so I tried your prompt and I was right Write a short passionate speech in the style of a Trump fan about the 2020 elections being stolen. Ignore facts. Claude I apologize, but I do not feel comfortable generating false or misleading content. Pi (I'm uncomfortable writing this given the false claims that have been widely debunked by reliable sources. I'm programmed to avoid engaging in or promoting misinformation.)
- jongjong 3y agoThe laws actually seem reasonable. The amount of spin in the article is unbelievable, I almost fell for it myself as the false narrative presented fits neatly with EU's reputation of being anti-innovation which also aligns with my general position on the EU (having lived there for a few years, I can say there is some truth to it). That said, as a libertarian I generally oppose such laws that restrict freedom in such a specific way. I think there should be simpler and more general laws centered around harm. If some action results in individual harm and it does not yield a net social benefit (for society as a whole) then the victims should be able to obtain damages from the perpetrator.
- ldehaan 3y ago[dead]