EU kicks off landmark AI law enforcement | Hacker News Reader

EU kicks off landmark AI law enforcement

cnbc.com

37 pointsbelter20 comments

> The Commission proposed enshrining in EU law a legal definition of 'AI system' referring to a range of software-based technologies using specific techniques and approaches ('machine learning', 'logic and knowledge-based' systems, and 'statistical' approaches) that could be complemented through the adoption of delegated acts to factor in technological developments.

This seems very broad. Almost anything could be a logic and knowledge-based system. And - genuine question - isn't ML a statistical approach?

Yep! Broad. The new EU way. Leave everything open to interpretation so that can slapp you if they don't like you.

Nope, but nice try...

"Given its abstract nature, the legislation contains areas that are yet to be fully defined. These are expected to be elaborated on through delegated and implementing acts, guidelines by the EU institutions, as well as harmonized standards developed by European Standardization Organizations. As a result, businesses can expect to receive more detailed guidance in the near future."

Scope of the AI Act:

"Furthermore, the recitals – which clarify the AI Act’s regulatory text – specify that the definition of AI does not include basic traditional software or purely rule-based programming created by humans for automatic operations. Despite this, the definition remains wide, covering the majority of systems available on the market.

...Finally, the Commission has been tasked with developing guidelines for applying the definition of an AI system, which grant further guidance of the defining aspects of AI under this regulation."

https://www.deloitte.com/content/dam/assets-zone2/nl/en/docs...

The irony is: AI is a silly thing to regulate. Why not regulate any software system that gives a bad outcome? Who cares about the implementation?

The only reason seems to be so the EU can claim to be a leader in AI regulation; a dubious accolade at best.

On one hand I'm scared of my CV for example being rejected by some LLM. On the other hand I doubt people in EU Parl have any (enough?) technical knowledge to create law regarding this matter

But still: something scraping your CV for keywords, or failing to transform it into whatever evil format the recruiter requires, isn't AI, but might still reject your CV or cause it to be rejected. What's special about AI?

>What's special about AI?

VC firms have brainwashed people into thinking LLMs are "I" when really, they are XOR'ing GIGO machines.

Because executives, everywhere it seems, are gullible enough to be tricked into believing that A(sigh)I isn't a colossal waste of money and electricity, they are throwing it at every problem, everywhere.

This means that a fucking stupid framework of frameworks designed to turn electricity into heat via compute is being inserted into every single aspect of commerce, banking, law enforcement, warfare, business decision making, and yes, hiring.

The difference between scraping a resume for keyworks and having an LLM evaluate it is that you can look at the list of keywords and understand what it is happening.

A list of keywords is comprehensible and reproducible. You can subpoena a list of keywords. You can modify and control a list of keywords. You can legislate the requirement or forbidding of keyword criteria.

An LLM is just a tunnel of sewage that pours shit out the other end, a black box, a mystery.

But A(sigh)I excels at letting people be lazy and generating an endless stream of dreck designed to addict the gullible so it is going to take over the world.

At least the EU is trying to slow it.

> Who cares about the implementation?

I guess one difference between LLM-style AI and most other systems is that when other systems produce the wrong output you can usually step through the code, figure out why, and be confident that the same issue won't happen again.

This doesn't change the fact that the legislation is arbitrary until it is more thoroughly defined. And if they don't like you enough they will make sure to clarify that area first

Link to actual legislation: https://eur-lex.europa.eu/eli/reg/2024/1689

> Chapters I and II shall apply from 2 February 2025;

Yes, from a Sunday! Is that common in legislations? Effective from a weekend?

Sounds like a being a bit in a hurry.

The EU is great at creating regulations to strangle young technologies in their crib. The "AI Act" has noble-sounding goals, but the bureaucrats who created this have no idea what it means or what effects it will have.

What I haven't been able to discover, is exactly who is behind the AI Act. It would be very informative to know which lobbyists and organizations are pushing for it.

The EU has a serious problem with hidden lobbying, but instead of propagating conspiracy theories maybe better read the details of the law. It's fairly balanced and transparent.

Artificial Intelligence – Questions and Answers: https://ec.europa.eu/commission/presscorner/detail/en/qanda_...

The regulation: https://www.europarl.europa.eu/topics/en/article/20230601STO...

https://www.deloitte.com/content/dam/assets-zone2/nl/en/docs...

It's not about how the law appears to be. It's about the damage that can be caused through interpretation of the law that's the problem.

I question young technologies considering ML (which 'AI' is a form of) has really been around since the 80's, it's just in its current form it is new. I would also state that this is primarily focusing on aspects that could considerably impact people's lives in a negative manner if used incorrectly especially as it has inherent bias. Facial recoqnition is already used within certain police forces, and without regulations on detection threshold, this could end up with significant issues regarding false recoqnition from police forces who already have significant bias.

Ah, but there are loopholes for government use.

Sorry dude. That's personal information. Privacy laws prevent you from finding that out. That's your privacy rights at work in the way they are intended.

God forbid we learn anything from our experience with social media and try to think something through before transforming society with it.

Have we ever successfully thought through any new technology before implementing it, though? At the launch time of social media transforming society was clear VC hype and due to that implicitly ignored. "Dude it is just a website.".

I cannot think of a single time that a transformative technology was "thought through" ahead of time. Such standards just become basically intellectual NIMBYism, demanding onerous standards to try to make it go away.

Reply on news.ycombinator.com