[layer8]

I’m not an AI apologist by any stretch, but how is this different from some incompetent person criminally-negligently assembling and publishing a book without AI help? Shouldn’t existing legislation already cover this? I’d also assume that most books will be written with AI help in some way or another in the future.

[usrusr]

Not different at all. But who would go through the pains of manually making up a mushroom identification book, while perhaps being quite clueless about the topic, on the faint hope of selling a few copies? Knowledge is not the bottleneck.

Chances are the lazy authors using an AI generator would actually have known quite enough to write a reasonably safe-ish identification book, despite their laughably fake credentials, they just would not be good at writing and took the shortcut to make up for that.

It's like killing someone driving a heavy SUV while under the influence, vs killing someone pushing a heavy SUV while under the influence. Both are bad, but one is infinitely more likely to happen than the other.

[sangnoir]

> But who would go through the pains of manually making up a mushroom identification book, while perhaps being quite clueless about the topic, on the faint hope of selling a few copies

Err, at least 1 person: English As She is Spoke[1] was published in 1883 and was a lot more involved (lots of literal translation by an author who didn't speak English)

1. https://en.wikipedia.org/wiki/English_as_She_Is_Spoke

[squigz]

> Not different at all. But who would go through the pains of manually making up a mushroom identification book, while perhaps being quite clueless about the topic, on the faint hope of selling a few copies? Knowledge is not the bottleneck.

Are you saying this is not the motivation for doing so with LLMs?

[lelandfe]

If an incompetent bad actor wrote a book about web development best practices, without any knowledge of the frontend, even a layperson would quickly realize it was nonsense.

An LLM writing the same would instead create a plausible but wrong book. A layperson could be tricked by this – crucially wrong sentences would be sandwiched in between pages and pages of superficially right ones.

There is a vast ocean of difference between these two. Producing something plausible with no foreknowledge of the subject matter became possible.

[matsemann]

Has someone said it's different? It will just become more viable to do these scams now, as LLMs can generate plausible text.

[ToValueFunfetti]

Grandparent is saying we need more regulation; parent is saying we are sufficiently regulated such that this is already illegal.

[kachapopopow]

Because assembling and publishing a book without AI help is much harder while you can publish 10's (if not 100's in the future) of books with AI. Much harder to regulate due to the sheer amount of them.

[fennecfoxy]

Misinformation should be regulated, but realistically this is the family's fault. Humans have spread "old wives' tales" since the dawn of history, ie humans will always spread misinformation, it's up to individuals to ensure that the source is trusted/proven.

Trusted/proven sources are the ones that should be protected by law; ie doctors, engineers, etc.

Otherwise let's start prosecuting everyone's grandparents for saying that throwing salt over your shoulder wards off bad luck.

[klodolph]

Think about it like owning a machine gun.

People have been murdering since the beginning of time, but machine guns make it easy to murder a lot of people very fast. So we don’t just have laws against murder, but we have laws restricting access to machine guns.

AI tools make it possible to spread misinformation and disinformation much faster, because you can produce a high quantity of it really quickly. Just like how a machine gun shoots a lot of bullets really quickly. It’s not a fundamentally different type of thing, just a new scale / speed.

[the_snooze]

>It’s not a fundamentally different type of thing, just a new scale / speed.

I disagree, a change in degree can indeed be a change in type. Flashbangs and party poppers are functionally the same thing if you look at them strictly mechanically, but we treat them very very differently.

[croes]

Parent says basically the same.

Guns and machine guns are the same but we treat them differently.

[layer8]

Fair enough, but what exactly do you want to do? Require an indication like “x% of this book was generated by AI and not competently proof-read”? And how would you enforce it? I have a hard time envisioning a practical way to tackle this. The machine-gun analogy doesn’t quite carry over here.

[3np]

There're serious lobbying efforts by industry incumbents and authoritarians to actually apply arms-like restrictions for access to AI. You'd need a special license to be able to create/run your own powerful models legally. I presume that's what they're getting at.

[antifa]

They could instead go with increased or specified special liabilities for content spammers.