HoskingcMar 19, 2023
Darius Kazemi

It seems that US Supreme Court Justice Neil Gorsuch and I are in agreement -- I have long claimed that the output of a bot is a speech act by its creator. If my bots say something shitty, that is on me. I think the creator of an AI should be liable for its outputs.

https://www.washingtonpost.com/politics/2023/03/01/ai-chatbots-may-have-liability-problem/

AI chatbots may have a liability problem

The Supreme Court stumbled on a futuristic tech debate

The Washington Post
Mar 1, 2023 at 5:05pmWeb
Show threadDarius KazemiMar 1, 2023

Opposing arguments are like this one, also quoted from the Washington Post article above:

"generative AI typically produces content only in response to prompts or queries from a user; these responses could be seen as simply remixing content from the third-party websites, whose data it was trained on"

If you are taking that broad a definition of "it's just a remix" then buddy, I have news for you about literally all creativity, language, and culture. By that logic, no one is liable for anything

Show threadDarius KazemiMar 1, 2023

Obligatory reference to "Everything is a Remix":

https://www.everythingisaremix.info/

Everything is a Remix

Watch the legendary series that reveals how creativity happens. Raise your creative game with tools, techniques and inspiration.

Everything is a Remix
Show threadDarius KazemiMar 1, 2023
Btw follow @willoremus for interesting reporting and analysis like the original linked article, you won't regret it
Show threadclar fonMar 1, 2023
@darius AI is just a corporate mixtape
Show threadfelix stalderMar 1, 2023

@darius Yeah. Frankly, I don't get how "it's just remix" could be construed as the entity doing the remix not being responsible for the quality of the remix.

This does not even hinge on the question of whether everything is a remix or whether the original creation exists. (I strongly lean toward the former, even if there are very original remixes....).

Show threadDarius KazemiMar 1, 2023
@festal right. It's just a very weak argument from someone desperate to deny liability
Show threadJESSE JANOWIAK HAS MOVEDMar 1, 2023
@darius My devil’s advocate argument would be that the user’s prompts are not just triggering responses, they are actively forming them. There is only so much safeguarding that an AI developer can implement against a user intent on making the AI say something dangerous or libelous.
Show threadJESSE JANOWIAK HAS MOVEDMar 1, 2023
Based on that reasoning, AI chat conversations are collaborative creations. I don’t know if there is a legal distinction there, but I would hope so.
Show threadgerbiMar 1, 2023
@darius AI (or more exactly machine learning) just moves the responsibility to whomever chooses the training data. And not even a conscious AI changes that, because even a conscious AI had someone choosing training data for it.
Show threadAli AlkhatibMar 1, 2023
@darius the apparent crisis this seems to be causing people is a little baffling. i'm honestly a little curious what outcome people thought we would be heading toward that would've been at all coherent.
Show threadDarius KazemiMar 1, 2023
@ali people will bend over backwards to come up with the most barely coherent framework to ensure they are not liable for their actions
Show threadAli AlkhatibMar 1, 2023

@darius i can't help but wonder if this is why tech companies are so weirdly keen to entertain goofball claims that AIs are sentient

if they can just clear that gorge, they'll be home free! lol

Show threadJeff CraigMar 1, 2023
@ali @darius Most people who claim AIs are at all sentient only do so up until the point that others start suggesting that, in that case, they should have Rights.
Show threadKevin Karhan Mar 1, 2023

@darius +9001%

This is the same in #Germany where one cannot deny or disavow #copyright and that one is always responsible for the outputs of one's product unless they can evidence otherwise [i.e. sabotage by the end user]...

Show threadKevin Karhan Mar 1, 2023
@darius Otherwise Neonazis would've spawned hoards of holocaust denial bots ages ago.
Show threadÉibhear 🔭Mar 1, 2023
@darius Does that not also mean that the creator should benefit from its outputs, too? Such as #copyright?
Show threadDarius KazemiMar 1, 2023
@eibhear I would think yes (though you would legally want to treat it the same as any remix)
Show threadÉibhear 🔭Mar 1, 2023
@darius I know there have been attempts throughout the world to get courts to agree that AI-generated images (in particular) can be copyrighted, and the general feeling is that they can't, on the grounds that humans are not involved in the creation.

If creators of an AI can be liable, it would be easy to argue that they should be granted #copyright on AI-generated works too. If they could get copyright, it would not be good.
Show threadDarius KazemiMar 1, 2023
@eibhear a big problem is that we are dealing with several different horrible broken legal apparatuses (including copyright) and trying to balance them for least harm
Show threadJohn CarlsenMar 1, 2023

@darius @eibhear

I think existing US copyright law is good enough.

A chatbot will substantively transform its data set based on a query, and output it according to a prompt and its algorithms. I'd say that makes its operator liable for its ouput (and possibly violating copyrights when creating its data set).

In contrast, a web search should accurately quote and cite sources, deflecting liability to those sources.

Show threadJesseMar 1, 2023
@darius Seems to me the bigger issue with this approach will be determining who created a given bot.
Show threadandeuxMar 1, 2023
@darius What exactly is the alternative? If creators are not responsible for the outputs of algorithms, then ... nobody is?
The people who get all worked up about the existential threat of "superintelligent AI" seem awfully blasé about not-so-intelligent systems spouting libel, giving inaccurate medical advice, or running people over with no accountability.
Show threadChip StewartMar 1, 2023
@darius I wrote about this & summarized a number of approaches a few years ago in my sci-fi & future media law book, with examples from folks like @marklemley & @gunkel. In short - maybe assign liability to programmers or hosts, but maybe also terminate the bots themselves?
Show threadCadence Larissa Beth BeresfordMar 1, 2023

@darius Who is responsible for what an AI creates? The AI's creator, the AI's user or the AI itself?

I say the answer is yes, all the above. What it is trained with, is important, what it is asked to do, is important and it's own capacity for reason, is important.

Show threadFrank CoteMar 1, 2023
@darius it's an interesting point. I hope you're not a pompous arrogant bastard like he is.
Show threadDarius KazemiMar 1, 2023
@Frankc1450 bad people can be sometimes correct!
Show threadFrank CoteMar 1, 2023
@darius yes, very true.
Show threadAnthony DiPierroMar 1, 2023

@darius Seems the question for 230 protection is whether the information was provided by another information content provider.

Which depends on the information. Was it a hallucination or was it just repeating someone else?

Show threadSJMar 2, 2023
@darius @anildash okay now it’s going to get interesting
Show threadjack the nonabrasiveMar 2, 2023

@darius @irwin Yet the output of a generative ML model isn’t copyrightable. It seems like, if this becomes precedent, holding a bot creator who uses generative tech to create the bot output is a wedge that could be used to pressure copyright for generative tech.

Which might not be desirable. Unless it’s a derivative work and the author owes royalties to holders of copyright on the training corpus?

Show threadAvi Rappoport (avirr)Mar 2, 2023
@darius Those who make the profit must also take responsibility — @j2bryson has been saying this for years
Show threadJules 🍺Mar 4, 2023
@darius I was going to say that it would mean we are responsible for what our kids say, but then I remembered that these ai bots are not independent entities unlike our progeny so yes the corporations are liable.