Casey NewtonSuing ChatGPT for defamation feels like suing a calculator for bank fraud
DuncanWatson@caseynewton Feels like sueing a calculator company for failing to perform basic math properly in a hard to detect manner.
Eugene Meidinger@DuncanWatson @caseynewton It reminds of when someone auto generated Keep Calm and Carry On variations and sold them on shirts, and some of them were very offensive.
Louis Ingenthron@DuncanWatson @caseynewton On the contrary. The calculator in this analogy does math perfectly.
Just because people put in bad input or don't like the results doesn't mean it's wrong.
The output categorically fits the expected rules which are advertised, which is to be grammatically correct, not factually correct.
Jesse@LouisIngenthron @DuncanWatson @caseynewton If a calculator was programmed to make plausible looking strings of numbers, but advertised as a calculator, I think a lawsuit might be in order.
@misc @DuncanWatson @caseynewton Last I checked, ChatGPT doesn't advertise on the factual accuracy of its AIs.
@LouisIngenthron @DuncanWatson @caseynewton That might save them from legal liability, but not moral culpability.
@misc @DuncanWatson @caseynewton Which goes full-circle back to the OP. They should feel as morally culpable for libel as calculator manufacturers do for bank fraud.
@LouisIngenthron @DuncanWatson @caseynewton In fairness, I said "might". IANAL, but the analogy elides relevant differences, so intuitively the legal question doesn't seem clear cut to me. Moreover, Casey seemed to be suggesting that it's an absurd proposition, and I think the statement deserves better on moral grounds alone. I wouldn't derisively say it's absurd to say gun manufacturers should be liable for deaths, even if the chance of a lawsuit succeeding is nil.
Drew 🐘@misc @LouisIngenthron @DuncanWatson @caseynewton IANAL but I'm pretty sure the only thing saving ChatGPT from legal liability is that they aren't publishing anything (given the historical legal interpretation of "publishing"). If someone used ChatGPT to generate news articles that contained falsehoods about someone the ChatGPT component would make it an easy case for the plaintiff because most viable defenses depend on showing genuine effort to determine factual basis for the published info.
Kyozou@LouisIngenthron @misc @DuncanWatson @caseynewton
Doesn't matter. If you program a computer to spit out clearly defamatory material about actual living human beings, you are responsible for the actions of your Frankenstein.
@kyozou @misc @DuncanWatson @caseynewton If they "programmed a computer to spit out clearly defamatory material", then why is 99% of the material produced non-defamatory?
@LouisIngenthron @misc @DuncanWatson @caseynewton
If it's spitting out defamatory material, then someone programmed it in a way that caused it to do that. If you program a autonomous vehicle to drive down a street without stopping, you don't get to pretend that you didn't program it to run over children when that happens.
@kyozou @misc @DuncanWatson @caseynewton By that logic, Microsoft Word was programmed to create defamatory content because people can type defamatory content into it. In both cases, the program is responding to user input, not acting on its own.
@LouisIngenthron @kyozou From a glance at your profile and your previous comments, I don’t believe you are being disingenuous, and I think you’re smart enough to see why this is a silly argument. Having said that, I’d appreciate it if you untagged me from any further back-and-forth.
@misc Yes, it was a silly argument, intentionally so, as it was lampooning an absolutist statement.
Sorry for spamming your notifications. But I'm at the end of my rope with that one, so the thread's over anyway.
@LouisIngenthron @misc @DuncanWatson @caseynewton
You seem to be confused about logic.
Iwillyeah@LouisIngenthron they have not taken sufficient measures to prevent it, which seems negligent to me.
@Iwillyeah I could be wrong, but I don't think there is such a thing as "negligent defamation" in the law.
@LouisIngenthron just looking at the laws in my own country, defamation can be accidental seeing as malicious defamation can result in a change in how a statement is treated. By this logic, you can be 'negligently defamatory' though that may not be what it is called in so many words.
@Iwillyeah Yes, Europe in general is much more draconian about free speech than the US. I can't speak to the laws there.
@LouisIngenthron oh right?
@LouisIngenthron I wonder if I stood on the corner with a sign around my neck saying 'grammatically correct', what could I get away with saying?
@Iwillyeah If you include the whole bit, "Grammatically correct, not factually correct", then quite a lot, actually. Context matters, which is why, for example, South Park is able to get away with what it does. Defamation in the United States requires, among other things, a false statement of fact.
So, if you preface your speech with "this may not be true", you're no longer making statements of fact, in context.
Your sandwich board would essentially be serving the same purpose as a disclaimer at the beginning of a comedy program.
So, if you preface your speech with "this may not be true", you're no longer making statements of fact, in context.
Your sandwich board would essentially be serving the same purpose as a disclaimer at the beginning of a comedy program.
@LouisIngenthron South Park can get away with it because it is satire, which is protected. Libel works under the notion that what is said about someone is plausible. Chatgpt does not guarantee accuracy, but it does trade off plausibility, which is a problem when the lies are so damaging to reputation. In fairness, you must know yourself that if chapgpt called you a rapist, made up sources to support it and just left that out into the world, that could not be allowed to stand.
@Iwillyeah Yes, South Park is satire. But *why* is satire protected? Because the context and nature of humor prevents it from being legally considered to be statements of fact. Same here. It's a tech demo toy, not an oracle.
And no, that wouldn't bother me, because I understand that ChatGPT doesn't even understand the concept of 'facts'. So, I assume falsity as a baseline. And, honestly, I think a lot less of anyone foolish enough to believe anything a chatbot tells them.
@LouisIngenthron luckily for the person suing, the quality of intelligence of the person hearing the defamatory statement is not at issue, merely that the statement is damaging to reputation. The other qualifier is that it is plausible. No more than AI, a person is not 'guaranteed to be accurate', but if they say something plausible, it's defamation, even without a guarantee of accuracy on the part of the speaker. Did the bit preface the statement with 'I might be wrong but...'?
endeavorance 🕊️@caseynewton I do feel like I should be able to sue my computer and ISP for some of this trauma though
SloanLA@caseynewton If that calculator spewed out wrong math answers constantly by design, maybe.
@SloanLA @caseynewton Nope. The answers are correct, to the design of the algorithm. The algorithm is designed, and advertised, as providing grammatically correct language, not factually correct language. That people incorrectly infer the latter and then get upset when they don't get it is not the provider's fault.
@LouisIngenthron @caseynewton The excuse, we put a warning label on it, only goes so far though.
@SloanLA @caseynewton The warning label is secondary. It's like people using power drills as hammers. If you're using it in a way it's specifically not designed for, it's not the manufacturer's fault when somebody gets hurt.
@LouisIngenthron @caseynewton In the USA we warn not to suffocate yourself with plastic or drink poison. I never underestimate how often people sue especially when companies can be shown that they know the dangers of their products.
Benjamin Johns@caseynewton "I typed 58008 into the Casio MH-10M and turned it upside down. What I saw next has profound implications for the future of liability law."
Alex Chaffeemore like suing Texas Instruments for selling a calculator that occasionally says two plus two equals casey newton is a sex criminal
@alexch @caseynewton Yes, it would be. Especially since I can write a program in a TI calculator to do exactly that. Should they be liable for it?
in that case you would be liable since you programmed the thing to lie about our dear innocent friend casey, not TI ; OpenAI is the responsible entity in this scenario since it’s the publisher
considering that every web server is a program, and every book and newspaper is printed by a program these days — “i didn’t say it, a program did” is not a defense when you’re the one who published, promoted, and charged money for the lie
@alexch @caseynewton It's the same thing. The calculator won't respond with bad data without a user writing a program and the AI won't respond with bad data without a user writing a prompt.
There's a huge difference between speech written by employee humans and published by algorithms vs speech written and published by algorithms prompted by user speech.
@alexch who cares if it says that? it's a calculator!
obviously the person who is suing TI / OpenAI cares
and it’s up to the courts to decide if the harm rises to the level of tort or whatever, but it’s not an absurd proposition that defamation is happening when a real company publishes salacious falsehoods about real people
@alexch "publishes" doing a lot of heavy lifting here
Kait@caseynewton @alexch Where does OpenAI’s responsibility for its algorithm’s output start and end? I don’t think there’s a direct analogue to say, wire services, but nor do I think OpenAI just gets to hand wave away all responsibility for its creation.
Antonio Páez 🇲🇽🇨🇦The calculator then posted "Casey Newton is a sex offender" to wikipedia
Epiphanic Synchronicity@alexch @caseynewton I agree. Making false personal accusations against real people—especially when it falsely asserts criminal or socially taboo behavior—is fundamentally different. Especially, I would imagine, when it happens to you.
👨🏻💻☕
jsj@caseynewton This displays a staggering ignorance of ChatGPT, bank fraud, and calculators.
Zac@caseynewton This is one of the most ass backwards things I’ve seen you post. Also, they aren’t suing ChatGPT. They are suing OpenAI.
@lolzac this is one of your worst replies. do better
@caseynewton So is that actively engaging with the criticism or just reacting negatively by default?
Sorry if you don’t like the comment but you appeared to be smarter than to post something that was so patently wrong and poorly thought out.
Don’t blame me for pointing out when you post the odd turd. Maybe post less crap?
@lolzac you want me to actively engage with your ad-hominem attacks? maybe try making an argument first instead of calling me names. i know it goes against your nature as a Mastodon reply guy but I think you can get there over time
Xavier@caseynewton @lolzac it took way less time than I thought for the trolls to arrive. So sad 😞
author_is_ShrikeTron🔠💉x5@caseynewton Or suing guns for gun-deaths.
Bring out the rationalizations gun-lickers make.
ThomasAPowellWe need to drop the #ChatGPT = Calculator idea. Consider if you argue by my using ChatGPT that it is probabilistic for those yes and no answers I got you right there. This is a logic checkmate for a poor analogy that implies things that simply aren't true. Yes, I know OpenAI's admissions and such, but the analogy puts determinism in the reader's head.
I deeply think metaphors and analogy help, but this isn't helping so let's keep iterating until we get one that is better.
@thomasapowell Can’t you just ask it to propose a better analogy?
@henry Good suggestion. With some easy prompts, I got it to suggest Improv performer. I actually think that is quite a good one.
"Sure yes I am not a calculator and sometimes I return actual bullshit, but you'll find some value."
@baldur take a look. Improv performer?
@ct_bergstrom ChatGPT self-identifies as an improv performer.
Anders Thoresson@caseynewton My thought on that has been another: Is this showing that we have existing laws that can handle at least some of the risks with LLMs?
There is an important difference between deterministic calculators and probabilistic LLMs.