The attorney general, James Uthmeier, a Republican, said the messages suggested that ChatGPT “offered significant advice to the shooter before he committed such heinous crimes.” He pointed to several exchanges, including ones in which the suspect asked about a gun’s power at short range and which ammunition might be used for it.
“My prosecutors have looked at this, and they’ve told me if it was a person on the other end of the screen, we would be charging them with murder,” Mr. Uthmeier said at a news conference in Tampa.
Two adults died and six other people, including at least one student, were injured last April at the shooting near the student union at Florida State, a public university with an enrollment of more than 43,000 in Tallahassee. The suspect, Phoenix Ikner, who was then a 20-year-old student at the university, faces multiple charges of murder and attempted murder and is in jail awaiting trial.
Mr. Uthmeier first announced on April 9 that his office would be opening an investigation into OpenAI and ChatGPT. On Tuesday, he said a civil investigation that his office initiated early this month into the company’s potential liability would continue, along with the criminal investigation.
The things they are saying ChatGPT said sounds like they would have been also answered equally well by googling. information about firearms how the country reacts to mass shootings when busy times...
The things they are saying ChatGPT said sounds like they would have been also answered equally well by googling.
information about firearms
how the country reacts to mass shootings
when busy times are at school
It doesn’t sound like they are alleging that ChatGPT encouraged the shooting.
I think you're right on the facts, but I also think it's probably a little thornier than that in practice. Laws on aiding and abetting, as far as I understand them, don't require that the aid is...
I think you're right on the facts, but I also think it's probably a little thornier than that in practice. Laws on aiding and abetting, as far as I understand them, don't require that the aid is unique information. Meaning that if I tell you how to commit X crime, with cause to reasonably believe that you were planning to commit a crime, it doesn't absolve me that you could've gotten the information elsewhere. But - most laws (US federal statute, anyway) also require that I wished to bring the criminal act about and sought to make it succeed. That's going to be difficult to prove, I think, because this is in part a philosophical question about whether AI can "want" anything. There are also going to be technicalities, I think, on whether chatgpt is a publisher (reporting extant information) or generating new information or what.
To a certain extent I'm taking the AG at his word when he says that his attorneys would charge a human, based on the transcripts. That might be foolish of me. Who knows.
The point I'm making here though, is that if I go and Google those items, Google is clearly not expected to determine my intent and then report me to the police, despite the fact that they log all...
The point I'm making here though, is that if I go and Google those items, Google is clearly not expected to determine my intent and then report me to the police, despite the fact that they log all my queries into a database and analyze them for all kinds of things, such as shopping intent. Why should some other text input box that I type questions into have different rules?
And I guess another concern here is, I want the default state of things to be that if the government wants to see things I say, they should find their own probable cause and get a warrant. Companies shouldn't have a positive legal duty to report communications to the government.
If you type a question into Google, the websites supplying those answers are who would be liable and it's likely a prospective shooter would consult more than one so the responsibility is...
If you type a question into Google, the websites supplying those answers are who would be liable and it's likely a prospective shooter would consult more than one so the responsibility is diffused. Google doesn't claim that its search engine directly tells you anything, it refers you to information sources (unless you mean the AI summary, I'm not sure about that). By comparison, answers ChatGPT give you are presented as coming from the chatbot itself.
When I ask ChatGPT questions, it usually does a web search and then gives me citations. I doubt a website that has information on guns or actually almost anything related to these queries would...
When I ask ChatGPT questions, it usually does a web search and then gives me citations. I doubt a website that has information on guns or actually almost anything related to these queries would bear any liability whatsoever.
Philosophically and legally, I agree that companies shouldn't have a positive legal duty to report communications. But unless I've missed something (please correct me if so), I think the question...
Philosophically and legally, I agree that companies shouldn't have a positive legal duty to report communications. But unless I've missed something (please correct me if so), I think the question at hand is something more like "should gpt have refused to answer" (and by extension, is that even possible if the user was only implying things? and what about jailbreaking?) and not "should it have reported the user to the police."
As far as whether Google and gpt should have different rules - I don't know. I can think of a few frameworks arguing in favor and against. There's clearly something materially different about a source of collated information (gpt) than the library itself (Google), though, if we're talking about the actual 'information' side of things, and I think the precedents support that.
It depends on how that went though asking the busy times at school is very different than asking the best times to cause the most publicity and/or damage. Asking general questions about guns is...
It depends on how that went though asking the busy times at school is very different than asking the best times to cause the most publicity and/or damage. Asking general questions about guns is very different than asking what will kill the most people the quickest.
Idk about Google, but if I send a comment about committing mass violence on Snapchat the FBI will tell my local PD and someone(s) will come talk to me. If I asked the same question to my campus chatbot, someone would be coming and talking to me. I know because I've seen it happen to students.
Should ChatGPT be responsible for not reporting that information? For providing the specific "how to commit murder" details being asked for? If it ever were to be a person we would need to consider if we could hold it accountable for crimes, but since it's not, what basic safety rails are they missing this time? Because this was predictable to anyone with any risk management oriented brain cells, much less people with billions of dollars to spend on, say, internet risk management experts. Did a lawyer or expert recommend something that the company didn't follow through on? Did an employee raise this concern and get told to shut up?
We'll see. I suspect it goes beyond questions asked to Google and into those more conversational questions that made his intentions known - at least based on the initial information.
Man, this is going to be complicated. No idea what to think about the potential consequences. It seems fair, though, for lack of a better word:
Man, this is going to be complicated. No idea what to think about the potential consequences. It seems fair, though, for lack of a better word:
My prosecutors have looked at this, and they’ve told me if it was a person on the other end of the screen, we would be charging them with murder,” Mr. Uthmeier said at a news conference in Tampa.
Realistically they're looking at the people in charge - what safety rails are on this product you sell to ensure that it won't teach someone how to murder people better? We already established...
Realistically they're looking at the people in charge - what safety rails are on this product you sell to ensure that it won't teach someone how to murder people better?
We already established they were pretty shit at not letting it tell people how to kill themselves.
The things they are saying ChatGPT said sounds like they would have been also answered equally well by googling.
It doesn’t sound like they are alleging that ChatGPT encouraged the shooting.
I think you're right on the facts, but I also think it's probably a little thornier than that in practice. Laws on aiding and abetting, as far as I understand them, don't require that the aid is unique information. Meaning that if I tell you how to commit X crime, with cause to reasonably believe that you were planning to commit a crime, it doesn't absolve me that you could've gotten the information elsewhere. But - most laws (US federal statute, anyway) also require that I wished to bring the criminal act about and sought to make it succeed. That's going to be difficult to prove, I think, because this is in part a philosophical question about whether AI can "want" anything. There are also going to be technicalities, I think, on whether chatgpt is a publisher (reporting extant information) or generating new information or what.
To a certain extent I'm taking the AG at his word when he says that his attorneys would charge a human, based on the transcripts. That might be foolish of me. Who knows.
The point I'm making here though, is that if I go and Google those items, Google is clearly not expected to determine my intent and then report me to the police, despite the fact that they log all my queries into a database and analyze them for all kinds of things, such as shopping intent. Why should some other text input box that I type questions into have different rules?
And I guess another concern here is, I want the default state of things to be that if the government wants to see things I say, they should find their own probable cause and get a warrant. Companies shouldn't have a positive legal duty to report communications to the government.
If you type a question into Google, the websites supplying those answers are who would be liable and it's likely a prospective shooter would consult more than one so the responsibility is diffused. Google doesn't claim that its search engine directly tells you anything, it refers you to information sources (unless you mean the AI summary, I'm not sure about that). By comparison, answers ChatGPT give you are presented as coming from the chatbot itself.
When I ask ChatGPT questions, it usually does a web search and then gives me citations. I doubt a website that has information on guns or actually almost anything related to these queries would bear any liability whatsoever.
Philosophically and legally, I agree that companies shouldn't have a positive legal duty to report communications. But unless I've missed something (please correct me if so), I think the question at hand is something more like "should gpt have refused to answer" (and by extension, is that even possible if the user was only implying things? and what about jailbreaking?) and not "should it have reported the user to the police."
As far as whether Google and gpt should have different rules - I don't know. I can think of a few frameworks arguing in favor and against. There's clearly something materially different about a source of collated information (gpt) than the library itself (Google), though, if we're talking about the actual 'information' side of things, and I think the precedents support that.
It depends on how that went though asking the busy times at school is very different than asking the best times to cause the most publicity and/or damage. Asking general questions about guns is very different than asking what will kill the most people the quickest.
Idk about Google, but if I send a comment about committing mass violence on Snapchat the FBI will tell my local PD and someone(s) will come talk to me. If I asked the same question to my campus chatbot, someone would be coming and talking to me. I know because I've seen it happen to students.
Should ChatGPT be responsible for not reporting that information? For providing the specific "how to commit murder" details being asked for? If it ever were to be a person we would need to consider if we could hold it accountable for crimes, but since it's not, what basic safety rails are they missing this time? Because this was predictable to anyone with any risk management oriented brain cells, much less people with billions of dollars to spend on, say, internet risk management experts. Did a lawyer or expert recommend something that the company didn't follow through on? Did an employee raise this concern and get told to shut up?
We'll see. I suspect it goes beyond questions asked to Google and into those more conversational questions that made his intentions known - at least based on the initial information.
Man, this is going to be complicated. No idea what to think about the potential consequences. It seems fair, though, for lack of a better word:
Realistically they're looking at the people in charge - what safety rails are on this product you sell to ensure that it won't teach someone how to murder people better?
We already established they were pretty shit at not letting it tell people how to kill themselves.
What we do know though, is that there will be no consequences for gun manufacturers, changes to gun laws, etc.