> While attorneys swear an oath to set aside their personal prejudices, biases, and beliefs to faithfully uphold the law and represent their clients, generative artificial intelligence is the product of programming devised by humans who did not have to swear such an oath. As such, these systems hold no allegiance to any client, the rule of law, or the laws and Constitution of the United States (or, as addressed above, the truth). Unbound by any sense of duty, honor, or justice, such programs act according to computer code rather than conviction, based on programming rather than principle
Minor quibble: I think the majority of bias would creep in when training, not necessarily during writing the code. If trained on biased data you'll get biased results. RLHF by it's nature is likely to introduce bias, but maybe bias most are ok with, but maybe not lawyers. Later, while actually serving responses, it's possible that OpenAI/others introduce checks for certain things (e.g. does the input contain the word 'gun'? If so refuse to answer) that are also sources of bias. In theory you could also taint the responses by changing the prompt, but that seems unlikely.
All this just to say that I think the judge is right to do this, his bias argument is a bit of a miss.
The whole paragraph except for the first and last sentences may have been composed using an LLM.
> All attorneys appearing before the Court must file on the docket a certificate attesting either that no portion of [...] or that any language drafted by generative artificial intelligence was checked for accuracy, using print reporters or traditional legal databases, by a human being [...] held responsible under Rule 11 for the contents of any filing that he or she signs and submits to the Court, regardless of whether generative artificial intelligence drafted any portion of that filing
Asking the attorneys to re-acknowledge that documents they file are their official entries into the record, no matter what programs are used to generate them, makes sense in principle as a way to preempt the "algorithm told us to" argument.
I think the context here suggests that the training is considered part of the overall task of "programming," though people in the industry use that term of art differently in the context of ML development. At any rate the creators of the model or tool were under no oath, nor is (or can be) the tool itself, and the results must be considered in the light of that fact.
Also basic computation can just be done via basic lookup tables. It's a principle that even FPGAs use. One the basic building blocks of an FPGA is a LUT. If you want to make a configurable basic gate just have a mux address a table of 4 bits, and you can then configure that to be any 2 input logic gate. FPGAs also have the word programming in their name.
As for an ML model it's basically a table of weights each connected to each other. In a sense we use "training" to find the weights to program into each node of each layer of the network.
I would still say an ML model is a program, just that humans did not directly specify each value this program. If we go back to an FPGA people rarely manually instantiate the raw building blocks instead it synthesized and routed most often by other software. Although VHDL and Verilog and such are closer to a program than training. Even with normal software code is compiled to machine instructions instead of directly specified. We still say your programming a computer even your using a compiler.
Just because the end results are based off of statistics and your input data I would still say training is a form programming. Just training tells you it was very indirect.
That aside I would agree that model or tools, and creator were under no oath so the reason here makes sense.
Single thing or human is always biased. Because that's how mind works. Out of zillions of possible guesses and decisions it uses only a small coherent subset. Which is always a non-representative subset. I.e. it's biased.
This is a local procedural rule for one federal judge in Texas. It affects a few dozen people (mostly lawyers). Why was it on HN #1? Are people misinterpreting the title, thinking it's something that remotely matters?
Is it not fascinating to watch changes to our world in real time due to genAi? This is small but its likely the very first of many rules that will become huge
It's a good and well-explained rule. It's worth reading regardless of prevalence, and one judge making one procedural change to start with is one of the ways in which the Law can start to evolve and change to adapt to new circumstances.
> Accordingly, the Court will strike any filing from an attorney who fails to file a certificate on the docket attesting that the attorney has read the Court’s judge-specific requirements and understands that he or she will be held responsible under Rule 11 for the contents of any filing that he or she signs and submits to the Court, regardless of whether generative artificial intelligence drafted any portion of that filing
Surely an attorney should be held responsible for any and all of the contents they file to a court. How much of that is generated/prepared by junior or trainee attorneys, paralegals, or researchers etc.
>> As such, these systems hold no allegiance to any client, the rule of law, or the laws and Constitution of the United States (or, as addressed above, the truth).
Note that while AI systems can be misaligned in general and can hold no allegiances, they are not above the law. Nothing is above the law.
This is super reasonable, as all it demands at minimum is that a human has reviewed and checked the work of any AI-generated filing.
That’s an obvious necessity - which, if my attorney failed to do, I would fire him before leaving the courtroom! How could anyone think that was ok, in court, where consequences of an error could be severe.
Mandating truthful disclosure to the public and customers is very good! Just like with required list of ingredients. I made a petition for this much more generally:
Why do people create petitions with only 10 signatures? How much impact do you think an online petition with a goal of 10 signatures can have? (Maybe the goal is dynamically updated but it still feels weird)
Sharing a link on HN doesn't demand the same level of effort nor does it purport to have the same impact as creating a petition on the platform "change".
Change.org dynamically updates the goal I think, it's usually slightly higher than the current signature count, I guess to make people think their signature will help hit that "important" goal.
Yeah, that's what I thought, but I still think it does the opposite job at first, making the petition owner seem to have no confidence in his petition.
In federal courts, when you file something under your account (often, attorneys have staff do the clicking to file), it is the equivalent of signing. And what you sign, you vouch for.
This standing order reiterates what any practitioner _should_ know. It a rule which states that you should follow the rules! But as we learned from the Air Avianca filing, at least one lawyer is missing something very basic about what it means to sign a filing in federal court.
```
All attorneys appearing before the Court must file on the docket a certificate attesting either that no portion of the filing was drafted by generative artificial intelligence (such as ChatGPT, Harvey.AI, or Google Bard) or that any language drafted by generative artificial intelligence was checked for accuracy, using print reporters or traditional legal databases, by a human being. These platforms are incredibly powerful and have many uses in the law: form divorces, discovery requests, suggested errors in documents, anticipated questions at oral argument. But legal briefing is not one of them. Here’s why. These platforms in their current states are prone to hallucinations and bias. On hallucinations, they make stuff up—even quotes and citations. Another issue is reliability or bias. While attorneys swear an oath to set aside their personal prejudices, biases, and beliefs to faithfully uphold the law and represent their clients, generative artificial intelligence is the product of programming devised by humans who did not have to swear such an oath. As such, these systems hold no allegiance to any client, the rule of law, or the laws and Constitution of the United States (or, as addressed above, the truth). Unbound by any sense of duty, honor, or justice, such programs act according to computer code rather than conviction, based on programming rather than principle. Any party believing a platform has the requisite accuracy and reliability for legal briefing may move for leave and explain why. Accordingly, the Court will strike any filing from an attorney who fails to file a certificate on the docket attesting that the attorney has read the Court’s judge-specific requirements and understands that he or she will be held responsible under Rule 11 for the contents of any filing that he or she signs and submits to the Court, regardless of whether generative artificial intelligence drafted any portion of that filing.
```
> While attorneys swear an oath to set aside their personal prejudices, biases, and beliefs to faithfully uphold the law and represent their clients, generative artificial intelligence is the product of programming devised by humans who did not have to swear such an oath.
While I like the idea, the reasoning here is kinda silly, even if some developer had sworn such an oath it is entirely unclear how that should affect a AI. We just don't have good techniques to prevent llms from hallucinating.
Not just the developer, but anyone who posted any sort of text in the training data. I assume that includes a pretty wide majority of English speakers. That makes the likelihood of actual agreement with the oath for all “participants” in the AI training low, to say the least.
That point about swearing an oath is a good one. It’s a reasonable position (assuming it’s the court’s position) that an AI (assuming this is an “AI”) being asked to provide testimony or a legal argument is doing so on behalf of all who trained it. Why would “my” training argue in “your” case? What if one doesn’t agree with the oath?
One of hardest, time consuming, and most important parts of being a judge is understanding the issues presented before you. Be it generative AI or communal animal husbandry.
There are clearly gross examples of judges failing in this duty, but others take that duty extremely seriously and I've read a number of judicial opinions on technology-related cases that demonstrated remarkable, nuanced grokking of the matter at hand.
Lordy this is dumb. There's been exactly one recorded instance of a lawyer using ChatGPT in a brief and not checking it. This isn't a meaningful problem.
There has been one news article that has gained enough traction to get widespread notice. That’s not even remotely the same as “one recorded instance.”
The order, already posted on this thread, lists multiple instances where generative AI would be useful in the law, but also a number of issues that I haven’t seen discussed before.
...Yet. This isn't a meaningful problem, yet. SimpsonsMeme.jpg
Sarcasm aside, there are already cases (I think at least two) of this happening in the wild, why not nip it in the bud before some high profile event really catapults it into the public eye?
In court, there is no meaningless problem or detail. Especially any detail pertaining to the structuring of legal arguments. And even more especially any details pertaining to the soundness of legal arguments and evidence. Legal experts obsessing over stuff like this is the only reason it's even vaguely possible to conduct a fair trial at all.
There's only been one big one which made the news. So far. There have almost certainly been more which have gone under the radar, or which weren't caught.
http://web.archive.org/web/20230530223247/https://www.txnd.u...
https://archive.ph/JpSsf
> While attorneys swear an oath to set aside their personal prejudices, biases, and beliefs to faithfully uphold the law and represent their clients, generative artificial intelligence is the product of programming devised by humans who did not have to swear such an oath. As such, these systems hold no allegiance to any client, the rule of law, or the laws and Constitution of the United States (or, as addressed above, the truth). Unbound by any sense of duty, honor, or justice, such programs act according to computer code rather than conviction, based on programming rather than principle
Minor quibble: I think the majority of bias would creep in when training, not necessarily during writing the code. If trained on biased data you'll get biased results. RLHF by it's nature is likely to introduce bias, but maybe bias most are ok with, but maybe not lawyers. Later, while actually serving responses, it's possible that OpenAI/others introduce checks for certain things (e.g. does the input contain the word 'gun'? If so refuse to answer) that are also sources of bias. In theory you could also taint the responses by changing the prompt, but that seems unlikely.
All this just to say that I think the judge is right to do this, his bias argument is a bit of a miss.
The whole paragraph except for the first and last sentences may have been composed using an LLM.
> All attorneys appearing before the Court must file on the docket a certificate attesting either that no portion of [...] or that any language drafted by generative artificial intelligence was checked for accuracy, using print reporters or traditional legal databases, by a human being [...] held responsible under Rule 11 for the contents of any filing that he or she signs and submits to the Court, regardless of whether generative artificial intelligence drafted any portion of that filing
Asking the attorneys to re-acknowledge that documents they file are their official entries into the record, no matter what programs are used to generate them, makes sense in principle as a way to preempt the "algorithm told us to" argument.
I think the context here suggests that the training is considered part of the overall task of "programming," though people in the industry use that term of art differently in the context of ML development. At any rate the creators of the model or tool were under no oath, nor is (or can be) the tool itself, and the results must be considered in the light of that fact.
Also basic computation can just be done via basic lookup tables. It's a principle that even FPGAs use. One the basic building blocks of an FPGA is a LUT. If you want to make a configurable basic gate just have a mux address a table of 4 bits, and you can then configure that to be any 2 input logic gate. FPGAs also have the word programming in their name.
As for an ML model it's basically a table of weights each connected to each other. In a sense we use "training" to find the weights to program into each node of each layer of the network.
I would still say an ML model is a program, just that humans did not directly specify each value this program. If we go back to an FPGA people rarely manually instantiate the raw building blocks instead it synthesized and routed most often by other software. Although VHDL and Verilog and such are closer to a program than training. Even with normal software code is compiled to machine instructions instead of directly specified. We still say your programming a computer even your using a compiler.
Just because the end results are based off of statistics and your input data I would still say training is a form programming. Just training tells you it was very indirect.
That aside I would agree that model or tools, and creator were under no oath so the reason here makes sense.
I define bias as anything I don't like.
Since I don't think I'm alone in defining bias this way, how about we all decide on a more objective standard such as "factual accuracy" ?
Single thing or human is always biased. Because that's how mind works. Out of zillions of possible guesses and decisions it uses only a small coherent subset. Which is always a non-representative subset. I.e. it's biased.
This is a local procedural rule for one federal judge in Texas. It affects a few dozen people (mostly lawyers). Why was it on HN #1? Are people misinterpreting the title, thinking it's something that remotely matters?
It follows from recent HN discussions:
ChatGPT-Authored Legal Filing “Replete with Citations to Non-Existent Cases" (62 points, 71 comments) https://news.ycombinator.com/item?id=36092509
Lawyer cites fake cases invented by ChatGPT, judge is not amused (343 points, 306 comments) https://news.ycombinator.com/item?id=36097900
A man sued Avianca Airline – his lawyer used ChatGPT (167 points, 129 comments) https://news.ycombinator.com/item?id=36095352
It's also likely the first of similar local rules in other courts, and it wouldn't surprise me if an equivalent rule is eventually added to the FRCP.
Is it not fascinating to watch changes to our world in real time due to genAi? This is small but its likely the very first of many rules that will become huge
It's a good and well-explained rule. It's worth reading regardless of prevalence, and one judge making one procedural change to start with is one of the ways in which the Law can start to evolve and change to adapt to new circumstances.
> Accordingly, the Court will strike any filing from an attorney who fails to file a certificate on the docket attesting that the attorney has read the Court’s judge-specific requirements and understands that he or she will be held responsible under Rule 11 for the contents of any filing that he or she signs and submits to the Court, regardless of whether generative artificial intelligence drafted any portion of that filing
Surely an attorney should be held responsible for any and all of the contents they file to a court. How much of that is generated/prepared by junior or trainee attorneys, paralegals, or researchers etc.
>> As such, these systems hold no allegiance to any client, the rule of law, or the laws and Constitution of the United States (or, as addressed above, the truth).
Note that while AI systems can be misaligned in general and can hold no allegiances, they are not above the law. Nothing is above the law.
Except, say, nature. Can’t much legislate gravity out of existence, can we.
This is super reasonable, as all it demands at minimum is that a human has reviewed and checked the work of any AI-generated filing.
That’s an obvious necessity - which, if my attorney failed to do, I would fire him before leaving the courtroom! How could anyone think that was ok, in court, where consequences of an error could be severe.
Mandating truthful disclosure to the public and customers is very good! Just like with required list of ingredients. I made a petition for this much more generally:
https://www.change.org/p/mandate-disclosures-to-mitigate-spa...
Why do people create petitions with only 10 signatures? How much impact do you think an online petition with a goal of 10 signatures can have? (Maybe the goal is dynamically updated but it still feels weird)
I don’t think people know in advance the petition will not attract enough attention.
Why do people make comments with only 17 views?
Why do people make Hacker News submissions with zero upvotes?
Zing! Your reply is def better than mine :)
Sharing a link on HN doesn't demand the same level of effort nor does it purport to have the same impact as creating a petition on the platform "change".
Change.org dynamically updates the goal I think, it's usually slightly higher than the current signature count, I guess to make people think their signature will help hit that "important" goal.
Yeah, that's what I thought, but I still think it does the opposite job at first, making the petition owner seem to have no confidence in his petition.
It does? That sounds like gaslighting users. Ironic in context of "Mandating truthful disclosure to the public and customers is very good".
The link only seems to open with an American IP (at least with my little tests).
For me doesn't open with a German IP but switching on a vpn, it opens with a US one.
Link appears to be broken.
Works for me, but you can also read the text of the order here: https://reason.com/volokh/2023/05/30/federal-judge-requires-...
(The linked blog post is where I discovered this, but I submitted the original source instead.)
Works for me too. Nice to see some judges have integrity.
In federal courts, when you file something under your account (often, attorneys have staff do the clicking to file), it is the equivalent of signing. And what you sign, you vouch for.
This standing order reiterates what any practitioner _should_ know. It a rule which states that you should follow the rules! But as we learned from the Air Avianca filing, at least one lawyer is missing something very basic about what it means to sign a filing in federal court.
``` All attorneys appearing before the Court must file on the docket a certificate attesting either that no portion of the filing was drafted by generative artificial intelligence (such as ChatGPT, Harvey.AI, or Google Bard) or that any language drafted by generative artificial intelligence was checked for accuracy, using print reporters or traditional legal databases, by a human being. These platforms are incredibly powerful and have many uses in the law: form divorces, discovery requests, suggested errors in documents, anticipated questions at oral argument. But legal briefing is not one of them. Here’s why. These platforms in their current states are prone to hallucinations and bias. On hallucinations, they make stuff up—even quotes and citations. Another issue is reliability or bias. While attorneys swear an oath to set aside their personal prejudices, biases, and beliefs to faithfully uphold the law and represent their clients, generative artificial intelligence is the product of programming devised by humans who did not have to swear such an oath. As such, these systems hold no allegiance to any client, the rule of law, or the laws and Constitution of the United States (or, as addressed above, the truth). Unbound by any sense of duty, honor, or justice, such programs act according to computer code rather than conviction, based on programming rather than principle. Any party believing a platform has the requisite accuracy and reliability for legal briefing may move for leave and explain why. Accordingly, the Court will strike any filing from an attorney who fails to file a certificate on the docket attesting that the attorney has read the Court’s judge-specific requirements and understands that he or she will be held responsible under Rule 11 for the contents of any filing that he or she signs and submits to the Court, regardless of whether generative artificial intelligence drafted any portion of that filing. ```
> While attorneys swear an oath to set aside their personal prejudices, biases, and beliefs to faithfully uphold the law and represent their clients, generative artificial intelligence is the product of programming devised by humans who did not have to swear such an oath.
While I like the idea, the reasoning here is kinda silly, even if some developer had sworn such an oath it is entirely unclear how that should affect a AI. We just don't have good techniques to prevent llms from hallucinating.
the hallucination issue was a separate clause. the court does not think that so swearing will fix hallucinations.
Not just the developer, but anyone who posted any sort of text in the training data. I assume that includes a pretty wide majority of English speakers. That makes the likelihood of actual agreement with the oath for all “participants” in the AI training low, to say the least.
That point about swearing an oath is a good one. It’s a reasonable position (assuming it’s the court’s position) that an AI (assuming this is an “AI”) being asked to provide testimony or a legal argument is doing so on behalf of all who trained it. Why would “my” training argue in “your” case? What if one doesn’t agree with the oath?
That's surprisingly well informed about tech for a judge.
One of hardest, time consuming, and most important parts of being a judge is understanding the issues presented before you. Be it generative AI or communal animal husbandry.
There are clearly gross examples of judges failing in this duty, but others take that duty extremely seriously and I've read a number of judicial opinions on technology-related cases that demonstrated remarkable, nuanced grokking of the matter at hand.
Lordy this is dumb. There's been exactly one recorded instance of a lawyer using ChatGPT in a brief and not checking it. This isn't a meaningful problem.
There has been one news article that has gained enough traction to get widespread notice. That’s not even remotely the same as “one recorded instance.”
The order, already posted on this thread, lists multiple instances where generative AI would be useful in the law, but also a number of issues that I haven’t seen discussed before.
> This isn't a meaningful problem.
...Yet. This isn't a meaningful problem, yet. SimpsonsMeme.jpg
Sarcasm aside, there are already cases (I think at least two) of this happening in the wild, why not nip it in the bud before some high profile event really catapults it into the public eye?
Would you rather it become a meaningful problem and overload the system before someone implements a simple check to ensure it doesn't?
In court, there is no meaningless problem or detail. Especially any detail pertaining to the structuring of legal arguments. And even more especially any details pertaining to the soundness of legal arguments and evidence. Legal experts obsessing over stuff like this is the only reason it's even vaguely possible to conduct a fair trial at all.
There's only been one big one which made the news. So far. There have almost certainly been more which have gone under the radar, or which weren't caught.
More than one. Here's two sources indicating at least four more lawyers out there using fake ChatGPT case citations.
https://reason.com/volokh/2023/05/27/a-partner-at-a-big-firm...
https://twitter.com/narrowlytaylord/status/16620971840770129...