6 ms·

Colombian judge used ChatGPT to make court decision

77 points by azefiel 4y ago

clbrmbr 4y ago
Could be interesting to use embeddings so you could get proper legal citations. This could be a great leveler, making it easier to self-represent…
- maxbond 4y ago
  Here's a video from Legal Eagle on the topic from yesterday: https://youtu.be/Tpq3hRt0pmw https://youtu.be/Tpq3hRt0pmw Tl;Dw it's a cold shower on this idea as it's implemented by the startup DoNotPay. There are a myriad of issues but the biggest one is that it isn't possible to foist the liability onto the customer and thus skirt the regulations around the unlicensed practice of law - if you run this service, you are either a law firm or you are a criminal enterprise. Maybe it could be made to work though, this particular startup doesn't seem to have the deep understanding of the law that would be necessary for the project to succeed. If one started from the assumption they were an AI-assisted budget law firm, rather than an Uber or AirBNB style "we're just gunnuh break the law and get away with it, until we can eventually lobby to change the law," perhaps it could work.
  - jkaplowitz 4y ago
    Not all countries comprehensively forbid the unlicensed practice of law - one prominent exception is England, where only a specific set of reserved legal activities require authorization: https://legalservicesboard.org.uk/enquiries/frequently-asked-questions/reserved-legal-activities https://legalservicesboard.org.uk/enquiries/frequently-asked... So a service like this would probably be legal there if they avoid the reserved activities.
jacooper 4y ago
TLDR: just to generate the legalese needed for the decision, not the decision it self.
- tyingq 4y ago
  >According to the court document, the legal questions entered into the AI tool included “Is an autistic minor exonerated from paying fees for their therapies?” and “Has the jurisprudence of the constitutional court made favorable decisions in similar cases?” It does say they were fact checked, but it also seems to leave room for questions. Like, for example, did ChatGPT choose the most appropriate answers/background, or just ones that fact-checked well enough? Might that matter later if the decision is rehashed or used as precedent?
  - donmcronald 4y ago
    > Like, for example, did ChatGPT choose the most appropriate answers/background, or just ones that fact-checked well enough? Yeah. Train it on a rulings that pre-date the civil rights movement and see how great it is.
    - mrweasel 4y ago
      On the other hand: Train it on rulings from the past 30 years, but remove all traces of gender, race, income, location and education. Then let the A.I. reveal a ton of bias in judges, prosecutors and juries.
- giancarlostoro 4y ago
  This is how I use chatgpt. “Write me an email explaining to stakeholders that we have x” and then I mention as much detail and whatever it spits out I tweak and go from there. I have been told I write amazing emails but my problem is I spend a lot of time thinking my email over way too much, so I like to see how condensed chatgpt gives me of a template so I can reuse it. I often take a five paragraph email and condense it back down to like three sentences before chagpt, now I can save myself the process.
  - hef19898 4y ago
    I would honestely have doubts giving an unvetted third party, openai through ChatGPT, access to any prorietery or business relevant information.
8note 4y ago
Oh interesting. I didn't realize chatgpt works in multiple languages
- jacooper 4y ago
  I mean it just studies relations between words, so I can be trained for any language as long as there is content for it.
  - cm2187 4y ago
    Though not sure it can distinguish colombian law vs any other spanish speaking country law.
- rapsey 4y ago
  Not only that it is probably the best translation engine ever made. It translates incredibly well. This is likely one of the industries that is going to be hit severely.
  - mistrial9 4y ago
    depending on the domain !! it is absolutely confident, right, wrong or in the middle.. actually "dangerous" if you think about a few contexts
  - drexlspivey 4y ago
    I tried using it to translate subtitles for a movie and the results were pretty bad. I found DeepL giving marginally better results but those were pretty bad as well. To the point that I uploaded the subtitles in opensubtitles.org and they were taken down because of very low quality.
  - fnordpiglet 4y ago
    It’s able to translate dead languages reasonably well, too. I asked it to translate various text into Pali and it did a good job, and did a reasonable job providing the text of various Buddhist sutta and decent translations into English.
cm2187 4y ago
I am not sure how it works in Colombia but in many countries, court judgements are lengthy documents that summarise all the aspects of a case before detailing the rationale for the decision. It feels awfully inefficient. Any tool that can help this process sounds like a good idea to me. It's no different to ChatGPT generating some dumb boiler plate code that you can adapt, correct and fine tune instead of having to write it from scratch.
- donmcronald 4y ago
  I think it's pretty scary. The judge is asking it to make judgment calls. > “Has the jurisprudence of the constitutional court made favorable decisions in similar cases?” Whether or not similar rulings are "favorable" could contain a lot of nuance. > It feels awfully inefficient. Good? I think it's important for a judge to have a comprehensive understanding of the current case and all relevant precedents. If they're going to use ChatGPT to inform their opinion, they might as well let it make the ruling. Imagine if the judge had another person that's not a lawyer read the relevant info and provide a summary / opinion. People would be appalled.
  - fnordpiglet 4y ago
    If it’s making reviewed boilerplate how is it making a judgement call? Having human beings write copious tedious boilerplate isn’t good. It takes the judges mind and applies to tedium rather than using judgement. Further under tedium humans make mistakes.
    - donmcronald 4y ago
      > “Has the jurisprudence of the constitutional court made favorable decisions in similar cases?” That's a judgment call because "favorable" is subjective. Someone who gets sent to jail for 1 year might tell you the prosecutor got a favorable ruling while the prosecutor considers it unfavorable because they wanted the person sentenced to 5 years.
      - 4y ago
whatgoodisaroad 4y ago
There's a way of looking at this as pretty much normal. I think we have yet to really face the fact that systems of common-law are ad-hoc LLMs (in contrast to statutory-law, which is more similar to standard programming). In particular, common-law uses an enormous "model" of case precedents that gets incrementally updated by human curators with minimal influence over it.
tropicaljacket 4y ago
> Although the Colombian court filing indicates that ... its responses were fact-checked The problem is that the fact-checkers themselves can make mistakes and chatGPT is very good at making text that looks correct, even to experts. See the Meta StackOverflow post on banning ChatGPT answers for example
- geraldyo 4y ago
  At that point we're back to the fact that humans make mistakes, so the AI is irrelevant. Had they not used AI, they could still make mistakes that look correct, even to experts
  - cscurmudgeon 4y ago
    It is all a matter of degree and not a binary situation.
theknocker 4y ago
[dead]
modzu 4y ago
cf: the plain language movement, and the book Language of the Law (1963)

77 points by azefiel 4y ago

Could be interesting to use embeddings so you could get proper legal citations. This could be a great leveler, making it easier to self-represent…

maxbond 4y ago
Here's a video from Legal Eagle on the topic from yesterday: https://youtu.be/Tpq3hRt0pmw https://youtu.be/Tpq3hRt0pmw Tl;Dw it's a cold shower on this idea as it's implemented by the startup DoNotPay. There are a myriad of issues but the biggest one is that it isn't possible to foist the liability onto the customer and thus skirt the regulations around the unlicensed practice of law - if you run this service, you are either a law firm or you are a criminal enterprise. Maybe it could be made to work though, this particular startup doesn't seem to have the deep understanding of the law that would be necessary for the project to succeed. If one started from the assumption they were an AI-assisted budget law firm, rather than an Uber or AirBNB style "we're just gunnuh break the law and get away with it, until we can eventually lobby to change the law," perhaps it could work.
- jkaplowitz 4y ago
  Not all countries comprehensively forbid the unlicensed practice of law - one prominent exception is England, where only a specific set of reserved legal activities require authorization: https://legalservicesboard.org.uk/enquiries/frequently-asked-questions/reserved-legal-activities https://legalservicesboard.org.uk/enquiries/frequently-asked... So a service like this would probably be legal there if they avoid the reserved activities.

jacooper 4y ago

TLDR: just to generate the legalese needed for the decision, not the decision it self.

tyingq 4y ago
>According to the court document, the legal questions entered into the AI tool included “Is an autistic minor exonerated from paying fees for their therapies?” and “Has the jurisprudence of the constitutional court made favorable decisions in similar cases?” It does say they were fact checked, but it also seems to leave room for questions. Like, for example, did ChatGPT choose the most appropriate answers/background, or just ones that fact-checked well enough? Might that matter later if the decision is rehashed or used as precedent?
- donmcronald 4y ago
  > Like, for example, did ChatGPT choose the most appropriate answers/background, or just ones that fact-checked well enough? Yeah. Train it on a rulings that pre-date the civil rights movement and see how great it is.
  - mrweasel 4y ago
    On the other hand: Train it on rulings from the past 30 years, but remove all traces of gender, race, income, location and education. Then let the A.I. reveal a ton of bias in judges, prosecutors and juries.
giancarlostoro 4y ago
This is how I use chatgpt. “Write me an email explaining to stakeholders that we have x” and then I mention as much detail and whatever it spits out I tweak and go from there. I have been told I write amazing emails but my problem is I spend a lot of time thinking my email over way too much, so I like to see how condensed chatgpt gives me of a template so I can reuse it. I often take a five paragraph email and condense it back down to like three sentences before chagpt, now I can save myself the process.
- hef19898 4y ago
  I would honestely have doubts giving an unvetted third party, openai through ChatGPT, access to any prorietery or business relevant information.

8note 4y ago

Oh interesting. I didn't realize chatgpt works in multiple languages

jacooper 4y ago
I mean it just studies relations between words, so I can be trained for any language as long as there is content for it.
- cm2187 4y ago
  Though not sure it can distinguish colombian law vs any other spanish speaking country law.
rapsey 4y ago
Not only that it is probably the best translation engine ever made. It translates incredibly well. This is likely one of the industries that is going to be hit severely.
- mistrial9 4y ago
  depending on the domain !! it is absolutely confident, right, wrong or in the middle.. actually "dangerous" if you think about a few contexts
- drexlspivey 4y ago
  I tried using it to translate subtitles for a movie and the results were pretty bad. I found DeepL giving marginally better results but those were pretty bad as well. To the point that I uploaded the subtitles in opensubtitles.org and they were taken down because of very low quality.
- fnordpiglet 4y ago
  It’s able to translate dead languages reasonably well, too. I asked it to translate various text into Pali and it did a good job, and did a reasonable job providing the text of various Buddhist sutta and decent translations into English.

cm2187 4y ago

I am not sure how it works in Colombia but in many countries, court judgements are lengthy documents that summarise all the aspects of a case before detailing the rationale for the decision. It feels awfully inefficient. Any tool that can help this process sounds like a good idea to me. It's no different to ChatGPT generating some dumb boiler plate code that you can adapt, correct and fine tune instead of having to write it from scratch.

donmcronald 4y ago
I think it's pretty scary. The judge is asking it to make judgment calls. > “Has the jurisprudence of the constitutional court made favorable decisions in similar cases?” Whether or not similar rulings are "favorable" could contain a lot of nuance. > It feels awfully inefficient. Good? I think it's important for a judge to have a comprehensive understanding of the current case and all relevant precedents. If they're going to use ChatGPT to inform their opinion, they might as well let it make the ruling. Imagine if the judge had another person that's not a lawyer read the relevant info and provide a summary / opinion. People would be appalled.
- fnordpiglet 4y ago
  If it’s making reviewed boilerplate how is it making a judgement call? Having human beings write copious tedious boilerplate isn’t good. It takes the judges mind and applies to tedium rather than using judgement. Further under tedium humans make mistakes.
  - donmcronald 4y ago
    > “Has the jurisprudence of the constitutional court made favorable decisions in similar cases?” That's a judgment call because "favorable" is subjective. Someone who gets sent to jail for 1 year might tell you the prosecutor got a favorable ruling while the prosecutor considers it unfavorable because they wanted the person sentenced to 5 years.
    - 4y ago

whatgoodisaroad 4y ago

There's a way of looking at this as pretty much normal. I think we have yet to really face the fact that systems of common-law are ad-hoc LLMs (in contrast to statutory-law, which is more similar to standard programming). In particular, common-law uses an enormous "model" of case precedents that gets incrementally updated by human curators with minimal influence over it.

tropicaljacket 4y ago

> Although the Colombian court filing indicates that ... its responses were fact-checked The problem is that the fact-checkers themselves can make mistakes and chatGPT is very good at making text that looks correct, even to experts. See the Meta StackOverflow post on banning ChatGPT answers for example

geraldyo 4y ago
At that point we're back to the fact that humans make mistakes, so the AI is irrelevant. Had they not used AI, they could still make mistakes that look correct, even to experts
- cscurmudgeon 4y ago
  It is all a matter of degree and not a binary situation.

theknocker 4y ago

[dead]

modzu 4y ago

cf: the plain language movement, and the book Language of the Law (1963)