Punjab and Haryana high court uses ChatGPT to decide bail plea
timesofindia.indiatimes.com
timesofindia.indiatimes.com
Interesting thing to do. Proves their court system (or at least one presiding judge) isn't afraid of AI in the courtroom.
Do Not Pay tried to run an experiment with an AI lawyer, but people got spooked and it didn't happen.
Different strokes - I like the India approach better so far.
Plus I'm pretty sure this violates "Open"AI's terms of service.
Edit: yep. https://openai.com/policies/usage-policies
>Disallowed usage of our models
>We don’t allow the use of our models for the following:
>Engaging in the unauthorized practice of law, or offering tailored legal advice without a qualified person reviewing the information
>OpenAI’s models are not fine-tuned to provide legal advice. You should not rely on our models as a sole source of legal advice.
>High risk government decision-making, including:
>Law enforcement and criminal justice
>Migration and asylum
Edit: better articles that detail the question posed to ChatGPT and the answer:
https://www.indianarrative.com/india-news/in-a-first-high-co...
https://sports.yahoo.com/first-indian-court-turns-chatgpt-07...
From the articles you quoted, the question asked to ChatGPT was: "What is the jurisprudence on bail when the assailants assaulted with cruelty?"
My reading of the terms and the situation:
> >Engaging in the unauthorized practice of law, or offering tailored legal advice without a qualified person reviewing the information
This was authorized practice of law, not tailored legal advice (the question was generic), and there was a qualified person reviewing the information.
> >OpenAI’s models are not fine-tuned to provide legal advice. You should not rely on our models as a sole source of legal advice.
This was not the sole source of legal advice.
> >High risk government decision-making, including: [...]
That one is tricky. Is the judicial system considered "high risk government decision-making"? Not too sure.
In my opinion, it was still a bad decision to ask a jurisprudence question to an AI instead of asking to a paralegal, whose job it is to research that, because there is no guarantee the response was accurate. On the other hand, since it was accepted by a judge, it has been human-reviewed and can probably be considered accurate.
My issue is that the judge thought it was a good idea to consult ChatGPT in the first place. There are other ways to get better answers that don't rely on a general LLM, like making a search index that includes every legal document for the case and consulting that. But using something that seems to be somewhat persuasive in its delivery is a cause for concern. It's like (weird example, sorry) as if the judge watches anime and made five of his favourite characters speak different verdicts and then chose the character he felt he liked the best. Almost reifying the source of the information and then considering human-ness a viable metric for judging how trustworthy a source is. Really reminds me of people that use LLMs as a "therapist" or for "roleplay", it's very disgusting.
> high risk government decision-making
IMO what that means is "high risk decision-making, like law enforcement and criminal justice, which is the kind a government engages in" and not that only some high risk decision-making pertaining to law enforcement (or criminal justice) by a government is prohibited.