ChatGPT lies. It makes up facts, sources, and nonsense arguments.
Lying to a judge is generally not a good idea. You can go from traffic ticket to contempt of court real fast if you start lying in court.
ChatGPT also assumes you're speaking the truth. If you ask it about a topic and say "that's wrong, the actual facts are..." then it'll change argumentation to support your position. You probably don't want your legal representation to become your prosecutor when they use the right type of phrasing.
> Lying to a judge is generally not a good idea. You can go from traffic ticket to contempt of court real fast if you start lying in court.
Then I suppose that's just the risk the defendant takes, isn't it? Let people use ChatGPT, if the rope they're given ends up hanging enough people, that'll be the end of that, won't it?
Also, everyone is ignoring the possibility that this same person could've had ChatGPT generate a script (general outline of arguments, "what do I say if asked this" type of stuff), memorized it, and used that to guide his self-defense. Fundamentally, no difference. No one would've known, and no one would've objected.
To me, this move is less "oh we need to protect people from getting bad legal advice from a robot" and "we're not even gonna let this thing be used a single time in court to keep our job from being automated."
> Fundamentally, no difference. No one would've known, and no one would've objected.
I'm not a legal professional but it seems obvious to me that there is a fundamental difference, namely the one you describe just before that sentence. The whole legal system is built around and under the assumption that all kinds of people want to trick it, and judges tend to be allergic to this kind of reasoning. Memorizing legal arguments and getting live legal advise from earphones in your glasses are not the same thing. Besides,even lawyers are advised not to defend themselves in court, and it would be generally very bad advise for anyone to do so.
> Memorizing legal arguments and getting live legal advise from earphones in your glasses are not the same thing.
Only in the strictest sense. Let's say the person memorizing ChatGPT's directions handles their case in the exact same manner as if it was being relayed to them live (i.e., the set of statements/questions from the judge lined up perfectly with what ChatGPT presented in its script). What then? Same outcome, different delivery method. We're kind of splitting hairs with the "live legal advise" thing. The defendant could bring a pile of law books with him and consult those without anyone blinking an eye. The objection seems to boil down to "well OK, if you want to represent yourself you better not consult an intelligent system to help you form your defense." Why not though? Seems more about job protection than anything else.
> Besides,even lawyers are advised not to defend themselves in court, and it would be generally very bad advise for anyone to do so.
And I say: let people discover the downside of using ChatGPT for defense if it's so inept. Bad outcomes are the best way to prevent widespread usage, not pre-emptive bans in the interest of keeping people from shooting themselves in the foot.
>Lying to a judge is generally not a good idea. You can go from traffic ticket to contempt of court real fast if you start lying in court.
If you knowingly lie because ChatGPT told you to that's on you. If you said something that you believed was true because ChatGPT said it to you then that's not perjury, it's just being wrong.
> You probably don't want your legal representation to become your prosecutor when they use the right type of phrasing.
When the worst case scenario is having to pay the parking fine, it might be worth taking this risk to avoid paying a lawyer.
Honest tangent that I'm dying to find an answer for. Is the assumption of truthful prompts something openAI decided should be there, or is the assumption something very deeply baked into this sort of language model? Could they make an argumentative, opinionated, arrogant asshole version of chatGPT if they simply let it off its leash?
Yeah, but notebooks are allowed in court and you could spill water on your notebook and confuse an 8 for a 0 and then read it out to court and it would be a lie.
Lying to judges is bad, so notebooks should be banned. Likewise, anything typed should be banned because we could have hit the wrong key, causing you to lie to the judge.
Unpaid traffic tickets are probably many people's gateway to their first arrest warrant. Then once in the system it is hard to escape. It is a real thing.
Then once this process starts you automatically get
* Suspended driver’s license
* Ineligible to renew your driver’s license
* Ineligible to register your vehicle
* Vehicle being towed and impounded
* Increased insurance premiums
Traffic tickets and parking tickets are two different things though. Unpaid parking ticket is unlikely to get you arrested anywhere, but an unpaid speeding ticket will eventually lead to some pretty unpleasant consequences which yes, might include arrest(to bring you in front of the judge and explain yourself).
ChatGPT: "Your honor, that couldn't have been me, as I drive a red Mustang, not a blue minivan, and was in Nepal climbing a mountain at the time.."
Defendant: "Your honor, that couldn't have been my, as I drive a red Mustang, not a blue minivan, and was in Nepal climbing a mountain at the time."
Prosecutor: "Um, this photo clearly shows your face in this blue minivan, and there's no evidence you've been to Nepal."
Judge: "I'm holding you in contempt of court, and sentence you to 7 days in jail for perjury."
8<----
I don't think the argument is that AI is never allowed to represent someone in court; just that before it happens, a sufficient amount of vetting must be done. At a bare minimum, the legal AI needs to know not to lead the defendant to perjure themselves.
I think the way forward might be an arbitration case, where they pay an actual legal expert in the right position to make binding decisions outside of the context of the normal law system.
In voluntary arbitration, you can bend the rules a lot more than in an actual court case.
For instance, as I discovered recently going through this process myself - here in UK when applying for British citizenship you have to disclose any court orders against you. Now here's a thing - if you were given a ticket for speeding, accepted it and paid it then that's it, no harm done. If it's less than 3 tickets in the last 5 years then you don't even need to list it on the application form.
However, if you went to court to contest it and lost, then you now have a court order against you - and that's an automatic 3 year ban on British citizenship applications, and even after that you always have to list it as a thing that happened and it can be used to argue you are of "bad character" and be used to deny you the citizenship.
So yes, failing to get rid of a traffic ticket(in the court of law) can absolutely ruin your life.
How is this known upfront? I've lived for over a decade in this country without knowing this, until I actually applied for my citizenship last year. I'm just lucky I never went to court to contest a speeding ticket(because I never got any) or I could have screwed myself over without even knowing.
Also define "small group" - nearly 200k people apply for British citizenship annually, and I bet most of them have no idea contesting a traffic ticket can cost them a chance at becoming citizens.
I used to drive a cab and people would tell me why they were in the cab quite often.
My favorite was this guy who got a ticket on a bicycle for not having a headlight, moved out of state and ten years later had his car impounded for driving without a license because they apparently suspended it for getting (or, more correctly, not paying) a ticket he got on a bicycle.
I’m guessing he never tried changing his license to the new state because Arizona licenses are good until you’re 65.
As the AI gets better people will trust it with more and more kinds of cases and cases with more increased complexity. If people want to pay for a real licensed lawyer they are still able to do so.
AI is just informed search, a dwarf sitting on shoulders of human knowledge. There were medical "expert systems" in 2000s, yet we still have doctors.
In my understanding, in most cases AI will be a glorified assistant, not an authoritative decision-maker. Otherwise in collides head-on with barriers and semis. I won't trust such system even with a parking ticket, yet alone my life.
We're just at the top of a hype-cycle now. AI can do new things, but not as well as we dream or hope.
Any kind of assistant can make mistakes. But a human assistant can be made to show their work and explain their reasoning to check their output. If ChatGPT says "this thing is totally legal" or "don't worry about that rash", how am I to validate it's "reasoning"? How do I know where it's drawing it's inference from?
> How does failing to get rid of a traffic ticket ruin someone's life?
The downside of contesting a traffic ticket is not “failing to get rid of the ticket”. The ticket amount amounts to a no contest plea bargain offer, not the maximum penalty for the offense, not to mentiom the potential additional penalties for violating court rules.
