Hacker News Reader

> A good lawyer is not just knowledgeable of law, but of people. He needs to get 12 people in front of him to agree with him.

While I agree with you that AI powered lawyer can have many downsides and drawbacks here you are building on a falsehood.

There is this illusions that computers are "book smart" but not "street smart". It is very deeply ingrained into our culture. Just think about Data from Star Trek. Very analytical, but struggles to understand emotions. Therefore also struggles to manipulate people's emotions, aka being emotionally persuasive.

This is a sci-fi trope, but is false as it comes to these generative models. (or can be false depending on training data) If you train the system with mock juries it can learn how to twist them around its finger. Text written by such a model can be as emotionally persuasive as one written by the best lawyer.

In fact in reality, as the tech is today, it would be easier to reliably implement the "reading reactions, and understanding what each of those 12 people needs to hear to have them act like he wants" part than the "follows flawless logic, legal technicalities to a T" part.

It is of course still a very perilous path to make an AI-powered lawyer. It is just important to not base our reasoning on sci-fi tropes

> it would be easier to reliably implement the "reading reactions, and understanding what each of those 12 people needs to hear to have them act like he wants" part

Come to think about it. This part can be valuable, even as a startup. You get mock juries, you act out legal cases in front of them. You collect every single thing the actor lawyers say, and how they say it, and you also collect full transcript of their deliberations.

You model the data, and then you sell the services to lawyers. They can practice in front of your model jury initialised with what can be known about the real jurors and can see traces of their deliberations. Not just one run of the model, but many runs if they want to.

An overwhelming majority of Federal cases don't involve a jury and instead consist of:

1. A scaling army of prosecutors and "professional witnesses" which are compensated/promoted based on conviction rate.

2. A single "freely" provided and extremely overworked defense attorney with three times the caseload of the prosecutor and where every bit of help they give out takes from the time of the other defendants they "defend."

3. As a result of 1&2 you are pressured to sign a plea and threatened with triple and quadruple punishments for the same activities. All while being stuck in a cell (why else would you need a free attorney)? And we are not talking the difference of 3 months and a year. Instead we are talking about the difference of 5 years and 20 years.

4. The result is an overwhelming number of factually innocent people in prison for crimes they never committed. From people who sat down and said, what happens if it goes wrong vs what happens when I do 5 years of adult daycare?

So yes, any kind of AI is likely to improve the situation as the situation is currently completely unacceptable to anyone who has seen a family member go through it.

If poor people can only get the bot, that means the Constitution was changed. I don’t see how that could be the tech’s fault. It’s more likely that the AI would free up public defenders from things like having to answer their clients’ questions, as they could just ask the AI, allowing them to be more effective.

> If poor people can only get the bot, that means the Constitution was changed.

The Constitution is already pushed beyond all reason for poor people anyway. You might have a "right" to a public defender, but good luck getting them to spend much more than 10 minutes on your case. (https://www.mprnews.org/story/2010/11/29/public-defenders) They're so overworked that it can be impossible for them to spend more than a tiny fraction of the time they actually need to do their job (https://www.kansascity.com/news/local/crime/article237036919...).

In the same way you have a "right" to speedy trial, but unless you're rich you can spend years (or even a decade) waiting behind bars (https://www.thecity.nyc/2022/8/17/23310771/why-some-spend-ye...).

It seems like we've already decided that if you can't pay the court fees you aren't entitled to access to our "justice" system (https://digitalcommons.law.seattleu.edu/sulr/vol44/iss3/11/)

I don't think it's too far fetched to imagine the courts telling poor people they have to settle for an AI to represent them unless they can fork over the money to hire one themselves.

If an AI is just as capable (if not more capable) than a real lawyer, we'll be seeing rich people and corporations using them instead of humans. Until we see that, I'll assume that if poor people are forced (by law or finances) to use them they'll be getting the short end of the stick.

> It’s more likely that the AI would free up public defendants from things like having to answer their clients’ questions, as they could just ask the AI, allowing them to be more effective

Public defenders would be more effective if they didn't have to talk to their clients?

I read it more as, clients have a mix of questions. Clients don’t go through the system everyday and likely have some general worries and procedural questions as a result. Clients also have a case and need to coordinate with their lawyer. AI could enable more of the latter to happen within the limited time the lawyer has for each client.

Seems like that problem could be fixed by giving clients a brochure and a video to watch that explains everything.

Sorry, I can’t ever write clearly on the phone, this site is very mobile-unfriendly. I meant that defendants would be able to talk to the AI about their case to learn as much as they want. A public defender can’t commit that much time to each of their cases.

The problem with this is "how do you assure the AI isn't bold face lying to the defendant".

For a natural text generation/chat system like ChatGPT, at the moment this is effectively impossible and frankly I don't see it being sufficiently fixed for something as complex as teaching law to a non-law-student any time in the near future.

I think these discussions always seem focused on the ideal state of the "AI is perfect and gives correct answers" and not the current state where the "AI constructs sentences that are structured correctly, but are filled with inaccuracies and illogical nonsense."

If we want to steelman AI why shouldn’t we also steelman the court system? Assuming a perfect AI is about the same as assuming a well functioning court system. Otherwise we should be assuming a flawed, potentially racist, confidently incorrect AI vs a flawed, excessively slow, definitely racist court system.

I don't see how it's the same.

So you make this as understood as possible and the lawyer can fill in the gaps and clarify.

It's not like a "jailhouse lawyer" will do any better. Though it might cost something you didn't expect, such as a new witness against you or a shower pal.