Hacker News Reader

Comment by deckiedan | Hacker News Reader

My grandmother passed away a few years ago, and some of her express wishes were not correctly stipulated in her will - so when things were being sorted out it got messy. Thankfully all the family were really chill and we could sort it out amicably - but I had the idea at the time:

If legal documents have to be parsed and understood almost like source code, could we also have a testing framework? So I could stipulate things I want to happen as tests (user stories) and then run it against the contract / legal document and confirm that it does actually allow what I want to happen?

You'd be restricted to a subset of legal language - and possibly you'd have to actually write documents in some kind of meta-language (probaby lisp...) and then turn it into English or whatever as an output format in the end...

But would this work?

View on HN

It wouldn't work because specifying requirements isn't the difficult part of law or really a problem now.

The problem is reconciling those requirements with reality. If contracts become functions, we don't make anything more neutral or unbiased. Same as today, the person who supplies the arguments to the function gets to control the output.

For example, earlier this year there was an interesting lawsuit [1]. An insurance contract excluded "acts of war." There was no dispute that that was a requirement.

The dispute was on whether or not a given action was an "act of war." Mondelez claimed that the NotPetya cyber attack was an act of war as the US government said it was the work of the Russian military. Zurich disagreed.

https://www.zdnet.com/article/notpetya-an-act-of-war-cyber-i...

Maybe it’s different for the contract law, but I definitely think that legislation could benefit from more formal language.

We recently had a disagreement between an NPS officer and another visitor about what constitutes an occupied campsite. Apparently having paid for it, having a receipt on hand and another receipt with your name on the board by the entrance with the site number next to it doesn’t mean it’s occupied, you also have to leave a personal item. When you look at the rules it does mention leaving a personal item, but it’s in the “How to pay” section indicating to me that it’s a precaution to prevent two people trying to pay for the same site at the same time. Alas, the ranger disagreed and we almost lost our site and it turned into a whole ordeal all because the author of the rules didn’t spell it out and it was up to everyone’s interpretation.

Writing rules in the form of code has the chance of taking out a lot of that interpretation. Like you said, people can still disagree about what’s “an act of war”, but even that can be more precisely defined in a way that agrees with the common sense in the majority of the cases while still being unambiguous in the exceptional ones.

Sounds like the campsite could improve their signage, but as a general principle, I very much disagree.

There is a fundamental tradeoff between specificity and generality, where the more specific a law is, the longer and more complex it must become to accommodate the nearly infinite number of situations that can occur.

Take a look at the bill of rights. The thing is about as vague as the English language allows and is less than 500 words.

When is a search "unreasonable?" What constitutes "due process?" Where is the line between a "peaceful" and non-peaceful "assembly?" How long of a wait before a trial is no longer "speedy?" When is bail "excessive?"

Is it even possible to enumerate all of the possible situations and amounts when bail is considered excessive? There are countless crimes and countless mitigating circumstances. How could anyone foresee all the possible situations that can ever happen?

Despite these questions, most Americans have some idea of what the amendments mean and they can be easily taught to high school students in a US history or civics class.

Compare with a highly specific law, the Affordable Care Act (not picking it, just an example), which is 350k words long. At most, people are likely to have read a summary of a few specific points, but the law is basically only knowable to career lawyers with significant domain knowledge.

So even if it is possible to be very specific in laws, doing so basically makes the laws so immense and complex that they become unknowable. And since laws govern human behavior, humans need to be able to know the laws.

You basically just run into Bonini's paradox. A set of laws that was sufficiently specific to have absolutely no ambiguity would be as complex as the universe itself. https://en.wikipedia.org/wiki/Bonini%27s_paradox

The Bonini’s paradox makes perfect sense, but I don’t think it applies to this case because the law is not trying to model anything, it stands on its own. It’s the same difference as trying to model the law and predict what is and what isn’t lawful based on some circumstances vs declaring that the model IS the law and it determines what’s legal or not.

Let’s take the tax code as an example. I’d argue that it already is a form of a program (maybe that’s why they call it code), but instead of something that I can run on my computer and explore and play with inputs to achieve a more beneficial result for myself, I have to instead pay a tax professional to do it. I think having the tax code executable would actually make it more accessible to people that are governed by it, not the other way around.

For the program you want regarding the tax code exists now. TurboTax, HR Block, free file etc...plug in numbers, see different totals is exactly how it works.

Of course there is still a ton ambiguity in answering the questions on the form. That has nothing to do with the format of the laws and everything to do with the inherit nature of translating reality into simple booleans and integers. If you wanted to remove all ambiguity, the form would need to be complex as reality.

Epistemology epigrams:

in theory, theory is the same as practice. In practice, they are different.

You cannot pass from the informal to the formal by formal means.

No. Because the definition of every word in your meta-language is subject to interpretation by the involved parties. There will be disagreements and then an independent adjudicator will have to provide its own interpretation (hopefully taking the parties view into account) to resolve the issue. Just like we have now.

No. The law is not a virtual machine and legal documents are not programming.

Believing so is a common misconception amongst engineers, but depending on it as such is likely to lead to disappointment, frustration, anger, needless bickering, extended conflict, and vexatiously long, hard to read, and mostly unenforceable contracts.

> No. The law is not a virtual machine and legal documents are not programming.

This is true, but I'd think the opposite point is more important.

Much of the grunge work I see the lawyers around me do really is about parsing out and semantically evaluating some clear-cut-but-complicated bit of text. As a result lawyers genuinely value producing (and succeed at) producing logically coherent, unambiguous text.

By contrast, most programmers are as sloppy as their compiler will allow them to be. This is bad enough in code, but it is much worse when we talk to each-other in English. Not only documentation, but internal design discussions etc, are usually some combination vague, meaningless and wrong.

There are companies already doing this for specific kinds of contracts, e.g. DealSumm [1] (https://www.dealsumm.com/). You give them a large corpus of lease agreements (e.g. for an apartment building you've just bought) and they give you a nice structured representation of the obligations, total revenue, all the upcoming deadlines, etc.

Due to the requirement of legal language to be consistent and explicit (e.g. defining most terms in an annex), it's actually a very easy workload for NLP techniques, and the software is in production and widely used. Don't know if there's a version specifically trained for wills, but it would probably not be very hard to put together.

Take a look at this: Formal Methods and the Law https://youtu.be/EshxZVMURt4

What a great idea. I feel like there are a dozen variations in this that might be effective.

How about running a bunch of documents through ML and teach it what contentious contracts look like?

Reply on news.ycombinator.com