I’m 1000% with you

Does this shows us how inhuman even the hardware interface to computers is?

I would say it's the opposite. The fact that we take these multi-dimensional ideas, and reduce them to expressions in one dimension, causes a lot of unnecessary cognitive load to extract information from the text. We like when that goes away.

Oh yes, I am saying the encoding is linear, not the underlying idea. I think we agree on that.

To elaborate on what @Jason Morris says, when you flatten a multi-dimensional idea, some relation that has parts close together necessarily pushes them apart, so along that dimension, whatever it is, you'll have to jump around.

The idea is to provide answers to legal questions in a way that answers "what are the relevant inputs", "why are those inputs relevant", "what is the answer in my fact scenario", "why is that the answer", "how could I get the answer to be different", "how else could I get the same answer", and more, all in an interactive interface.

Leaf nodes would be inputs, and as you change the inputs, the colours in the graph change to indicate the answer, and the reasons. You could also theoretically override intermediate nodes for exploratory purposes, and it would show that the inputs to that node were being ignored, somehow.

Some things I'm curious about: how to provide easy navigation of this, limiting what is displayed to what the user is interested in, interactively, without requiring the entire screen.

whether it works with non-boolean conclusions.

how to represent "irrelevant" inputs, etc.

feedback welcome. in particular, if anyone has actually seen anything like this in the wild, open source, let me know.

Why does it go right to left?

Because there is less information on the left, so it seemed like an easier place to "start". But I could be wrong. More intuitive the other way around?

Also, arguably the goal is the information that the user is most interested in.

I'm still having trouble following this in spite of the explanation. Perhaps the use case is too far away from what I usually think about. I wonder if it would help you get better feedback to explain what the concrete artifact is before you describe this particular visualization of the artifact.

Excellent point. It's a visualization of the explanation for why an automated legal reasoning tool reached a conclusion that a person was eligible for a benefit.

What does OAS stand for?

I think I understand the diagram. OAS is something like "Old Age Support" which will be some kind of service from the government with eligibility criteria. This eligibility criteria is explained in pages and pages of legislation. You can become eligible in different ways i.e. under section 3(2) (you have less than $X savings) etc

Is that right?

An IDE with "type checking" for laws as pure functions + FSM + separating objective input and subjective interpretation weights could be amazing...

...see all possible outcomes and why, clearly, before signing the paper, + with probability distributions if given in input. Backtrace and see what would need to change to fix it. See dependencies with existing legal framework, and outcome changes/interplay/affected parties when introducing new law/statement. Analysis connected to prejudicing cases.

I don't think pure functions are a good idea for modelling laws, because laws aren't written that way. But otherwise that's the basic idea. It also improves the drafting.

How would it influence drafting? Are you imagining automatic generation of legalese from such diagrams?!

No, the encoding that is used to generate the diagram is a model of the legislative text. The task of creating that model often reveals inadequacies in the legal text. The code can also be used to do things like search for loopholes in the abstract based on the model. So not these diagrams in particular, but it's benefit of the larger ecosystem of digitized rules.

Super interested in this space. One thought on the relevance question. There are a few ways for a rule/chain of inferences to be relevant/irrelevant. Suppose we look at the whole support tree for an Eligible conclusion. If a rule never shows up, it's totally irrelevant.

If a rule shows up in all justification paths, it's totally relevant: it's necessary, there's no way to be Eligible without it.

Between these extremes we have all manner of potential conditional relevancies. Say the case for those over 65 and the case for those under 65. The requirements could be totally different for those two groups, likely there will be some overlap though. Whatever overlaps is relevant regardless of age.

For me, this line of thinking resonates with two notions Bayesian Networks and query planning.

I think I would make it left-to-right or top-to-bottom. Top-to-bottom means it could be put side by side with the legislation (a bit like godbolt.org - which is, in a way, doing something similar).

I think I understand why you prefer this way - because the user is going to come to it primarily interested in "Am I eligible?", and can follow up with "but why? but why?" - "because you are eligible if you meet any of these three subsections, and you meet section 3(2), because you meet 3(2)a...".

For me though, I am perfectly comfortable with "why? why?" as going backwards against the normal direction of "calculation".

If it's possible, I think natural language explanations (i.e. expanding the "all", "any" could help.

This is how a helpful lawyer would explain it to somebody. They would point their way through the text, with a commentary like I put above "You need to meet one of these criteria, but you don't meet any".