LSAT 132, Section 3, RC Passage 4, Automated Legal Reasoning

RC Passage 4 Explanation

PASSAGE TEXT: Computers have long been utilized in the sphere of law in the form …

Paragraph Summaries

1. Computers are already useful in legal work, but AI legal reasoning systems have not lived up to early hopes. They struggle because legal rules are hard to interpret and apply.
2. Early systems treated law as a set of rules to apply to facts. But legal terms often require interpretation, and computers lack the broad real-world knowledge needed to handle new situations.
3. Case-based systems try to reason from precedents instead of just rules. But they still depend on preset similarity criteria, so they can’t independently figure out which case similarities actually matter.

Analysis

This is a very timely passage. You’d assume it was written in the 2020s or at least the late 2010s, but it’s actually from 2006!

As a side note, AI legal reasoning has improved since then, naturally. Many (if not most) law firms use legal AI tools now, and the field has shifted from “don’t use it” to “learn how to use it efficiently”. But it’s still far from perfect. As the passage points out, it can be notoriously bad at legal reasoning.

The reason it’s bad is because it’s hard to know what the law means when you have messy facts.

The rule-based systems fail for the obvious legal reason. A statute doesn’t always tell you whether a mobile home is a house or a vehicle. Sometimes the words are deliberately loose so the law can stretch to future cases. That’s great for human lawyers and judges; terrible for a computer that wants clean categories.

The case-based systems sound like the fix, because lawyers do reason by analogy to precedent. What this means is that if a similar thing happened in a previous case, you can use the ruling from that past case to determine what should happen in this case.

But the passage is pretty unimpressed. The machine can compare cases only according to factors the designer already built in. So the hard part hasn’t gone away. It’s just been moved back a step: who decides what makes two cases relevantly similar?

That’s the phrase that matters (relevantly similar). In law, similarity isn’t just “these two cases share five facts”. One shared fact might matter a lot while ten other facts are noise. The current systems can’t discover that for themselves.

The author isn’t anti-computer, that’s silly. Word processing, research systems, etc. are fine. The skepticism is about computers independently giving expert substantive legal advice.