In part 1,¹ we discussed LawNet AI, which features AI-generated summaries of cases not already headnoted (i.e. summarised) by Justice Law Clerks or legal editors.

Below, we discuss how headnotes can be misused.

--

Kenta Kusano's view is that AI-generated summaries are merely meant to help lawyers determine which cases lawyers should read in their entirety. Lawyers should continue to read entire cases, and no lawyer should rely on an AI-generated summaries to give advice or make a submission to Court.²

On this, we are in complete agreement.³

Headnotes - the low-tech predecessors to AI-generated summaries - often do not capture the Court's full legal reasoning,⁴ and can even be wrong.⁵ And that's before we consider whether their overuse risks impeding the development of young lawyers.⁶

We can probably expect the same from AI-generated summaries.

--

However, that's not to say that AI-generated summaries are useless.

A heuristic technique is a pragmatic problem-solving method that is not fully optimised, but is nevertheless "good enough".⁷

The use of headnotes and AI-generated summaries is a heuristic technique that can contribute to solving a legal research problem. I can see a potential use case where:
a) the lawyer doing legal research uses keyword Boolean searches to generate a list of potentially helpful cases;
b) for each case, the lawyer looks at the AI-generated summary to see if the case appears relevant;
c) if it looks relevant, the lawyer proceeds to read the case; and
d) if it look irrelevant, the lawyer turns to the next case.

Here, the use of AI-generated summaries cannot solve the problem by itself, but contributes towards finding the solution.

There is, however, a catch.

How confident is the lawyer that - based on its AI-generated summary - a case is truly irrelevant?

Let's assume that the vast majority of the AI-generated summaries are accurate.

Even so, how does the lawyer know whether the AI-generated summary for that particular case - which they are relying on to decide not to read that case - is accurate?⁸

One way to find out is to skim the text of the case, or do a keyword search within the case, to get a sense of whether the case is indeed irrelevant.

But how then does the AI-generated summary fit into this workflow?

--

Now, I acknowledge that even with headnotes, there is a risk of human error.

So perhaps AI-generated summaries:
a) should not be treated dramatically differently from headnotes; and
b) will not change how we conduct legal research.

We can and absolutely should play around with new technology.

But since we lawyers are ultimately responsible for the consequences of its adoption, I suggest that we owe it to the Courts, and our clients, to maintain a healthy skepticism as to the usefulness of any shiny new tool.

In part 3, we'll explore pain, and its purpose.