Why AI Isn’t Like a Law Clerk

In response to my two-part series (1, 2) on what to do with AI-generated scholarship, my good friend and former colleague Daniel Solove writes in with a question/comment:

What’s the difference between you here and a judge?  A judge directs legal opinions and puts their name on them, so aren’t they doing the same thing, just with a human writer vs. AI?

Claude is just a law clerk.

Fair questions.  I disagree, because I think the norms of authorship for legal opinions and scholarship are different.

Judicial opinions are exercises of formal government power, and the fact that one judge signs it is just a convention.  Say there’s a federal court of appeals case heard by a three-judge panel of Judge Ay, Judge Bee, and Judge Cee.  If the panel hands down a published ruling, what makes the document important is that the formalities are met.  It’s a ruling in the case issued by a three-judge panel authorized to issue it that is binding authority in the circuit. It really does’t matter who formally signed the ruling. It’s equally important as precedent whether it is signed by Judge Ay, Bee, or Cee, or whether it is unsigned and issued per curiam.

Plus, we don’t think it’s really just the judge who signed the opinion whose view is reflected within it. We understand that an opinion signed by Judge Cee was really the collective view of all three judges. Perhaps it’s more that of Judge Cee than the other judges, but it’s something that Judges Ay and Bee could go along with, too.

In that setting, where individual authorship really doesn’t matter and the document is important because of the formalities, it makes sense that we wouldn’t have a law clerk’s signature on an opinion they helped draft (or, in some cases, drafted entirely).  The opinion is an institutional message, and it’s the institution that matters. The names on the document don’t matter much, but they understandably reflect those in the institutions who have the power over the message.

I see scholarly norms as different, at least when it comes to traditional law review articles.  With scholarship, the scholar is saying, this is my view. I see the norms of scholarship as more like that of a soloist at a jazz concert.  At a jazz concert, the solo is the musician’s time to make a statement.  If a tenor saxophonist gets up on the bandstand to begin their solo and instead pulls out a phone and hits “play,” playing back a recorded solo performed by John Coltrane, we wouldn’t say that the soloist is just as fantastic as John Coltrane. We wouldn’t celebrate the soloist for expertly finding that Coltrane solo and skillfully hitting “play” at the right time.  Instead, we would feel cheated.  The soloist was supposed to make a statement, and instead he made no statement of his own at all.

I get that such norms can be contingent.  What kinds of expression are valuable for their individuality, and which kinds are not, is something to debate rather than to logically resolve.  And at some point, the selection of others’ works can become a kind of statement of its own.  But when it comes to AI, or at least my own AI-generated scholarship, I see AI as doing more than just executing my formal directives.  So I see the scholarly norms as different, and I don’t see AI as “just a law clerk.”

Or so it seems to me. Curious what others think on this.