Given TFA, it seems that having human determinations involved might run the risk of prejudicing the AI.

“Ladies and gentlemen of the jury, I actually asked ChatGPT and it said my client is not guilty.”