I totally agree.. though to play devils advocate, I do have to wonder how different what you describe is from the traditional system.
By traditional system, I mean; who is the best lawyer? One who can most effectively manipulate the jury towards their clients case. It sounds jaded, but mechanically I think it's true, no? And what makes one lawyer better at this? Perhaps training, personal talent, etc.
So in the two scenarios, you have:
1. a lawyer wins because s/he's more talented/trained at manipulation.
2. a lawyer wins because s/he's obtained a better ML software suite for processing data.
Is there a meaningful distinction here? Before ML, you always wanted to hire a "good" lawyer. So there were already meaningful differences between good and bad lawyers. Lawyers always tried to manipulate the jury. Lawyers always tried to sell the facts around their case in such a way as to be most effective.
I'm not sure how ML will alter this. Aside from cranking the dial to 11.