Predicting judicial decisions is not terribly impressive
Prometea "predicts the solution to a judicial case in less than 20 seconds with a 96 percent success rate"
And how would that compare to, say, a simple HMM trained on a few parameters such as the relative wealth of the defendant?
In fact, since the article suggests this was developed by a single programmer for the equivalent of $500, that may well be all it is. Or some other simple classification system, like Naive Bayes or a decision tree or an SVM. Frankly it sounds like a student project.
But the real issue is that predicting "the solution to a judicial case" just means you've automated whatever distribution was already in the system, improving nothing except the cost of maintaining whatever level of (in)justice was already there, while covering it with a veneer of "computer says so". It's a terrible, terrible idea.