Ultimately all AI from simple multilayer perceptrons to large neural networks
are ultra-high degrees-of-freedom pattern detectors. A suitably trained
network filled with all known/published case law, doctrines, strategies
would conceivably be able to predict the outcome(s) of a hypothetical legal
scenario with a high degree of confidence, and the optimal strategy. The
ultimate legal clerk on steroids, but being REPRESENTED by an algorithm is
another matter. I’d first ask the AI what my legal odds were if I used AI
representation and watch its circuits overheat.
The AI could also consider other factors that may or not be relevant: the judge or jury, the time of year, weather, characteristics of the defendant, etc.