The Black Box
Bench.
From Technology-Assisted Review (TAR) in discovery to algorithmic risk assessments in sentencing. When the judge is an algorithm, how do we preserve Due Process?
Predictive Justice & Sentencing
Tools like COMPAS (US) use historical data to predict recidivism risk. The controversy (State v. Loomis) lies in proprietary algorithms acting as "Black Boxes," preventing defendants from challenging the specific weighting of factors used to deny them parole.
In the EU, the AI Act classifies "Administration of Justice" systems as High Risk, mandating strict transparency logs and accuracy testing to prevent bias against minorities.
E-Discovery & TAR
The sheer volume of digital evidence makes manual review impossible. Technology-Assisted Review (TAR) using Predictive Coding is now judicially accepted (e.g., Da Silva Moore).
The "Seed Set" Challenge
The legal battleground has shifted from "Can we use AI?" to "How was the AI trained?". Parties now litigate over the composition of the "Seed Set" (the initial documents reviewed by senior lawyers to train the model). Bias in the seed set leads to bias in discovery production.
Judicial Analytics
Analyze a specific judge's ruling history on AI matters. Predict motion outcomes based on vector similarity of past cases.