Machine Learning (Legal Applications)
Algorithms that learn patterns from labeled legal data — relevance decisions, risk labels, outcome records — to make predictions on new documents or cases; TAR is the most established application.
Last reviewed: 2026/05/19
Definition
Why It Matters for Lawyers
How AI Tools Handle It
Frequently Asked Questions
- Q: What is the difference between a rule-based system and a machine learning system?
- A rule-based system applies explicitly programmed rules: "if the clause contains the phrase 'limitation of liability,' flag it." An ML system learns rules from examples: "given 1,000 labeled contracts, learn what features distinguish high-risk from low-risk limitation of liability clauses." ML handles variability and edge cases better; rule-based systems are more predictable and auditable.
- Q: How much training data does a legal ML model need?
- It depends on the model type and task. Traditional ML models may require thousands of labeled examples per class. Modern deep learning models with transfer learning can perform useful tasks with hundreds of examples. Few-shot LLM techniques can achieve reasonable performance with ten or fewer examples. Ask vendors specifically how much labeled data their models require for your use case.
- Q: Can I train a model on my firm's own data?
- Yes, with caveats. Custom model training requires sufficient labeled examples, data science expertise, and ongoing maintenance. Some platforms (Kira, Relativity) support custom model training within their products with accessible interfaces. Fully custom model development requires data science resources. Evaluate whether the performance benefit of firm-specific training justifies the investment. --- *Last reviewed: 2026-05-19 by LawyerAI Editorial Team.*
Related Concepts
Related Tools
- Luminance
Enterprise AI for portfolio-level contract analysis and institutional memory.
Related Reading
Last reviewed: 2026/05/19. Definitions are written by the LawyerAI Editorial team. We do not accept affiliate commissions; Featured placement is clearly labeled and does not influence editorial content.