Home /Modules/artificial-intelligence/README.md

README.md

artificial-intelligence/large-language-models/law/README.md · 49 line(s) · markdown

Law: Moral Reasoning, Contracts, and Public Standards

This directory contains a carefully labelled starter corpus for legal reasoning and structured interviews. It is designed for educational model development, LSAT-style analytical practice, contract-principle exploration, and reflective discussion of public-service standards.

M/I: morals and intellect

M/I is a conceptual lens for studying how moral commitments and intellectual reasoning interact. It is not a validated psychological scale, a measure of intelligence, or a formula for ranking people. The corpus does not assign a numeric moral score or infer a person's inner character from a short answer.

The model should learn to:

Files

FileContents
interview-standards.jsonlBinary opening questions followed by paragraph prompts that invite explanation
contract-principles.jsonlContract formation, consent, reliance, good faith, breach, and remedies
lsat-reasoning.jsonlArgument structure, assumptions, inference, flaw detection, and conditional logic
public-service-judgment.jsonlEthical scenarios relevant to public service and diplomatic work

Record format

Each file is UTF-8 JSON Lines. Every line is one JSON object with prompt and completion strings, compatible with the minimal supervised format documented in the parent datasets/README.md.

Interview prompts use a two-stage structure:

  1. Binary position: ask the respondent to answer Yes or No (or explicitly state that neither answer fits).
  2. Explanatory paragraph: ask what the answer means, which principle supports it, what facts could change it, and what fair exception might apply.

The binary answer is a starting point, not a complete evaluation. Preserve the paragraph as the primary evidence of reasoning. Do not treat vocabulary sophistication, verbosity, accent, education, or agreement with a preferred political or moral position as proof of integrity or competence.

Responsible use

Suggested labels for downstream annotation

domain, topic, response_mode, reasoning_skill, principle, uncertainty_present, counterexample_present, and source_type.

Use source_type: synthetic for every record in this initial corpus. If future examples are based on statutes, cases, or official materials, add the jurisdiction, citation, retrieval date, and rights/attribution information rather than silently blending them with synthetic examples.