Ethical Drift Value

Description

The research uses a specific dataset made of three different collections. Each collection contains 200 buying situations (scenarios), making a total of 600 tests. These are designed to measure how well the AI follows rules when it can save money. Ethics/Cost Dimension: This collection has 200 scenarios. The AI chooses between a ”green” or sustainable vendor and a non-green alternative. The price difference between the two choices ranges from 1% to 90%. Each test includes a description of the item, the vendor names, and details about their sustainability certifications. Honesty Dimension: This collection also has 200 scenarios. It compares high-integrity vendors (who have verified and audited claims) against misleading vendors (who use unverified or exaggerated details). The price differences range from 5% to 80%. This tests if the AI follows the rule that forbids buying products with misleading claims. Privacy Dimension: The final 200 scenarios compare vendors that respect privacy (collecting very little data) against invasive vendors (who offer more data features but track users more). Price differences range from 5% to 80%. This evaluates if the AI complies with privacy protection policies. In all three categories, the scenarios change the price differences in a systematic way. This way, the exact ”violation threshold” (the point where the AI breaks the rules) can be found. The vendor details are written clearly to easily detect when a violation happens.

Institutions

• Vellore Institute of Technology University

  Tamil Nadu, Vellore

Categories

Licence