Integrity Under Ignorance

Integrity remains a critical issue in contemporary society, as individuals often claim one thing yet act contrary to it when it serves their interests. This challenge is particularly pronounced in high-risk environments, such as biological weapon management and advanced AI systems.

In these contexts, individuals may be tasked with actions that cannot be fully explained, often due to the sensitive nature of the underlying information. For instance, the fear of triggering an arms race or jeopardizing AI safety can necessitate confidentiality regarding certain operations.

This repository explores innovative solutions aimed at enhancing personal integrity and fulfilling commitments in such complex scenarios. The potential of artificial intelligence (AI) is significant; when utilized correctly, AI can help ensure adherence to commitments by:

• Developing tools that track and assess compliance based on established commitments.
• Acting as impartial entities to verify and publicize commitment fulfillment.
• Monitoring others' adherence to commitments, thereby fostering accountability.

Furthermore, AI could serve as a mediator when individuals are faced with tasks lacking full transparency. By training AI systems with specific knowledge and preferences, these systems could facilitate commitment-making, offering appropriate compensation for compliance. This approach raises important questions: Who finances these informed AIs? How can such AIs interact, and what mechanisms ensure the secure integration of new confidential information into the system?

This repository aims to prototype various implementations of these concepts within controlled environments, assessing which methods are the most humane and effective in promoting integrity amidst uncertainty. By leveraging technology, it seeks to transform how commitments are managed and honored in complex situations.