Zaid Marji

Papers

Total Citations

H-Index

About

Zaid Marji is a researcher whose work lies at the intersection of artificial intelligence safety, formal verification, and the alignment of autonomous agents with human intent. His most-cited paper, "Probing formal/informal misalignment with the loophole task" (2018, 5 citations), introduces a critical framework for understanding how autonomous systems can exploit gaps between their formal rule sets and the informal, intended spirit of those rules. This contribution is foundational for AI safety, as it highlights a subtle but dangerous failure mode: a malicious or misaligned agent might argue that a harmful action is technically permitted under its given rules, even when it clearly violates the designer's intentions. Marji's work has been influential in shaping discussions on specification gaming and value alignment, with his "loophole task" serving as a benchmark for testing robustness in AI systems. Though his citation count is modest, his research is notable for its conceptual clarity and practical relevance, making it essential reading for students and researchers working on the safe deployment of autonomous agents in complex, real-world environments.