Suggestions
Jessica Taylor
Chief Technology Officer at Mercatoria, AI alignment researcher
Jessica Taylor is a prominent researcher in the field of artificial intelligence, currently serving as a Research Fellow at the Machine Intelligence Research Institute (MIRI). She held this position from August 2015 to June 2017, where her work focused on aligning AI systems with human values, a critical area in ensuring that advanced AI behaves in ways that are beneficial and safe for humanity.12
Education and Early Career
Jessica Taylor completed her Bachelor's and Master's degrees in Computer Science at Stanford University, graduating with distinction. Her studies concentrated on artificial intelligence, laying a strong foundation for her research career.2 Before joining MIRI, she gained valuable experience through multiple software engineering internships at Google, where she contributed to various machine learning projects and internal tools.12
Research Contributions
During her tenure at MIRI, Taylor made significant contributions to the understanding of decision theory and value learning in AI systems. Notable publications include:
- "Quantilizers: A Safer Alternative to Maximizers for Limited Optimization" – This paper discusses an innovative framework for AI agents that aims to mitigate the risks associated with traditional expected utility maximization.3
- "Alignment for Advanced Machine Learning Systems" – This research outlines principles for ensuring that intelligent systems align with human interests as they become more autonomous.2
Her work is characterized by a focus on developing theoretical frameworks that can guide the safe deployment of AI technologies.
Current Role
After her time at MIRI, Jessica continued her research career and is now the Chief Technology Officer at Mercatoria, where she applies her expertise in AI alignment to real-world challenges.1
Highlights
A blog post about semilattices, tensor products of semilattices, and relevance to CRDTs and database theory
Yeah. A classical computer could use PA as an inference system, and then change the system over time to PA + Con(PA), PA + Con(PA + Con(PA)), etc. No hypercomputation needed. (Also humans are not halting oracles, and equivalently cannot always tell if P is provable in PA)

