The institute
About HARPI
The Human Aligned Research and Policy Institute works to help prevent human extinction from advanced AI.
Addressing the risks of advanced AI.
HARPI brings together AI safety, mathematics, and policy. Our purpose is to understand how advanced AI could threaten human survival and what can be done to prevent that outcome.
This requires work at several levels: studying the behavior of AI systems, developing mathematical foundations, and examining the decisions made by governments and organizations. Technical progress, sound reasoning, and effective institutions all matter.
A broad research agenda.
Our interests include AI alignment and evaluation, mathematical reasoning under uncertainty, and policy responses to catastrophic risk. These areas inform one another, while each also raises questions that deserve investigation on their own terms.
Mathematics is part of HARPI’s remit as a discipline in its own right. Foundational work can be valuable before its eventual applications are known.
Research areas and publicationsEvidence, argument, and revision.
We aim to make research clear enough to be challenged. A proof should state its assumptions. An empirical claim should identify its evidence. A policy recommendation should explain the values and tradeoffs behind it.
Research should remain open to criticism and correction. We also support anonymous and pseudonymous authorship, allowing researchers to contribute without making their identity or affiliation the basis for evaluating their work.
Our rationale for anonymityGet involved
Contribute to the work.
Share research, offer criticism, or explore a collaboration.