AI safety · Mathematics · Policy
Our research spans the technical, mathematical, and policy questions that matter for preventing human extinction from advanced AI.
Library
A Case for Exclusive Resource Dedication Under Catastrophic AI Risk
When should society dedicate all its available resources to AI alignment? This paper sets out and justifies an allocation rule for that decision.
Inspired by Absolute Alignment Allocation
A fourth-dimensional traveler separates the present into two worlds. Starting with the same machines and unfinished questions, they take different paths toward superintelligence.
Areas of inquiry
Different disciplines bring different tools to the problem of advanced AI. HARPI’s scope extends across these three areas.
How can advanced AI remain safe?
We are interested in how increasingly capable systems can be aligned with human intentions, and what evidence would justify confidence in their safety.
What mathematical foundations do we need?
Our scope includes formal approaches to uncertainty and decision-making, alongside fundamental questions in mathematics itself. Foundational results can be valuable before their eventual applications are known.
How should institutions respond?
We examine how governments and organizations can respond to the possibility of catastrophic AI harm, including decisions made under uncertainty and competitive pressure.
Contribute
Research proposals, technical criticism, and collaboration are all ways to contribute.
Zenodo
GitHub