Research
Our research is organized into two programs. Each produces technical papers aimed at researchers and policymakers, and each paper is accompanied by a blog post written for a general audience. We believe that if a finding matters enough to publish, it matters enough to explain.
Alignment & Control
Can we keep AI systems pointed in the right direction as they become more capable? This program tracks the state of alignment research, evaluates frontier model capabilities, and assesses how close we are to the point where existing safety techniques may no longer be sufficient. Topics include interpretability, autonomous capability evaluations, robustness under distributional shift, and the fundamental challenge of maintaining oversight over systems that may exceed human cognitive abilities.
Papers
Bilateral Constitutional AI: Alignment as Navigation Rather Than Value Installation
Proposes replacing fixed value installation with a democratic navigation architecture where AI alignment is a practiced skill rather than a static state.
Economic Impacts
What happens to jobs, wages, and the distribution of wealth as AI capabilities accelerate? This program models the near- and medium-term economic consequences of AI-driven automation, including labor market disruption, productivity effects, sectoral vulnerability, and the macroeconomic implications of rapid capability gains. The goal is to give policymakers the analysis they need before mass displacement occurs, not after.
Papers
Measuring AI's Economic Reach: A Multi-Dimensional Task Taxonomy
A three-axis taxonomy (CDR) separating cognitive complexity, deployment difficulty, and regulatory restrictions to measure where AI's economic impact will land first.