✳flâneur — a map of the web's best reading

Whitepaper — AI • Objectives • Institute

ai.objectives.institute · 6,888 words · saved by 1 readers

We argue that, rather than a sharp break, there is continuity between problems in AI alignment and those in economic regulation, institutional design, and personal auton- omy. There are therefore under-appreciated opportunities for ideas and interventions to flow between research communities in both directions. On one hand, the AI safety community may be able learn from successes and failures in aligning other kinds of superhuman optimizing entities, such as corporations and political parties, which are susceptible to biases analogous to those in AI systems. On the other hand, advances in AI (such as large language models) and breakthroughs in AI safety (such as advances in cooperative inverse reinforcement learning and reinforcement learning from human feedback, improvements in measuring Goodhart’s law, and techniques for developing systems without explicit objective functions) could provide new tools for improving ex- isting regulations, institutions, and self governance, or helping

Whitepaper Contents Introduction AOI’s Research Areas Alignment of Markets, AI, and Other Optimizers Direct Isomorphisms Between Neural Networks and Markets Scaling Cooperation with AI Assistance Human Attention and Epistemic Security Conclusion Download PDF AI Objectives Institute Whitepaper * A Research Agenda for the Production of a Flourishing Civilization February 2023 Abstract We argue that, rather than a sharp break, there is continuity between problems in AI alignment and those in economic regulation, institutional design, and personal autonomy. There are therefore under-appreciated op

Explore this link on the map →

related reading