Towards Superintelligence Alignment
agendareviewed

MAI + CIP (institutional alignment)

Can thick values and deliberative processes stay identifiable, contestable, and robust under strategic pressure and AI mediation?

Introduction

MAI and CIP pursue full-stack alignment through thick values, alignment assemblies, and collective constitutional AI—aligning AI–institution systems rather than model behavior in isolation. The agenda asks whether deliberative processes stay identifiable and contestable under strategic pressure and AI mediation. Legitimate institutional output does not imply correction-channel integrity  or institutional-selection-gating that survives optimization.

Who carries it: Meaning Alignment Institute (Joe Edelman et al.); Collective Intelligence Project (Divya Siddarth, Saffron Huang et al.)

What they aim to do. Align AI–institution systems and legitimate value aggregation—not model behavior in isolation.

The hard question. Can thick values and deliberative processes stay identifiable, contestable, and robust under strategic pressure and AI mediation?

What they produce. Full-Stack Alignment research, alignment assemblies, collective constitutional AI, and thick-value / institutional design programmes.

Key terms. Key terms include full-stack alignment, thick values, alignment assemblies, deliberative alignment, and collective constitutional AI.

Related field cruxes. Value Learning; Audit Independence; Goodhart Selection; Extrapolated Volition

What they contribute. Institutional amplification as a failure mode; operational deliberative alignment; counterfactual value commitments distinguished from surface preferences.

How this project treats it. A legitimate process does not imply correction-channel integrity ; consensus output does not imply bearer-persistence  under optimization.

Map clustering

AISafety.com map listings that roll up to this agenda:

See the coverage matrix for evidence tagged to this agenda, and the glossary for shared terms.