"Reverse Alignment" initiative argues society—not just models—needs to be aligned for AI

profjamesevans · x · 2026-11-06

Scholar James Evans and the Sociotechnical Grand Challenges Coalition launched "Reverse Alignment" (reversealignment.ai), arguing that beyond model alignment, institutions, norms, skills, and governance must be deliberately redesigned so societies can safely absorb AI.

The core argument: nearly all AI effort and capital goes to making systems more capable and aligning them with human values, but even a perfectly aligned model will underdeliver or cause harm if released into unprepared economies, governments, and information systems. Historically, industrialization, electrification, and the internet each required decades of institutional invention before benefits arrived.

The project identifies three predictable failure modes:

Productivity without prosperity: work is reorganized faster than people can retrain, so gains never reach most workers;
Execution without verification: as the cost of generating text, images, and code collapses, the scarce input becomes human capacity to check, attribute, and stand behind output;
Capacity without constraint: the same tools concentrate power unless rights of appeal and contest advance alongside them.

The authors claim the same designs double as defenses: attestation webs resist agent swarms, bridging resists manufactured consensus, and surprise-weighted credit resists synthetic flooding.

Original post →

"Reverse Alignment" initiative argues society—not just models—needs to be aligned for AI

More from AGI Musings