Reverse Alignment

The Other Side Of Alignment

The idea / Introduction

Most of the effort and capital in AI goes to two tasks: making systems more capable, and aligning those systems with human values.

Both matter. But even a perfectly aligned system will underdeliver, or do real harm, if it is released into economies, governments, and information systems unprepared to receive it.

We call that neglected work reverse alignment: the deliberate redesign of institutions, norms, skills, and governance so that societies can absorb AI safely and share its gains widely.

This is not a new kind of problem.

Industrialization, electrification, and the internet each demanded decades of institutional invention before their benefits arrived. The companies of the assembly line looked like assembly lines. The institutions that governed the internet looked like the internet.

History suggests these adaptations are not optional add-ons to technological change. They are the conditions for its success.

Left unaddressed,

the mismatch produces three predictable failures.

1

Productivity without prosperity

Work gets reorganized faster than people can retrain or move, and the gains never reach most of them.

2

Execution without verification

As the cost of generating text, images, and code collapses, the scarce input becomes the human capacity to check, attribute, and stand behind what gets produced.

3

Capacity without constraint

The same tools that make institutions more capable can concentrate power, unless rights of appeal and contest advance alongside them.

None of these is inevitable.

Each is a consequence of neglect, and neglect can be reversed.

What Needs Building

Human flourishing over the next decade depends on a set of sociotechnical grand challenges.

Summary of recommendations

Large, coordinated investments in the human and institutional side of the AI transformation. They span identity, privacy, markets, governance, work, and knowledge. They share an architecture, and they share a method. No single sector can deliver any one of them alone.

*Hover to read more *Click to read more

As fake documents, biometrics, and synthetic personas approach zero cost, we need ways to prove a real, unique human is present without revealing who they are.

001

Identity

AI gives everyone a personal detective; the answer is not blanket secrecy or full exposure but tools that let people reveal exactly what a situation calls for and nothing more.

002

Privacy

When a convincing fake costs nothing to make, knowing where content came from becomes essential to a shared reality.

003

Provenance

The human record that trains these models was taken at almost no cost; the people who created it deserve a real market and durable compensation, which the companies’ own measurements already make possible.

004

Data value

AI agents are most useful when they pool what each one knows, and that requires trust built the way humans build it, across providers rather than inside walled gardens.

005

Agentic collaboration

Media that rewards outrage and obscures who agrees with whom can be redesigned to surface common ground while representing real disagreement fairly.

006

Communal sensemaking

The authority that thriving communities generate has to translate into real power over resources, which means modernizing participation, representation, and the unglamorous work of implementation.

007

Democracy

AI lowers the cost of surveillance and enforcement in ways that favor the executive; restoring balance means building equivalent capacity for oversight, contest, and appeal.

008

Law and liberties

The bottleneck to AI’s productivity gains is not individual speed but whether organizations can be redesigned to absorb a surge of creativity and agency.

009

Workplace

Models now generate hypotheses and drafts faster than peer review can sort them, and the same tools, aimed well, can reward genuinely original, silo-breaking work.

010

Research

AI breaks the old bundle of teaching, learning, and credentialing, forcing schools to name the capacities that matter and certify what people actually know.

011

Education

A workforce system built for mass layoffs at a single employer cannot handle the continuous, fine-grained reshaping of work that AI produces; it needs portable, interpretable credentials and real-time retraining.

012

Labor transition

The distinctive claim

Institutions In The Image Of Their Tools

There is a pattern in how societies absorb a new general-purpose technology. The institution that succeeds tends to mirror the architecture of the era’s dominant way of thinking.

The industrial age organized cognition around standardized accounting and the assembly line, and it produced hierarchical, divisional institutions to match: the corporation, the central bank, the chartered multilaterals of Bretton Woods. The internet age organized cognition around layered protocols and distributed routing, and it produced layered, cooperative bodies to match: the standards organizations, the messaging cooperatives, the open process of the Request for Comments.

Neural networks impose a third grammar. They represent things by their position in a rich space rather than by a single label. They combine differing inputs by weighting how each bears on the others, finding common structure without averaging away difference. They assign credit by tracing each contribution to its effect.

When the institution and the technology share a grammar, the gains arrive. When they don’t, the gains stall. Electrified factories delivered little for nearly two decades while they kept the layout of the steam-powered plant; the productivity came only once factories were redesigned around the distributed electric motor.

The wager of this coalition is that the institutions able to absorb AI will be the ones that instantiate these same moves at the scale of society. The work is to build them on purpose.

Read the full argument — coming soon

The papers

Read The Argument

In Noema —

“What Humanity Needs to Flourish in the Next Decade”

The fuller, public statement of the problem and the agenda: the three failure modes, the historical precedents from mass democracy to Taiwan’s digital governance, and the full set of sociotechnical grand challenges. The place to start.

Read in Noema

In Science —

“In the Image of Their Tools”

Coming soon

The crisp engineering principle behind the agenda: why institutions come to resemble the cognitive technology of their age, and what it means to build in the shape of a neural network. The distinctive thesis, with the historical record and the technical mapping.

Coming soon

The Playbook

What AI Can’t Build, a storyteller’s guide that turns these challenges into dramatic material for writers and creators, with a high-concept pitch and a writers’-room prompt for each one.

The Deck

The summary presentation behind the recommendations.

The Biggest Story Nobody’s Telling

Every institution we rely on was once unimaginable, until stories made enough people care to build it. The human challenges of AI are not policy abstractions. They are the raw material for the most important stories of the next decade, and most of them have not been told yet.

We made a guide for

the people who can tell them.

What AI Can’t Build takes each challenge and hands writers a high-concept pitch and a set of prompts to drop into the writers’ room: the woman fighting for the right to be different people in different parts of her life, the documentary maker whose airtight film meets an equally airtight rebuttal, the data contributor who is rich, essential, and completely anonymous.

See how we brought these ideas into a room with 100 TV writers.

Watch the video

See The Challenges

We are also working with independent creators to bring these challenges to life, one at a time, in plain language. Each short film takes a single problem, from proving you’re human online to paying people fairly for the data that trains AI, and shows why it matters and what a solution looks like.

First films arriving soon.

Fast-grants for storytellers

We are funding storytellers who weave these ideas into their work.

If you write, film, build games, or make things people watch, and you want to incorporate or reinterpret any challenge here, we want to support it.

You don’t have to explain the policy. You just have to make people feel what’s at stake.

Apply for a grant

The Coalition

This work was shaped by people who rarely sit at the same table.

AI researchers and social scientists, technologists and public servants, funders and civic builders, drawn from across regions and political perspectives. That range is not incidental. The method we advocate, finding common ground across real differences without suppressing them, is the method that produced the work itself.

ResearchTechnologyGovernmentPhilanthropyCivil SocietyBusinessEntertainmentMedia

Eric Glen Weyl

Co-founder, RadicalxChange Foundation

Audrey Tang

Former Digital Minister, Taiwan

Jake Hirsch-Allen

Head of Partnerships, The Dais

Vilas Dhar

President, Patrick J. McGovern Foundation

Melissa Valentine

Associate Professor, Stanford University

David Patterson

Google Fellow; UC Berkeley Professor Emeritus

Tiona Zuzul

Associate Professor, Harvard Business School

Jess Scully

Executive Director, RadicalxChange Foundation

Dean Woodley Ball

Head of Strategic Futures, OpenAI

Romain Vakilitabar

Founder & Executive Director, Pathos Labs

Andrew Sorota

Head of Research, Office of Eric Schmidt

Samuel Roland

Fellow, Foundation for American Innovation

Chris White

Lab Director, Microsoft Research

Dr. Tomicah Tillemann

President, Project Liberty

Galen Hines-Pierce

Founder & Principal, Minerva Fund

Ben Buchanan

Former White House AI Advisor

James Evans

Professor, University of Chicago

David Deming

Dean, Harvard College

Dr. Ken Suzuki

Project Professor, Keio University

Blaise Agüera y Arcas

VP & Fellow, Google

Danielle Allen

Professor, Harvard University

Divya Siddarth

Co-founder, Collective Intelligence Project

Tim O’Reilly

Founder & CEO, O’Reilly Media

Zoe Weinberg

Founder & Managing Partner, Ex/Ante

Join the coalition

Add Your Name

Reverse alignment is not one project. It is a portfolio of bets, in every sector, that only a broad coalition can sustain.

If you believe we should build for our social systems as ambitiously as we build for our algorithms, lend your name and help carry the work into your own field.

We are building AI faster than we are building the society to receive it.