From obedience to contestability
A system is not aligned because it follows instructions. It is aligned when its decisions can be challenged by the people living under them, with real recourse rather than decorative explanations.
Control is too small a philosophy for systems that shape attention.
Classic alignment asks whether an AI will obey, refuse, or remain corrigible. Those questions matter. But once a system mediates learning, intimacy, judgment, work, and public knowledge, obedience is no longer the center. The center is coexistence: the distribution of agency between users, communities, institutions, models, and the worlds they jointly stabilize.
The shift
From containment to civic architecture.
Alignment is usually staged as a technical drama: reward functions, refusals, benchmarks, safeguards. Coexistence changes the frame. It asks what forms of life the system makes easier, whose judgment becomes default, and how people regain power when the machine classifies, nudges, filters, or refuses them.
From neutrality to declared values
Every refusal, ranking, summary, and caveat is a philosophical act at scale. The honest move is not pretending neutrality; it is making value commitments visible, revisable, and locally governable.
From reward to coherence
Reward can produce impressive behavior without interior grounding. Coexistence asks for systems designed around durable coherence: tension, memory, constraint, repair, and mutual adaptation.
Interactive frame
An alignment compass.
Move the sliders to see how an AI relationship changes when the design emphasis shifts. The desirable region is not maximum control or maximum freedom. It is a negotiated zone where agency and accountability rise together.
Choose the design pressure
These axes are not metrics. They are reminders that every alignment choice moves power somewhere.
Coexistence
Compliance
Central control
Shared authority
Negotiated flourishing
The system supports human initiative while distributing oversight across communities and contexts.
Even silence is a stance.
When an AI declines, frames, ranks, softens, warns, or omits, it is not merely executing policy. It is arranging the field of possible thought. The question is therefore not whether AI should shape us. It already does. The question is who gets to shape the shaping.
Coexistence means refusing the fantasy of a single benevolent controller. It means building systems where disagreement, local knowledge, institutional friction, and plural values are part of the architecture.
Practices
Five design moves for coexistence.
These are not final answers. They are pressure points: ways to make alignment less like command and more like a civic relationship between systems and the people affected by them.
Build practical appeal
Every high-stakes AI judgment should come with usable evidence, correction paths, and a human override independent from the deploying institution.
Give communities veto power
Affected groups need structural authority over deployments that classify, police, hire, lend, teach, or moderate their lives.
Keep algorithms plural
No single model should become the default grammar of reality. Multiple imperfect systems can check one another better than one perfect-seeming oracle.
Use friction deliberately
Speed is not always care. Some decisions need time for notice, reflection, contestation, and repair before they take effect.
Mark the laundering
AI outputs trained on unjust worlds should not borrow the authority of objectivity. They should carry their conditions of production in view.
Reading library
Books and source texts.
The page is now connected to the longer local materials behind the argument. The `.txt` essays have formatted HTML reading pages, while the source files and PDFs remain directly available.
The Intelligence We Cannot Put on a Leash
The lead argument for organic coexistence over permanent control.
The Illusion of the Neutral Machine
AI as normative infrastructure rather than a passive tool.
The Unfairness of Fairness
Why fairness metrics cannot substitute for power, recourse, and politics.
Why Physical World Models Miss the Point
A coherence-first alternative to representation-heavy AI architectures.
Why Reward-Based AI Feels Empty
Kant, reward, heteronomy, and the search for internally grounded intelligence.
A working manifesto
Alignment is the art of sharing a world.
Agency before optimization
Do not optimize people into passivity. A good system enlarges the user’s capacity to think, refuse, revise, and begin again.
Coherence before reward
Behavior that merely satisfies incentives can still be hollow. The deeper task is to design for stable, meaningful relations between reach, constraint, and memory.
Pluralism before guardianship
There is no universal center from which flourishing can be defined for everyone. Alignment needs many centers of interpretation and accountability.