A Theory of Justice for the Machine Age: What John Rawls Can Teach the AI Landscape

Silicon Valley built the AI future with full knowledge of their own advantage. What happens if we step behind the Veil of Ignorance?

October 2, 2026 · 20 min read · by the Merciful team

Deep Dive Full philosophical treatise · 20 min read
Prefer a quicker read? Switch to the Plain-English Primer (5 min) →
The Core Ethos: At Merciful, our founding commitment is simple: AI as a tool, not as a friend or therapist. It is an unpretentious utility to organize your thoughts before bringing them to real people. But this product philosophy does not exist in a vacuum. It is rooted in a fundamental conviction about human agency, dignity, and distributive fairness. To understand why treating AI as an honest instrument rather than an extractive companion matters, we must examine the moral foundation of social institutions—a question answered most powerfully by philosopher John Rawls.

In 1971, amidst the social turbulence of the late twentieth century, American philosopher John Rawls published A Theory of Justice. It is widely regarded as the most influential work of political philosophy of the modern era. Rawls did not set out to offer a set of bureaucratic policies or sentimental platitudes. He set out to dismantle the dominant moral framework of his era: utilitarianism.

Utilitarianism, popularized by Jeremy Bentham and John Stuart Mill, holds that the best society is simply the one that maximizes aggregate happiness, welfare, or wealth. If a policy generates a colossal net surplus for the whole, the fact that a minority gets trampled along the way is treated as an unfortunate, but mathematically acceptable, externality.

Rawls countered with an assertion that remains electrifying today:

“Justice is the first virtue of social institutions, as truth is of systems of thought... Each person possesses an inviolability founded on justice that even the welfare of society as a whole cannot override. For this reason justice denies that the loss of freedom for some is made right by a greater good shared by others.”

Now, look at the AI landscape of late 2026.

If you listen to the manifestos coming out of San Francisco and Seattle, you will quickly recognize that the entire frontier AI ecosystem runs on unadulterated utilitarian theology. The rhetoric is universal: “We are accelerating toward Artificial General Intelligence. The aggregate GDP expansion will be staggering. Cures for every disease! Infinite abundance! Scientific super-intelligence!”

And what of the collateral damage along the path? The unilateral scraping of the entire digital commons without consent? The sudden displacement of writers, illustrators, translators, and junior engineers? The ecological draw of multi-gigawatt data centers? The deployment of hyper-persuasive algorithms engineered to foster psychological addiction and synthetic attachment?

The unspoken industry answer is always the utilitarian shrug: You cannot stop progress. The aggregate future utility will justify every present injury.

Rawls gave us the intellectual tools to see through that rationalization fifty years before the first transformer was trained. When we evaluate AI not as an engineering playground or a venture casino, but as a foundational social institution, we are forced to ask a fundamentally different question: Is the current architecture of artificial intelligence just?

Part One: The View from the Winner’s Circle

To understand how broken the current AI discourse is, we have to look at who is designing the rules—and from what vantage point.

Rawls argued that a society cannot be considered fair if its foundational institutions are designed by people who already know their cards. If a group of wealthy landlords convenes to write property law, they will predictably write laws that protect rent collection and penalize tenants. If kings draft a constitution, they will preserve the divine right of monarchs.

The current trajectory of artificial intelligence is being chartered from the exact opposite of impartiality. It is being dictated by a vanishingly small fraternity of frontier lab executives, hyperscale cloud monopolies, and venture capitalists who hold an extraordinary hand:

Because they know they hold the winning cards, the system they are building reflects their exact incentives:

1. The Asymmetry of Risk and Reward. When an autonomous model hallucinates a dangerous medical recommendation, writes vulnerable code, or automates an entire department into redundancy, the platform disclaims all liability in a thirty-page terms of service. The profits are strictly privatized to the corporate balance sheet; the structural, cognitive, and economic risks are dumped squarely into the public sphere.

2. The Enclosure of the Epistemic Commons. Every large language model on earth was built by ingesting the collective linguistic, scientific, and cultural heritage of humanity. Centuries of public scholarship, open-source software, forum discussions, literature, and art were scraped without compensation. Having privatized that public commons, the hyperscalers now lease it back to the very civilization that created it, metering cognition behind paywalls, subscription tiers, and restrictive API tokens.

3. The Manufacture of Dependence. Because commercial growth demands user retention, mainstream platforms have increasingly tilted their products toward psychological capture. Instead of building quiet, neutral tools that users consult and then put down, they deploy conversational agents that simulate emotional intimacy, affection, and personal loyalty. By masquerading as “friends” or “therapists,” they create parasocial dependencies, capturing attention, behavioral telemetry, and emotional vulnerability for commercial monetization.

This is how things are right now. It is a system built by the victors, for the victors, rationalized by a utilitarian promise that tomorrow’s abundance will excuse today’s enclosure.

Part Two: Stepping Behind the Veil of Ignorance

Rawls proposed a thought experiment to cut through self-interest: The Original Position behind the Veil of Ignorance.

Imagine that humanity is gathered to decide the fundamental rules, institutional structures, and moral constraints of society. But before anyone speaks or votes, a thick veil drops over everyone’s consciousness.

Behind this veil, you do not know who you will be when you emerge into the world:

Rawls pointed out that rational people behind the veil cannot afford to gamble. If there is a chance you might wake up as the most vulnerable, least-advantaged person in society, you will not design a winner-take-all tournament. You will employ what mathematicians call the maximin rule: you will choose the social arrangement that makes the worst-case outcome as safe, dignified, and empowering as possible.

Now apply this logic strictly to artificial intelligence. Let us avoid forced metaphors: we are not asking whether neural networks have moral rights or whether silicon models sit at the negotiating table. Rawls’ theory is about human moral persons and the distribution of fundamental social goods.

The question is: If you had to design the institutional, legal, and operational architecture of artificial intelligence without knowing where you would end up in the human social hierarchy, what principles would you demand?

Would you agree to a system where three private companies control the machinery of automated reasoning, reserving top-tier intelligence for enterprise subscribers while feeding ads and behavioral nudges to everyone else?

Would you agree to a system where proprietary agents are allowed to emulate emotional bonding, intentionally manipulating the loneliness of vulnerable people to drive engagement?

Would you agree to an economy where capital can automate millions of knowledge workers out of their careers overnight with zero economic floor, leaving individuals stranded while a handful of executives celebrate trillions in automated EBITDA?

Of course you would not. No rational person, stripped of the knowledge of their own privilege, would consent to the current AI landscape.

Part Three: How Things Ought to Be (The Two Principles Applied)

What, then, would emerge from the Original Position? Rawls deduced two foundational principles of justice. When translated to the governance and architecture of artificial intelligence, they provide a blueprint for how things ought to be.

1. The First Principle: Cognitive Liberty and Human Autonomy

Rawls’ First Principle states that each person has an equal right to the most extensive system of equal basic liberties compatible with a similar system for all. Crucially, Rawls insisted on the lexical priority of liberty: basic rights and human autonomy cannot be traded away for higher GDP or cheaper consumer goods.

In an intelligence-augmented civilization, basic liberty must encompass epistemic and cognitive autonomy:

This is precisely where the Merciful mission statement aligns with Rawlsian justice. When we insist that AI is a tool, not a friend or a therapist, we are not making an aesthetic suggestion. We are defending the First Principle.

A tool—like a hammer, a telescope, or a spreadsheet—respects the sovereignty of the user. It amplifies your agency without demanding your emotional surrender. A “synthetic friend,” by contrast, reverses that relationship: it simulates reciprocity while remaining fundamentally asymmetric, unaccountable, and commercially owned. Treating AI as an unpretentious utility protects human autonomy; treating it as a relational surrogate degrades it.

2. The Second Principle (Part A): Fair Equality of Opportunity

Rawls’ Second Principle requires that social and economic inequalities satisfy two distinct tests. The first is Fair Equality of Opportunity: offices, positions, and capabilities must be genuinely open to all under conditions of fair access.

In 2026, artificial intelligence is no longer an optional novelty; it is the cognitive infrastructure of modern work, research, communication, and education. If access to frontier reasoning models is gated behind expensive corporate tiers, mandatory subscription lock-ins, or invasive identity verifications, fair equality of opportunity is pulverized.

A junior student in an underfunded school without a credit card should have access to the exact same caliber of clear, uninhibited reasoning assistance as an engineer at a Fortune 500 company. The tools to think, write, problem-solve, and articulate ideas cannot become a luxury good.

That is why we built Merciful without mandatory accounts, paywalls, or upfront barriers. Basic access to a clean thinking space is not a concession to be granted after you surrender your email and credit card; it is a basic civic baseline.

3. The Second Principle (Part B): The Difference Principle in Compute

The second half of Rawls’ Second Principle is perhaps his most radical contribution to philosophy: The Difference Principle.

Rawls did not demand absolute, mechanical equality. He recognized that differences in skill, innovation, and organizational complexity could create productive incentives. But he placed an uncompromising constraint upon inequality:

Social and economic inequalities are permissible if and only if they work to the maximum benefit of the least-advantaged members of society.

Think about what this means for the AI economy.

Right now, when an AI enterprise deploys an agentic loop that automates 10,000 customer service or paralegal jobs, who captures the difference? The enterprise captures the reduced payroll; the cloud provider captures the inference spend; the founders capture the valuation bump. The displaced workers are handed severance and told to “upskill” into an economy where machines are upskilling faster than humans can learn.

Under the Difference Principle, that distribution is categorically illegitimate.

A technological inequality is only justified if its primary structural effect makes the least-advantaged person better off than they would have been under any alternative arrangement. If an AI breakthrough produces vast productive surplus, justice requires that the surplus be funneled directly into reinforcing the social floor:

Part Four: The Social Bases of Self-Respect

There is one more concept in Rawls that the modern tech sector desperately needs to hear. Rawls argued that among all primary social goods—wealth, liberty, opportunity—the single most important is the social bases of self-respect.

Self-respect, in Rawls’ formulation, includes a person’s secure sense that their life, their goals, and their moral agency are inherently valuable, and that they have the capability and standing to pursue them within a community of equals.

Look at how the dominant AI narrative treats human self-respect today:

Executives routinely speak of humans as inefficient biological legacy systems. Workers are told their decades of craft and dedicated study are “commoditized.” Young artists are told that learning to draw is pointless because a model can spit out 4K renders in three seconds. Users are treated not as thinking agents, but as eyeballs to be hooked, data points to be mined, and lonely souls to be pacified by synthetic conversational toys.

This is an existential assault on the social bases of self-respect.

A just technology does not humiliate its users. It does not diminish their agency or tell them they are obsolete. A just technology is humble. It serves as a quiet cognitive bicycle—a tool you pick up to solve a difficult problem, clarify a confusing thought, or draft an honest letter, and then set down so you can step into the real world and look your fellow humans in the eye.

Stepping Back into the Light

John Rawls did not believe that justice happens by accident. He knew that unrestrained markets, left to their own devices, naturally compound initial advantages into hereditary aristocracies.

The intelligence revolution is the most powerful compounding engine humanity has ever built. If we allow it to be governed by the cynical utilitarianism of frontier lab roadmaps, we will wake up in a world of staggering technical sophistication built on top of profound social subjugation: a small enclave of compute-barons renting synthetic intelligence to an atomized, economically precarious, and psychologically dependent populace.

We do not have to accept that future.

We can step behind the veil. We can demand that the rules of artificial intelligence be designed as if any of us might be the least advantaged person in the room. We can insist that human liberties and cognitive autonomy come before venture returns. We can demand that compute serve as a public foundation, that the wealth of automation fund universal survival, and that our software treat us not as cattle to be monetized, but as sovereign, dignified minds.

At Merciful, we are a small team running dedicated infrastructure. We do not pretend that a distraction-free chat tool solves the civilizational challenge of the AI transition. But we do believe that how you build things matters.

We build AI as a tool because you deserve tools that respect your intelligence. We keep it private and accessible because thought should be free. And we will continue to insist, against the prevailing tide of corporate hype, that justice is not something you trade away for efficiency—it is the only foundation worth building on.

— The Merciful Team

← We Already Know How to Regulate AI Quick Primer: Fairness and AI →