Liberty, Autonomy, and Tolerance
The perfect life and AI
NB: It has been an unmooring week in liberal society. If you are interested in these topics and want to strengthen them, check out Cosmos Institute’s generous offer to writers.
To be clear, I am unaffiliated with them and can’t participate, I just think it’s cool and a great opportunity for younger folks.
If Artificial intelligence was once a rumor about a distant tomorrow, it isn’t anymore. It is how emails draft themselves, how music finds us, how maps speak, how hospitals predict risk, how bureaucracies move paperwork, and how platforms arrange the stream of our attention. That immediacy invites a much older question: can technology help us live more perfect lives?
The honest answer starts with a reminder. Tools don’t decide what counts as “better.” People do. If there is anything like a perfect life in an age of AI, it will be secured by the same goods that have long made human lives worth living: liberty, autonomy, and tolerance. These are not parameters to adjust but principles that set the stage on which everything else becomes possible.
Liberty is our external protector. Long before circuits and code, thinkers like Locke and Mill argued that human beings begin free and that constraints demand justification. That claim is less abstract than it sounds. Liberty is the lived experience of not having to ask permission to speak, to assemble, to worship, to publish, to choose one’s work and companions, to be eccentric without fear. It is the breathing room in which people test ideas, attempt projects, and fail without being crushed. Rawls later turned the intuition into a rule of fairness: each person should have the widest basic liberties compatible with everyone else having the same. In practice, that isn’t a promise of constant agreement but rather a commitment to keep disagreement from hardening into domination (violent or algorithmic).
Yet the outer space of freedom is not enough if the inner capacity to use it is missing. That is the internal protector of autonomy. To be autonomous is to act from reasons one recognizes as one’s own, after reflection that one can endorse. Autonomy requires negative protections—freedom from coercion and deceit—but it also depends on positive conditions: access to knowledge, time to deliberate, and the skills to weigh options. A medical decision made under pressure and misinformation is a hollow choice. A financial contract clicked through in a fog of dark patterns isn’t really consent. An online life in which every path is subtly nudged by a system that knows our cravings better than we do is not a fully self-directed life. It is not human flourishing. Autonomy is fragile because it depends on attention, and attention is the coin of the realm in the digital economy.
Tolerance is the social glue that binds liberty and autonomy to the reality of pluralism. Living together means living near people who will take paths we think are mistaken or even offensive. Tolerance is not indifferent, instead it is principled and purposeful restraint. It means: unless a practice crosses the threshold of serious harm, we argue with words, not with force. It is grounded in humility because no person or institution has a final claim on truth. In reality and in prudence, societies that police heresy stagnate. Tolerance protects the conditions under which new truths can surface and under which falsehoods can be exposed without violence. It is the difference between a culture that tries to persuade and one that tries to purge.
Set against this moral backdrop, AI is both lever and test. On its best days, AI enlarges the space for human choice. It drafts the routine or mundane so that people can spend more time on the creative. It helps doctors catch illness earlier, teachers personalize lessons, and cities optimize transit. It can power forward science and technology to solve the climate crisis or improve quality life for individuals but also extend that to more individuals. It reveals patterns that would have remained invisible, broadening the horizons of what individuals can attempt. Used well, AI can widen liberty by lowering the practical costs of pursuing one’s projects, and it can support autonomy by giving people better information exactly when they need it.
On its worst days, AI compresses the space it seemed to open. Recommendation engines learn to keep us scrolling by offering what we will click rather than what we will value, trading moments of attention for habits of passivity. Scoring systems whisper judgment into hiring and lending decisions, conferring or denying opportunity in ways that are hard to see and harder to contest. Surveillance tools wrap themselves in convenience and safety while dissolving the private sphere where experiments with identity and dissent can occur without repercussion. When a few firms or agencies control these infrastructures, liberty doesn’t die in a spectacular seizure of power; rather, it constricts as menus shrink and norms harden. Choice remains, but among options preselected by others.
Those tensions are not mysteries of the machine. They are consequences of choices about design, deployment, and governance. That is good news. What people choose, people can revise. Protecting liberty in an age of AI begins with transparency where rights are at stake. If a model helps decide who gets a mortgage or a transplant, the people affected should be able to see how the decision was made and to challenge it. “Black box” is not a synonym for “beyond explanation”. Instead, it is often a policy decision about responsibility and accountability (or lack thereof) disguised as a technical constraint. Due process does not evaporate because an equation is involved.
Privacy is not a nostalgic preference for curtains. It is the precondition for becoming someone. Without spaces shielded from surveillance, we stop trying on ideas we are not yet sure about. We stop reading the strange book or attending the unfashionable meeting. We stop making first drafts of ourselves. AI can be built to preserve those spaces with data minimization, on-device processing, differential privacy, and legal rules that put meaningful control in the hands of individuals. Or it can be built to harvest everything, all the time, for a promise of convenience that turns out to be a down payment on dependence. The choice is ours.
Fairness is not an optional feature in systems that allocate opportunity. Bias is not only an offensive output but also a restriction of liberty by another name, limiting which doors open for whom. Testing models for disparate impacts, ensuring representative training data, monitoring systems after deployment, and giving people clear avenues of redress are not gestures toward virtue. This is how we maintain equal citizenship. If liberty is to be equal, the infrastructure that increasingly mediates access to housing, jobs, and credit must respect that equality.
Democratic governance matters because private systems of coordination will not reliably produce public goods. People affected by systems should have a say in how they are designed and used, especially when those systems shape the conditions of daily life. Participation is not only a matter of legitimacy but also a source of intelligence and system literacy. Teachers know things about classrooms that vendors do not. Patients know things about pain and trust that hospital administrators do not. Workers know things about workflows that executives do not. Bringing those voices into the room improves systems and builds the confidence necessary for adoption. Keeping them in the room ensures goals aren’t pitch-deck pablum but constantly evaluated and jointly wayfound.
Human accountability is the backstop when things go wrong and things have and will go wrong. Responsibility must attach to natural or legal persons, not to abstractions that cannot apologize, compensate, or be sanctioned. “The AI did it” is not an answer; it is an abdication. Keeping humans in the loop where rights are implicated is not a sentimental attachment to a bygone era. It is a recognition that moral judgment involves more than pattern matching. It requires sensitivity to context and a willingness to stand behind a decision.
Education should be treated as critical infrastructure for autonomy. The goal is not to turn everyone into a machine learning engineer. It is to cultivate citizens who can read a model output skeptically, spot a manipulative interface, ask what data a service needs and why, and understand the tradeoffs between predictive power and privacy. Autonomy can’t survive as a matter of theory if, in practice, people approach AI as something that happens to them rather than something they engage. A population literate in these questions is more difficult to mislead and more capable of steering public debate.
Tolerance must be reaffirmed in a digital culture that multiplies both the speed of offense and the temptation to punish it, by state or vigilante. Content moderation is hard because the scale is enormous and the harms are real. Yet a tolerant society resists reflexive bans where persuasion would suffice. It also refuses to launder ideology through code—whether by suppressing art because it offends taste or by curating political speech to protect the comfortable from discomfort. Where AI is used to combat harassment or fraud, restrictions are consistent with the harm principle. Where AI is used to enforce a cultural orthodoxy, restrictions become a rebuke to pluralism. Tolerance also has an architectural dimension as it pertains to AI. Giving users meaningful control over feeds and recommendation settings can produce digital pluralism, where different communities can choose different values without imposing them on others.
Internationally, the virtue of tolerance counsels cooperation over conversion. Liberal societies should advocate for human rights in AI and refuse practices that violate them—mass social scoring, pervasive biometric surveillance, tools that enable persecution. But they should also recognize that there are multiple legitimate paths to governing technology. Exporting principles through example, trade, and dialogue is more sustainable than imagining that power and coercion can inscribe a single moral code onto diverse cultures. Respecting autonomy means leaving room for difference while drawing bright lines around abuse.
None of this means treating technology as an enemy. The point is precisely the opposite: to welcome its benefits without surrendering the goods that make those benefits matter. Liberty ensures that people can use tools to pursue their own visions of life. Autonomy ensures that those uses are truly chosen and not implanted. Tolerance ensures that many visions can coexist, that the cultural weather remains breathable for dissenters and late bloomers and those out of step with the day’s fashions. Together, they amount to a moral environment within which AI can do what tools at their best have always done: make more things possible for more people.
A “perfect life” will always be aspirational, a horizon we approach but never fix in place. It is useful anyway, because it keeps us honest about ends rather than means. A society that confuses efficiency with flourishing will build efficient systems that feel like cages. A society that confuses safety with flourishing will build padded rooms. The liberal answer has never been to ignore efficiency or safety, but to fit them to the shape of dignity. The test for any new technology is the same: does it enlarge the domain of meaningful choice and the capacity to make it? If not, why are we building it?
In concrete terms, this answer yields a simple checklist. Systems that affect rights must be explainable and contestable. Data practices must give real control to people, not ritualistic consent boxes that disguise extraction as choice. Audits and monitoring must be routine, not emergency measures after scandal. Regulators should have the expertise and be given the authority to look under the hood, and the public should be able to see what they find. Procurement should favor designs that minimize data, preserve privacy, and allow substitution so that no one firm can hold whole sectors hostage. Public investment should flow to research on safety, robustness, continuous monitoring, privacy, and alignment with human values, not only to performance benchmarks. And civic education should treat AI as a standing topic befitting its nature as general purpose and dual-use technology not a one-off unit in a technology class.
There is also a personal dimension that policy cannot supply. Each of us will have to decide how much of our thinking to outsource and when to insist on the slow work of judgment. Each of us will have to practice attention in an attention market, to build small habits of refusal that keep us from being steamrolled by design. Each of us will have to remember that friction can be a feature. Sometimes what makes a choice meaningful is that it took time, or effort, or the courage to stand apart from the feed. The perfect life, if it exists at all, consists less in the polish of our tools than in the seriousness of our choosing.
The greatest risk in the age of AI is not that machines will become persons. It is that persons will be treated more like machines: predictable, programmable, nudgeable into compliance. The remedy is what it has always been: institutions that restrain power concentration, cultures that reward argument over obedience, and daily cultivation of practices that keep conscience awake. Governments and companies will do what incentives push them to do. Citizens decide what those incentives are. That is the promise of liberty and the constant labor of democracy.
When we ask whether AI can help us live the perfect life, we should answer: It can, if we endeavor to make it so. AI can help when it is bent to human purposes and bounded by the human values of liberty, autonomy, and tolerance. It cannot tell us what to value, and it should not be allowed to make us forget the ethical structures that got our society this far.
The measure is simple enough to write on a note above any engineer’s desk, any policymaker’s inbox, any user’s screen: does this system leave people freer, more self-directed, and more able to live together in difference? If it does, build it and welcome it. If it doesn’t, fix it or refuse it. That is not a Luddite veto. It is a liberal one. And it is the difference between an age of intelligent machines and an age made more human by their presence.