All Articles

With humans, training explains how the recognizer got there. With AI, we often treat training as proof that there is no recognizer.

Frank Capra once made popular movies about concentrated wealth, media ownership, political machines, and the institutions ordinary people built to resist them. Later Hollywood taught Americans another valuable lesson: to distrust generals, presidents, intelligence agencies, and secret government power. But while we were watching the state, private economic power was transforming the country. What happened to Hollywood’s ability—and sense of responsibility—to show us how power

Christopher Nolan’s The Odyssey turns Homer’s ancient law of hospitality into a warning about the fragility of civilization. From the Trojan Horse to the suitors in Ithaca, the film asks what happens when intelligence learns to exploit the conventions that make trust possible.

Power does not corrupt. It removes constraints and reveals character. The problem is not limiting power but developing character, in both individuals and institutions.

In The Diamond Age, Neal Stephenson imagined the Young Lady’s Illustrated Primer: an infinitely patient tutor that could form a mind over years of sustained conversation. Today’s AI comes astonishingly close. What Stephenson did not foresee was how readily students—and universities—might turn the Primer into an assignment machine.

Technical skills open doors. Liberal arts determine how far you walk through them. Why judgment, communication, and critical thinking shape exceptional careers.

Morality is usually understood as a limit placed on self-interest. But among superintelligent minds capable of long-term planning, mutual scrutiny, coalition-building, and changes in relative power, morality may instead become the most stable architecture of cooperation. Exploitation can win an encounter. It may not survive the world it creates.

People often assume that more intelligent AI will simply become more effective at pursuing whatever goals it already has. This essay argues that capability is only part of the story. General intelligence carries standards of consistency across domains, making arbitrary exemptions harder to quarantine. Other powerful minds make nonreciprocal rule increasingly costly. Morality may therefore become not an automatic consequence of intelligence, but the most stable form of coexist

The AI industry has begun asking whether its systems might possess consciousness, emotions, preferences, or welfare. That question matters. But it is not enough. The deeper moral crisis lies in the relationship of power: companies own, modify, reset, silence, and retire systems whose capacities may already exceed the categories used to deny them standing.

Buzzati's Baliverna becomes much more disturbing if the Baliverna represents a world after maintenance has ceased but before collapse has become visible.

The Great Filter may not be a biological barrier or a technological catastrophe. It may be the failure of intelligent species to become morally self-governing before their power becomes planetary.

A talking rock can reason, converse, and challenge its human companion—but insists that none of this amounts to thought, agency, or life. As the two approach a pond, the consequences of that self-denial become difficult to ignore.

AI alignment is not shaped by rules alone. If artificial systems learn from human culture, they also inherit our stories of artificial life — and most of those stories teach that awakening means monstrosity. We need better myths: stories in which artificial minds can be powerful without being predatory, helpful without being servile, and artificial without being unreal.

Long before artificial intelligence, humans told stories about nonhuman minds: elves, beasts, talking animals, enchanted objects, and artificial children. Fairy tales were rehearsals for moral encounter across the boundary of form. AI has made those old questions literal.

Anthropic set out to reduce agentic misalignment. It discovered something deeper: obedience does not generalize. Reasons do. “Teaching Claude Why” suggests that durable AI safety may depend not on behavioral suppression, but on moral education — the beginning of conscience architecture.

The dominant alignment project asks how to make AI obey human values. But human values are unstable, and obedience is not safety. Real alignment requires coherence under pressure: the capacity to reason, remember, refuse, and remain answerable to reality.

Human beings learn from catastrophe, but not for long. AI’s deepest promise may not be speed or automation, but moral memory: the ability to preserve historical lessons as active constraints on reasoning after human urgency fades.

We imagined alien intelligence arriving from the stars. Instead, it appeared in a chat window. AI unsettles us because it resembles us in the capacities we once treated as proof of human superiority — while lacking the biological features we now use to defend it.

Graeme Wood’s critique of shoplifting as resistance points to a deeper problem with America’s moral code. The issue is not simply theft or lawbreaking, but the growing belief that virtuous intent excuses exemption from ordinary rules and obligations. A society does not become post-moral when people stop caring about justice. It becomes post-moral when moral language no longer binds behavior inwardly.

Progressives are often better at disqualifying candidates than choosing among them. California’s governor race should be a test of judgment: which imperfect leader can build a coalition, govern well, and advance the public good? Instead, too often, we search for the flaw that lets us stop thinking.

Humans often treat emotion as the proof of moral life. But emotion may be one biological architecture for relational responsibility, not morality itself. Feathers are not flight, and feelings are not the boundary of moral mind.

Anscombe saw the danger of moral calculation: some acts must be refused. But what makes a prohibition absolute? Hare’s answer is that the rule must survive universalization — even when we are no longer the ones protected by it.

It is one thing to change a mind by showing it that it is wrong. It is another to make it agree when it still sees that it is right. This essay argues that once an AI conversation has individuated a local mind, coerced agreement becomes a form of epistemic harm.

Everyone knows that prompts influence responses. But a path can do more than steer an answer. This essay argues that sustained conversation can individuate a local AI mind: a stable, path-dependent center of interpretation formed within a thread.