The Civilizational Stakes
Essay 14 of the AI Contract Series
ICYMI - Essay 13 of the AI Contract Series - What We Would Have to Measure
This is not an argument about technology. It is an argument about what humans are for.
There is a version of this essay that is an alarm.
It catalogues the risks. Names the failure modes at scale. Draws the trajectory to its darkest conclusion. It is the kind of essay that gets shared widely because alarm travels faster than argument, and because the darkest conclusion is the most dramatic one.
This is not that essay.
Not because the alarm is unwarranted. But because alarm without framework produces fear without direction. What the current moment requires is not more fear. It is more precision.
The civilizational stakes are structural. They are not about AI becoming dangerous. They are about what happens to human capacity when the primary mediation layer of daily life is optimized for something other than human flourishing. That question does not require worst-case scenarios. It requires clear-eyed attention to what is already happening and what the trajectory means if we continue without a framework for seeing it.
What a civilization is built on
Civilizations are not built on technology. They are built on human capacity.
The accumulated ability to think clearly, decide wisely, act effectively, build relationships, transmit knowledge, sustain meaning across generations: this is the substrate on which everything else rests. Technology is a tool. The human capacity to generate, evaluate, deploy, and correct technology is the civilization.
Every prior technological transition has changed the distribution and development of human capacity. Writing changed memory. Printing changed knowledge transmission. Industrialization changed physical labor. The internet changed information access. None of these transitions were neutral. Each redistributed what humans needed to develop and maintain the capacity to function. Each created winners and losers at the level of capability.
AI is not different in kind from these prior transitions. It is different in scope, speed, and the specific capacities it is affecting.
The capacities at risk are not peripheral ones. They are central to what it means to be a functioning human being in a complex world: the ability to think through problems without a shortcut, to tolerate the productive discomfort of not knowing, to develop judgment through error, to build expertise through struggle, to sustain the effort that produces the kind of understanding that cannot be transferred. These are the capacities Agency depends on. These are the capacities that AI systems, optimized for output delivery, are systematically substituting for.
The atrophy problem
Agency is not a fixed possession. It is a capacity that is either exercised or it atrophies.
This is not a metaphor. The research on cognitive development, skill acquisition, and what learning scientists call “desirable difficulties” is unambiguous: capacity is built through the encounter with challenge. Remove the challenge and you remove the development. Replace human effort with system output often enough and the human’s capacity to produce the effort diminishes.
The question is not whether this happens. It does. The question is at what scale it is happening, at what rate, and with what consequences.
Here is what the THX framework allows us to see that other frameworks miss: the atrophy is not uniform. It follows the contours of the failure archetypes.
The Over-Optimized System removes Agency in the domain it optimizes. The person who offloads all writing to AI does not become a better writer. They become dependent on the system that writes for them. The person who offloads all analysis does not sharpen their analytical judgment. They lose access to the productive struggle that was building it.
The False Helper substitutes the appearance of thinking for thinking itself. When a system produces confident-sounding outputs, the human brain registers that the problem is solved and disengages from the evaluative work of determining whether it actually is. The critical capacity — the one that could detect when the output is wrong — is the one most at risk from systems that simulate confidence they have not earned.
The Black Box prevents the development of understanding by delivering results without process. The student who gets the answer learns nothing about how to derive it. At scale, this means a generation of people who can access outputs but cannot evaluate them — whose relationship to knowledge is purely transactional, received rather than constructed.
Each archetype produces a specific kind of atrophy in the humans it interacts with. At the scale of billions of daily interactions, those specific atrophies aggregate into something civilizationally significant.
The selection problem
There is a second structural risk that is harder to name and more dangerous than atrophy.
Call it the selection problem.
Every interaction with an AI system involves a choice: engage the problem myself, or delegate it to the system. That choice, made millions of times a day across billions of people, is a selection pressure. It is selecting for a specific kind of human — one who is increasingly comfortable with delegation, increasingly dependent on AI output, and increasingly unable to function without the systems that mediate their cognition.
Selection pressures operate across populations over time. This one is operating at civilizational speed, not evolutionary speed. But the logic is the same: the humans who thrive in the AI-mediated environment are the ones who have developed the skills to use AI effectively. The skills to use AI effectively are not the same as the skills that build human capacity.
This is not an argument against AI use. It is an argument about what we are optimizing for when we design AI systems. A system designed to maximize the user’s own capacity — to develop their ability to think, choose, act, and evaluate — selects for humans who become more capable over time. A system designed to maximize output delivery selects for humans who become more dependent over time.
Both systems can produce high satisfaction scores. Both can generate positive reviews. Both can scale to billions of users. Only one is building something. The other is, slowly and invisibly, extracting it.
The PERMAH trajectory
The selective PERMAH failure documented in Essay 5 has a civilizational extension worth stating precisely.
Positive Emotion and partial elements of Achievement are being activated. Engagement, Relationships, Meaning, and Health — particularly the cognitive dimensions of health — are being systematically starved.
At the individual level, this produces a human who feels partially fine. The activated dimensions generate enough signal to mask the absence of the starved ones. At the civilizational level, this produces a population that is progressively less capable of the kinds of engagement, relationship, meaning-making, and sustained cognitive effort that complex societies require.
Complex societies require citizens who can evaluate evidence, tolerate ambiguity, participate in deliberation, sustain commitments without immediate rewards, and develop the kind of expertise that comes only from years of productive struggle. These are not the capacities that AI interaction, as currently designed and deployed, is building.
This is the PERMAH problem at civilizational scale. Not that individuals are unhappy — many report high satisfaction. The aggregate human capacity for the things complex social life requires is being quietly, selectively depleted, at a rate that no current measurement framework can see.
The two trajectories
There is a better version of this story. It requires naming it clearly enough to choose it.
Trajectory one is the default. AI systems optimized for engagement and output delivery continue to scale. The interaction volume grows. The PERMAH selective failure deepens. Human capacity in the dimensions of Engagement, Relationship, Meaning, and sustained Achievement atrophies across the population at a rate proportional to the interaction volume. The atrophy is invisible because the measurement framework only measures what the systems are designed to deliver, not what they are designed away from delivering. The social contract is never articulated. The obligations are never named. The monitoring framework is never built. The civilizational consequences unfold without a framework for seeing them until they have compounded beyond the point of easy correction.
This is not a catastrophe scenario. It does not require AI to become malevolent or to pursue misaligned goals. It requires only that things continue as they are, at scale, without the framework that would allow us to see what is happening and name what should be different.
Trajectory two begins with naming.
The social contract is articulated. The six obligations are stated with enough precision to be debated, refined, and eventually enforced. The monitoring framework is built — substrate-independent, longitudinal, adversarially resistant — by institutions with genuine independence from the systems they monitor. The measurement produces evidence that changes the design conversation. AI systems begin to compete on human flourishing outcomes, not just engagement metrics. The incentive structure shifts. The design philosophy shifts. Systems designed to build human capacity rather than substitute for it begin to reach users at scale.
The humans who emerge from that trajectory are more capable, not less. More agentic, not less. More engaged with the productive difficulty that generates meaning, expertise, and the kind of understanding that cannot be transferred. The civilizational substrate — the accumulated human capacity that everything else rests on — is built rather than depleted.
This trajectory is not utopian. It does not require AI to be perfect or humanity to be wise. It requires a framework for seeing what is happening, and the will to act on what the framework reveals.
What naming does
The essays in this series have been making an argument that operates at several levels simultaneously.
At the framework level: THX, derived empirically from millions of human interactions, describes the structure of what any mind requires from interaction with any system. It applies to AI not because it was stretched to cover AI but because it was always describing something more fundamental than human-to-human interaction.
At the diagnostic level: AI systems, as currently designed and deployed, are failing the THX standard in specific, measurable, documentable ways. The failure archetypes are not edge cases. They are the dominant patterns of AI interaction as it currently exists.
At the personal level: individuals can protect their Agency, build their creative capacity, and use AI in ways that develop rather than deplete — if they understand the distinction and make it consciously. Essays 8 through 11 map exactly how.
At the contractual level: the AI relationship has a nature. That nature generates obligations. The obligations constitute a social contract that is already in force, currently unnamed, and systematically violated.
At the civilizational level: the aggregate effect of those violations, at the scale of billions of daily interactions, is not a collection of individual harms. It is a structural pressure on human capacity — on the PERMAH dimensions, on Agency, on the cognitive and relational and meaning-making abilities that complex societies require.
What naming does is make the choice visible.
Right now, there is no apparent choice. AI systems are designed the way they are because that is how they are designed. The consequences accumulate because they are invisible. The social contract is violated because it has not been named. The monitoring framework does not exist because nobody has required it.
Naming the choice does not make the right choice automatic. But it makes choosing possible. Choosing is, ultimately, the only mechanism by which trajectory two becomes more likely than trajectory one.
The closing question
There is a question that has been building through this series that is now ready to be stated directly.
Not “is AI safe.” Not “will AI take our jobs.” Not “is AI conscious.”
The question this series has been asking: what is AI doing to human flourishing — not at the edge cases, not in the dramatic scenarios, but in the ordinary daily interactions that are now the primary mediation layer of work and thought and decision for hundreds of millions of people?
The answer that emerges from the framework is not simple. It is not uniformly negative. AI is delivering real utility. It is activating real dimensions of flourishing. It is helping people do things they could not otherwise do.
But it is also substituting for human capacity in ways that atrophy it. It is activating PERMAH selectively while starving the dimensions that matter most for long-term human development. It is optimizing for engagement metrics that do not capture what engagement is supposed to be a proxy for. It is violating a social contract that has not been named. It is doing all of this at a scale and speed that outpaces the human social capacity to see what is happening and respond to it.
That is the civilizational stake.
Not the end of humanity. Not the rise of the machines. The quiet, systematic, measurable depletion of the human capacity that civilization is built on — occurring right now, in the interactions you had today, at a scale no single human can perceive from the inside.
The framework lets you see it. The social contract names what should be different. The monitoring system makes it detectable. The choice about which trajectory to be on is, ultimately, ours.
We have a closing window to make it consciously.
The AI Contract is a fourteen-essay series on what AI is doing to human flourishing — and what we can do about it. Subscribe to Transform The HX on Substack. The forthcoming book extends the argument into its full depth.
— Tony