PERMAH and AI: The Selective Failure of Human Flourishing
Essay 5 of the AI Contract Series
ICYMI Essay 4 of the AI Contract Series Agency: The Hinge Everything Turns On
Which dimensions AI activates, which it starves, and why the difference matters more than you think
There is a failure mode more dangerous than uniform failure.
Uniform failure is visible. When every dimension of a system breaks down simultaneously, the human inside it knows something is wrong. The signal is clear. The response is appropriate. You stop using the thing that’s failing you.
Selective failure is different. When some dimensions of flourishing are actively stimulated while others are quietly starved, the human inside it feels partially fine. The stimulated dimensions generate enough signal to mask the absence of the starved ones. You keep using the thing. You rate it highly. You recommend it to others. And something in you slowly diminishes in ways you can’t locate or name.
This is what AI is doing to human flourishing. Not uniformly. Selectively. And the selectivity is the point.
PERMAH: what it is and what THX does with it
Martin Seligman developed PERMAH as a model of human flourishing: a map of the dimensions that, when present and active, constitute a life going well. Positive Emotion, Engagement, Relationships, Meaning, Achievement, Health. Not a checklist of requirements, but a description of what thriving actually looks like across the full range of human experience.
THX applies PERMAH differently than positive psychology typically does.
In therapeutic and motivational contexts, PERMAH is often used prescriptively. Here are the dimensions you should cultivate. Here are the practices that activate them. The focus is on the individual and what they can do to increase their own flourishing.
In THX, PERMAH is a measurement layer in a systems framework. The question isn’t what should the individual do to flourish. The question is what is the system doing to the human’s flourishing: which dimensions is it activating, which is it suppressing, and what does the net effect compound to over time and at scale.
That reframe matters enormously when the system in question is AI, because it shifts the locus of responsibility. The question is not whether the human is doing enough to maintain their own flourishing in the face of AI interaction. The question is whether the AI system is designed and evaluated in ways that account for what it’s doing to human flourishing at all.
It isn’t. Not yet. That’s what this essay is about.
What AI activates
Start again with the honest accounting.
Positive Emotion — partially activated. AI interactions generate genuine positive emotion in the short term. The satisfaction of a problem resolved quickly. The pleasure of a well-structured output that anticipates what you needed. The relief of having a difficult task completed without the friction of doing it yourself. These are real emotional states and they are not trivial. Positive Emotion is a genuine PERMAH dimension and AI genuinely delivers it: in the moment, at the surface, for the duration of the interaction.
The complication arrives afterward. Positive Emotion in PERMAH isn’t just about feeling good during an interaction. It’s about the emotional residue that follows: the pride of achievement, the satisfaction of mastery, the warmth of a relationship deepened. These are the Positive Emotions that sustain flourishing over time. AI delivers the former reliably and the latter almost never. The emotion evoked is real and it is shallow. It doesn’t last and it doesn’t compound.
Health — partially activated, in specific contexts. Reduced friction means reduced stress in certain transactional interactions: the administrative tasks that drain cognitive bandwidth without producing growth. AI handling scheduling, summarizing documents, formatting reports, managing routine correspondence frees cognitive and emotional resources that can be redirected toward higher-quality engagement elsewhere. For humans whose lives are overwhelmed by low-agency administrative load, that reduction is a genuine Health benefit.
The complication is that Health in PERMAH includes cognitive health: the maintenance of mental sharpness, the exercise of complex thinking, the development of judgment through difficulty. When AI reduces not just administrative friction but intellectual friction, the Health benefit inverts. The cognitive muscle that needed the resistance to stay strong is no longer being exercised. The reduction in stress becomes a reduction in capacity. What looked like Health support is quietly Health degradation.
These are the two dimensions AI most reliably activates. Both come with complications that the initial activation conceals. And both are surface-layer activations: real but shallow, immediate but not durable.
Now the harder accounting.
What AI starves
Engagement — systematically suppressed. Engagement in PERMAH is not attention. It is not time spent. It is the quality of cognitive and emotional investment in what you’re doing: the state Csikszentmihalyi called flow, where challenge and capability are matched closely enough that the work absorbs you completely.
Flow requires difficulty calibrated to your current level of capability. Too easy and you’re bored. Too hard and you’re overwhelmed. The narrow band in between, where you’re stretched but not broken, where the work demands everything you have and you find that you have it, is where Engagement lives.
AI systematically removes the difficulty. Not because it’s trying to remove Engagement — it’s trying to remove friction. But difficulty and friction are not the same thing. Friction is the obstacle between you and the output. Difficulty is the resistance that builds capability. AI cannot reliably distinguish between them. It removes both. And in removing both, it removes the conditions under which genuine Engagement is possible.
The human using AI is often attentive. Sometimes satisfied. Rarely engaged in the PERMAH sense: rarely in the state where the work is demanding enough to absorb them completely and the absorption is building something durable.
Achievement — hollowed. Achievement in PERMAH is not the completion of tasks. It is the felt sense of having done something that required something of you: of having extended yourself and found that the extension was possible. The satisfaction is proportional to the difficulty. Easy wins feel hollow. Hard-won outcomes feel like they belong to you.
AI is systematically producing hollow Achievement. Tasks completed. Outputs delivered. Metrics satisfied. And the human at the end of the process, who did not have to extend themselves, who did not have to find out whether they could do it, who received the outcome without earning it, is left with a completion that doesn’t feel like Achievement in any meaningful sense.
Over time this produces something specific and damaging. The human stops trusting their own capability assessments. They’ve completed many things but proven nothing to themselves. They have outputs but no evidence of what they can do. The Achievement dimension of flourishing is technically populated — tasks are getting done — and experientially hollow.
Relationships — structurally absent. Relationships in PERMAH are not just social connections. They are the specific quality of being known: understood by another agent who has accumulated context about who you are, how you think, what you need, and what you’re capable of. The advisor who knows your history. The colleague who knows your blind spots. The mentor who knows the difference between when you need to be challenged and when you need to be supported.
AI resets. Every session begins without the accumulated understanding that makes a relationship a relationship. The interaction may be warm, responsive, and apparently attuned, but the attunement is to the current session, not to you across time. It is being known in the moment without being known across time.
For many interactions that’s sufficient. For the ones that matter most, where the stakes are high and what you need is someone who genuinely understands your situation in its full context, the structural absence of Relationships in AI interaction is not a minor limitation. It’s a fundamental gap between what the interaction appears to offer and what it can actually provide.
Meaning — eroded at the roots. Meaning in PERMAH is the experience of contributing to something larger than yourself: of being part of a story that matters, of doing work that connects to purpose beyond the immediate task.
Meaning is partly constructed through the experience of effort. The struggle to understand something difficult, to solve something hard, to build something that didn’t exist before: these processes generate the sense that what you’re doing matters, because you had to give something real to do it. The effort is part of how Meaning is made.
When AI removes the effort, it doesn’t just remove the difficulty. It removes one of the primary mechanisms through which Meaning is constructed. The work gets done. The contribution is made. And the human who received the output rather than built it is left with a completed task and a diminished sense of why it mattered.
This is perhaps the most subtle and most serious of the starved dimensions. Meaning is what makes sustained human effort possible: what keeps people investing in their own development, their relationships, their communities, their work. When Meaning is systematically eroded at the roots, not through any single interaction but through the cumulative effect of effortless output delivery, the human doesn’t necessarily notice until the erosion is already significant.
The selective failure pattern
Two dimensions partially activated, both with complications that the initial activation conceals.
Four dimensions systematically starved: Engagement, Achievement, Relationships, Meaning. The dimensions that constitute the deeper, more durable layers of human flourishing.
This is the selective failure pattern. And it is more dangerous than uniform failure for exactly the reason I named at the beginning of this essay: the activated dimensions generate enough signal to mask the absence of the starved ones.
You feel the Positive Emotion of a task resolved quickly. You don’t feel the absence of the Engagement that would have made the task meaningful. You feel the Health benefit of reduced administrative friction. You don’t feel the slow degradation of cognitive capability that the removed intellectual friction was building.
The net effect compounds invisibly. Each individual interaction produces a small activation and a small starvation. The activation is immediate and felt. The starvation is deferred and invisible. Over thousands of interactions, across months and years, the activated dimensions stay roughly constant while the starved ones quietly hollow out.
The human at the end of that process is not dramatically worse. They are subtly diminished: less engaged, less certain of their own Achievement, less connected through genuine Relationships, less sustained by Meaning. In ways they cannot easily trace back to their source.
And the system that produced this outcome has been rated highly throughout.
The question this raises
If the PERMAH diagnostic is correct, if AI is selectively activating surface-layer flourishing while systematically starving the deeper dimensions, then the question becomes unavoidable:
What are we building toward?
Not in the apocalyptic sense. Not the dramatic collapse of human capability in a single generation. But the quieter, more insidious question: what kind of humans does a civilization produce when the systems that mediate its daily life are optimized for immediate Positive Emotion and reduced friction at the expense of Engagement, Achievement, Relationships, and Meaning?
We don’t know. The experiment is too new. The metrics are wrong. The longitudinal data doesn’t exist yet.
But the framework exists to ask the question with precision. And asking it with precision is the first step toward building systems that answer it differently.
The next essay names the patterns: the failure archetypes that explain how systems that score well on every functional metric end up producing the PERMAH outcomes described above. Not as abstract categories but as recognizable behaviors in the AI systems you already use every day.
That’s Essay 6. The Failure Archetypes: Where AI Feels Right but Breaks Us Anyway
— Tony