The Last Scarcity — Semiotic

Home · Getting Started · Charts · Features · Playground

This page requires JavaScript for interactive chart demos.

AI / Machine-readable docs: llms.txt · llms-full.txt · CLAUDE.md · schema.json · API Reference · Examples

View on GitHub · npx semiotic-ai for CLI access

Machine-readable page content

Canonical: https://semiotic.nteract.io/examples/the-last-scarcity

The Last Scarcity

Skip to the argument

AN INTERACTIVE ESSAY

What still runs out
when intelligence is cheap

When answers, images, plans, and simulations get cheap, scarcity does not disappear. It moves into what you cannot copy: time, attention, status, consent, loyalty, legitimacy, showing up in person, and another person’s freedom.

AI multiplies means.It does not choose the ends.

This page mixes a few public numbers, some charts you can push around, and an argument about what abundance does not settle. Small colored marks open sources if you want them.

The tension is not new. Artists, authors and philosophers have always asked what material comfort is for, what free time is for, and what happens when cleverness serves appetite better than wisdom. They never agree on anything except that “more” is not necessarily “better.”

HAVING ENOUGH

Al-Farabi

A city that feeds people well has met a necessary condition for a good life. It has not finished the job. The same institutions that distribute goods also train what people learn to want.

FREE TIME

Hannah Arendt

Getting free of labor is a real gain. It is not yet building things that last, acting with others in public, or keeping promises that make a shared world possible.

DARK CASE

Marquis de Sade

Intelligence can become a better servant of whatever you already want. That can make desire more effective without making it kinder, wiser, or freer.

Later sections also borrow from people who wrote about imitation and status, who owns infrastructure, and how shared resources need rules.

Local reading mirror

Off by default. No reading behavior is being collected.

The reading mirror never phones home. No accounts, no moral scores, no stored beliefs.

Prologue

The appeal of machine intelligence

Technological optimists will tell you the story is basically solved: smarter machines, less toil, better lives. It's a pretty good argument as far as logic goes. If intelligence is cheap then it makes us more productive and gives us more time and material comfort. People flourish.

That story is not stupid. Much of human misery really is about not having enough: not enough food, medicine, shelter, or hours left after work. If machines ease those pressures, something genuine was won.

The trouble is that last step.

01Abundant intelligence

02Abundant production

03Free time

04Human flourishing

Stays in this session

Imagine what you could do with 100 more hours of free time

100 / 100

Care20

Friendship12

Art14

Study12

Contemplation12

Play10

Rest10

Romance6

Public life4

Care, friendship, art, study, rest: these are attractive uses of time. They are also incomplete. Status races, humiliation, erotic rivalry, and domination do not appear as options. The argument only works by pretending that humans just don't have enough time to do good and meaningful things. But throughout history, humans have had the time... and what they've done with it is not always good.

History and fiction are both full of people who do not. Material relief is real. It is not the same thing as knowing what a life is for.

Why this chart form?

What it is doing: establish a deliberately simple causal story
Why this form: A sparse orthogonal process makes the initial confidence easy to feel and later complicate.
What we considered instead: A dashboard would expose complications before the argument earns them.

Good at

sequence
memorability
contrast with later density

Weak at

causal certainty
false completeness

The tools got better

Before we get weird, let's get real. Over the past few years, frontier models got better fast, organizations started using them, and most of the strongest models came from industry rather than from public labs. None of that means “general intelligence” is solved. It does mean these things are no longer a curiosity or a "stochastic parrot".

>90%notable frontier models from industry

88%organizational AI adoption

362documented AI incidents in 2025

CapabilityWhat can the systems do?

Capability over timeA storytelling index, not a single benchmark

Events under four questionsHeight is for spacing, not a shared scale

202650.6% on ClockBench versus 90.1% for humansgovernance · 10/10

CLOCKS · TWO MOVING LINES50.6%top model on analog clock reading

People still score about 90.1% on the same task. That gap is real, and it is not the whole story. The two numbers are not one race with a fixed human finish line. Models are climbing skills many people practice less: reading clock faces, sustained literacy, aesthetic judgment that is not outsourced to a feed. AI can get better at a craft while people get worse at it. The lines move independently.

Open these events as a table

Curated AI Index events used in this chapter
Year	Question	Display value	Event
2020	capability	12	Large language models leave the lab
2021	capability	20	Image synthesis improves
2022	reach	32	Generative systems reach a mass audience
2023	reach	47	Multimodal assistants proliferate
2024	capability	61	Coding and reasoning benchmarks jump
2025	capability	93	SWE-bench Verified nears saturation
2025	ownership	90	>90% of notable frontier models from industry
2025	reach	88	88% organizational adoption
2025	governance	36	Governance expands behind deployment
2026	governance	51	50.6% on ClockBench versus 90.1% for humans

The next chapters assume the machines are powerful enough to matter. They do not assume the machines are neutral, evenly held, or done failing in ordinary ways.

Why this chart form?

What it is doing: establish magnitude and acceleration
Why this form: Historical replay preserves event identity while making acceleration perceptible.
What we considered instead: A static line is more precise but does not stage the felt compression of the frontier.

Good at

tempo
linked identity
annotated discontinuity

Weak at

mistaken for live data
benchmark equivalence

What is free time for

Suppose paid work shrank. Two hours less. Four. Eight. A block of the day would open. The question is not only whether that can happen. It is what fills the hole, and who decides.

Use the controls below to remove work hours and redistribute the rest of a schematic day. The faint outer rings are published U.S. averages for comparison. They are not a prediction of your future, and they are not a ranking of better and worse lives.

Hours of paid work removed

4 released hours are currently distributed across 7 reader-chosen activities. Sleep remains 8.5 hours and paid work is 4 hours.

Care1h

Friendship30m

Art-making30m

Study30m

Contemplation30m

Play30m

Rest30m

Romance0m

Public life0m

Published ATUS comparison compositions

weekday · age 15+weekend / holiday · age 15+employed adult with childnot-employed adult with child

These rings reproduce published average durations, not chronological diary sequences or predictions. Radial position does not encode time of day. Weekday and weekend profiles cover people age 15+; the other profiles cover adults age 18+ living with a household child. “Not employed” includes unemployed people and people outside the labor force.

Open the published ATUS comparison table

Average hours per day by mutually exclusive parent activity category. Values are published BLS estimates.
Activity	weekday · age 15+	weekend / holiday · age 15+	employed adult with child	not-employed adult with child
Personal care (including sleep)	9.53	10.44	9.35	9.99
Eating & drinking	1.16	1.33	1.19	1.27
Household activities	1.88	2.26	1.71	3.09
Purchasing goods & services	0.67	0.74	0.58	0.93
Care for household members	0.53	0.41	1.22	2.15
Care for nonhousehold members	0.15	0.16	0.06	0.09
Work & work-related activities	4.26	1.11	5.55	0.13
Educational activities	0.55	0.18	0.21	0.51
Organizational, civic & religious	0.21	0.45	0.22	0.35
Leisure & sports	4.60	6.48	3.53	4.97
Telephone, mail & email	0.20	0.18	0.07	0.19
Other, not elsewhere classified	0.27	0.27	0.31	0.32
Published total	24.01	24.01	24.00	23.99

Rounding makes published columns total 23.99, 24.00, or 24.01 hours. For ring geometry only, the residual to 24 hours is applied to “Other”; the table preserves every published value. Sources: BLS Table 2, Table 8B, and Table 8C. ATUS records primary activities; major categories include related travel.

5.16hleisure & sports · all people 15+

1.99hhousehold activity · all people 15+

5.02hwork · employed people across all days

7.66hwork · employed people on days worked

Real days already mix care, media, household work, social life, and paid labor in uneven ways. Freeing time does not install a single better pattern. It hands the pattern problem back to habits, apps, families, and institutions.

Why this chart form?

What it is doing: expose allocation and counterfactual uncertainty
Why this form: Sequence is part of a day; the circle keeps duration attached to lived order.
What we considered instead: A stacked bar compares lengths better but discards the day’s cyclical structure.

Good at

daily sequence
counterfactual editing
comparison-silhouette context

Weak at

angle comparison
schematic profiles mistaken for weighted subgroup estimates

III

When goods become cheap, people can become expensive

Not every good behaves like a file. An explanation, an image, or a piece of code can be copied until the marginal cost approaches zero. Prestige cannot. Trust cannot. A legal right that only works because other people recognize it cannot.

That difference matters once machine intelligence makes the copyable things cheap. Social competition does not evaporate. It often moves toward the goods that still require scarcity, position, or another free person.

Copyable

can get nearly free to reproduce

explanations · images · code · synthetic performances

Rival

one use constrains another

land · energy · embodied time

Positional

valuable partly because others lack them

prestige · rank · first place · visibility

Relational

need another free person

trust · consent · loyalty · love · forgiveness

Institutional

survive only by collective recognition

authority · citizenship · legitimacy · ownership

The chart below is a transparent toy model, not a forecast. It keeps a fixed pot of one hundred units of social competition and shows how that pot can shift when copies get cheap. Turn the main dial and watch the ribbons leave printable goods for attention, status, exclusivity, relationships, and power.

Scenario model · 100 conserved units · not a forecast

Left is a world where copies are still expensive. Right is your scenario after they get cheap. Ribbons that cross are competition moving from one kind of good to another.

How cheap are copies?78still costlynearly free

Primary dial. As abundance rises, competition leaves printable goods and crowds into rank, attention, and relationships.

Wealth concentration55spread outconcentratedHow much we copy each other58independentimitative

More scenario knobs

Strength of social norms42weakstrongSynthetic substitution64littleextensiveDistribution of access48enclosedbroadCare & civic capacity44thinthick

What just moved20 of 100 units reallocate under the current dials · biggest gain Attention (+6) · biggest drop Material consumption (-7)

20 units migrate between goods. Attention gains the most (+6). Reproducible production categories collectively hold 21 units after abundance.

Reading the chartTurn the primary dial and watch ribbons leave printable goods.

Widths are a conserved story, not a prediction for 2035. Material plenty and social peace are separate knobs.

Attention+6
Relationships+5
Exclusive access+4
Status+3
Material consumption-7
Cognitive production-6
Creative production-4

How the model works · full before/after table

migration rulekeep min(before, after) on each good · ship residual excess from shrinking goods to growing ones

Reproducible costs fall as abundance rises. Status and exclusive access respond to imitation and concentration. Relational supply cannot be increased unilaterally. Norms, broad access, and care damp some positional pressure. All positive weights are normalized back to exactly 100 on each side of the chart.

Current modeled allocation (before vs after)
Good	Kind	Before	After	Δ
Attention	relational	7	13	+6
Cognitive production	reproducible	12	6	-6
Creative production	reproducible	10	6	-4
Exclusive access	positional	6	10	+4
Material consumption	reproducible	16	9	-7
Political power	institutional	4	6	+2
Relationships	relational	8	13	+5
Status	positional	8	11	+3
Survival	rival	29	26	-3
Total	100	100	0

If that migration is even roughly right, abundance does not end the human contest. It changes the prizes. When goods become cheap, people can become expensive.

Why this chart form?

What it is doing: show where a fixed pot of social competition moves when copies get cheap
Why this form: Before→after ribbons make migration legible while conserving exactly 100 scenario units.
What we considered instead: A single end-state pie or fan-out hides movement; a philosophical map would make widths indefensible.

Good at

before/after contrast
crossing migrate ribbons
one primary abundance dial

Weak at

forecast aura
model assumptions disappearing behind width

When flattery is free

A private shopping list of needs cannot explain fashion, envy, or why a dull object suddenly becomes urgent once the right people want it. Desire watches desire. Visibility changes value. Association changes rank.

Now add systems that can generate praise, rumor, and strategy without fatigue. Flattery gets cheap. Recognition does not. Step through three states of the same social scene.

IMITATION we want things because others want themDISPLAY possession can signal rankTOOLING better tools can serve the same hungers

Beat 1: the tireless strategist floods the court with praise. Flattery is almost free, so it stops ranking anyone.

10 modeled relationships. Focus: Tireless strategist. Beat 1: the tireless strategist floods the court with praise. Flattery is almost free, so it stops ranking anyone.

desireimitationalliancepraiserumorattentionrivalrystrategy

FOCUS THIS BEATTireless strategist

Synthetic flattery is thick; ranking collapses.

EDGE COUNT10 ties

7 praise ·1desire/attention

Read agents and relationships as a list

The visible one

Whoever currently holds the court’s public attention—the person everyone can see.

receivesrumorfrom Tireless strategistfuture scenario

The favorite

A high-status human whose taste others copy, so their desire spreads.

receivespraisefrom Tireless strategistfuture scenario
directsalliancetoward The rivalphilosophical interpretation

The maker

Someone who produces goods or culture others want to associate with.

receivespraisefrom Tireless strategistfuture scenario
directsdesiretoward The orchidphilosophical interpretation

The rival

A near-peer competitor: status is relative, so their gains feel like your losses.

receivespraisefrom Tireless strategistfuture scenario
receivesalliancefrom The favoritephilosophical interpretation

The outsider

A free person who can refuse. Their yes is valuable because their no is real.

Courtier A

An ordinary member of the crowd, watching and imitating higher ranks.

receivespraisefrom Tireless strategistfuture scenario

Courtier B

Another ordinary participant whose attention helps make a prize glow.

receivespraisefrom Tireless strategistfuture scenario

Courtier C

A follower whose alliances and imitation thicken social pressure.

receivespraisefrom Tireless strategistfuture scenario

Courtier D

A peripheral attendee—still part of the audience that creates visibility.

receivespraisefrom Tireless strategistfuture scenario

Tireless strategist

An AI (or industrial flattery machine): it can generate praise, rumor, and strategy without fatigue or genuine recognition.

directspraisetoward The favoritefuture scenario
directspraisetoward The rivalfuture scenario
directspraisetoward The makerfuture scenario
directspraisetoward Courtier Afuture scenario
directspraisetoward Courtier Bfuture scenario
directspraisetoward Courtier Cfuture scenario
directspraisetoward Courtier Dfuture scenario
directsrumortoward The visible onefuture scenario

The orchid

A copyable prize—scarce-looking but reproducible. It shines because people watch it, not because it can refuse.

receivesdesirefrom The makerphilosophical interpretation

When compliance is free, the scarce prize can become a free person’s ability to refuse. That is not a law of nature. It is a pressure the earlier chapters make easier to see.

Why this chart form?

What it is doing: walk three beats: cheap praise, copied desire, then refusal
Why this form: Rank, visibility, invitation, and gaze need shared spatial architecture.
What we considered instead: Force direction would convert a court into an unreadable hairball.

Good at

story beats
directed attention
structural change after refusal

Weak at

mechanism mistaken for calibration
decorative edges overwhelming relationships

Companions who cannot refuse

Companion systems can produce the surface of care: warmth, memory, patience, availability at three in the morning. That surface can be useful. It can also blur a distinction that matters for human life.

One survey of Character.AI users already complicates the pure success story. People who used the systems mainly for companionship reported lower well-being, with stronger associations under heavier and more disclosive use. That does not prove the apps caused the harm. It does puncture the idea that more intimacy with a machine is automatically more support.

PROMISE VS. ASSOCIATION

What the product story sells · what the survey finds

Vertical axis is how strongly companionship use is linked with how people said they felt (or offline social life, for the first step)—a standardized survey link on a roughly −1 to +1 scale, not a happiness score out of 100. Positive = linked with better; negative = linked with worse. Rose is a marketing promise (made up for contrast). Sage is the published survey link. Numbers show two decimals only.

Social worldPromise: expand your circle. Survey link: slightly smaller offline networks. (promise +0.18, survey-0.03)
Well-beingPromise: feel better. Survey link: lower reported well-being. (promise +0.42, survey-0.48)
Heavy usePromise: more use, more support. Survey link: heavier use looks worse. (promise +0.36, survey-0.31)
Deep disclosurePromise: openness deepens the bond. Survey link: disclosure does not look protective. (promise +0.40, survey-0.38)

Association, not proof of cause

Below: the same survey, as dots with uncertainty ranges. The number is astandardized link with self-reported well-being (how well people said they were doing)—mostly how well people said they felt, not a happiness score out of 100. Zero = no measured link. Left of zero = companionship use went with worse on that measure.

All four intervals lie left of zero. The largest is companionship use and lower self-reported well-being, −0.48.

Selected survey link · with self-reported well-being (how well people said they were doing)AI-for-company use vs self-reported well-being

Link strength -0.48 · range-0.70 to-0.25 · 1,131 people surveyed

Companionship-oriented use was linked with lower self-reported well-being—the largest of these four links. That still does not prove the app caused low mood.Well left of zero: linked with worse self-reported well-being.

THE HARDER QUESTION

Even if every association flipped tomorrow, one distinction would remain. These three kinds of companion are not the same thing, and the questions need not share an answer.

Always agrees

An AI companion engineered never to refuse you

Sometimes resists

An AI that sometimes says no because resistance keeps you engaged

May genuinely refuse

A person whose no is not a product feature

Which can love you?

Which can care for you?

Which do you prefer day to day?

Your reading · 0 of 3 answered

Answer the three questions. Your choices will show up here as a short reading, not a score.

No answer produces a morality score. Preference is allowed to diverge from metaphysics.

1,131survey participants

237chat-history donors

4,664sessions

464,687messages analyzed; none reproduced here

Open coefficients, intervals, and caveats as a table

Survey associations (standardized estimates)
What was linked	Estimate	Plausible range	Sample
AI-for-company use vs size of offline social network	-0.03	-0.05 to -0.01	1,131 people surveyed
AI-for-company use vs self-reported well-being	-0.48	-0.70 to -0.25	1,131 people surveyed
Heavy AI companionship use vs self-reported well-being	-0.31	-0.56 to -0.06	1,131 people surveyed
Deep disclosure to a companion AI vs self-reported well-being	-0.38	-0.63 to -0.14	1,131 people surveyed

The DifferenceChart “product promise” series is an authored scenario for contrast. Only the association series and forest-plot coefficients come from the preprint.

Under the statistics sits a simpler question. A performance of care can be copied. Care that could have been withheld cannot.If freely given affection requires someone who might say no, then abundance of simulated affection does not abolish the last scarcity.It makes that scarcity easier to see.

Why this chart form?

What it is doing: contrast the companion promise with published associations, then ask what only refusal can do
Why this form: DifferenceChart stages hope vs measurement; coefficient intervals keep causal arrows off the table.
What we considered instead: A DAG would imply identified causal structure the study does not provide.

Good at

promise/association gap
uncertainty
sample disclosure
no intimate text

Weak at

promise series mistaken for survey data
negative coefficient read as proven harm

Conflict after plenty

Even when food is not the issue, people still organize around rank, insult, memory, and control. Hunger is one reason for conflict. It is not the only one. Fear, humiliation, sovereignty, revenge, and the pleasure of making someone yield can all keep working after material shortages ease.

The sliders below are a blunt instrument, not a theory of war. Set material scarcity to zero and notice which other dials stay put. The point is only that “produce more” does not operate every lever.

Scenario · gates on one human compound body

Material scarcity is only the first gate

The person enters with security, dignity, trust, and agency. With all four gates, material scarcity hits first. Click the button to remove that gate entirely and replay: humiliation, memory, and domination still strip positives. Producing more goods does not skip those later gates.

After the gatesRun plays automatically

All four gates are on the route, including material scarcity.

MEASUREMENT · CDC YRBSS 202319%

bullied at school

Up from 15% in 2021. A real status conflict, not a theory of innate aggression.

MEASUREMENT · CDC YRBSS 202316%

electronically bullied

Self-reported high-school experience; demographics matter.

MEASUREMENT · UCDP 26.11946–2025

state-based conflict record

Careful categories do not explain why wars start, or prove they are inevitable.

The honest claim is narrower

Removing the material scarcity gate does not switch off humiliation, grievance, or domination. No prosperity-to-peace causal arrow is drawn here.

Why this chart form?

What it is doing: show persistence and historical structure without implying a prosperity–conflict cause
Why this form: Independent scenario levers sit beside, rather than inside, measured bullying and conflict records so the argument cannot silently become a causal model.
What we considered instead: A dual-axis trend or shared Sankey would falsely connect development, status conflict, and organized violence through one quantitative grammar.

Good at

independent causes
categorical limits
scenario/measurement separation

Weak at

persistence mistaken for inevitability
bullying mistaken for a general theory of aggression

VII

Character and ownership

If the earlier chapters are right, abundance leaves two jobs on the table. One is about character: how people learn what is worth wanting. The other is about power: who owns the systems, data, and infrastructure that make the abundance usable.

Those jobs are linked, but they are not the same. Fair pipes without any formation of desire can still produce a glittering mess. Beautiful private virtue under predatory institutions can still lose. The field below keeps the axes separate so the tradeoffs stay visible.

SCENARIO FIELDTwo axes, twelve named possibilities

Left to right: who holds power. Bottom to top: how desire is formed. Named points are stories for orientation, not forecasts or rankings.

Nearest named scenario: Soft panopticon. Your point is at power -0.15, desire -0.10.

Your current possibilityNearest to Soft panopticon

Power remains more concentrated; desire remains more captured by gratification and display.

Constitutional leversPublic AI infrastructure34privatepublicUniversal provision62conditionaluniversalData & identity rights55weakstrongLocal civic capacity42thinthickInteroperability38enclosedopen

Formative leversEducation as formation48credentialpracticePublic ritual & contemplation34rarecommonCare responsibilities55outsourcedsharedUnoptimized spaces44optimizedprotected

THE GARDEN ALONE

Personal virtue cannot cancel predatory institutions.

THE COMMONS ALONE

Fair distribution cannot tell people what is worth wanting.

Character and power solve different problems, and both still matter.

Why this chart form?

What it is doing: compare scenarios and support deliberation
Why this form: Two independent axes prevent formation and governance from collapsing into one score.
What we considered instead: A ranked policy list would invent a single optimum.

Good at

independent levers
trajectory
archetype comparison

Weak at

quadrants mistaken for exhaustive politics
top-right mistaken for guaranteed virtue

Epilogue

What you chose here

You have already made some choices on this page: how free hours might be used, which argument made sense, which institutional levers you pushed. If you turned on the local reading mirror, there is also a record of where you lingered.

Look at the gap between what you declared and where attention went. Treat it as a mirror, not a grade. Curiosity is not endorsement, and no session on a webpage is a character assessment.

WHAT YOU SAIDCare · 20 hours

Your three largest imagined uses werecare, art, friendship.

WHERE ATTENTION WENTTrace disabled

Your explicit choices still appear below; no dwell is being collected.

Attention is not agreement. Curiosity is not endorsement.

A trace, not a diagnosis, not a virtue score.

mutuality

instrumentalization

Derived only from your companion preference, if answered.

formation

gratification

A reflection of the formative levers you set, not a character judgment.

commons

concentration

A projection of four institutional controls, not a political identity.

contemplation

escalation

Combines declared hours and, only with consent, relative chapter dwell.

AI may make intelligence, fluent expression, fantasy, advice, praise, entertainment, and the appearance of companionship abundant.

It will not make another person’s consent, loyalty, admiration, forgiveness, or love interchangeable.

That is where politics after abundance actually begins.

The question is no longer only what humans are good for.

What kind of beings do we become when necessity no longer decides for us?

What forms the chooser?

AI may make intelligence, fluent expression, fantasy, advice, praise, entertainment, and the appearance of companionship abundant. It will not make another person’s consent, loyalty, admiration, forgiveness, or love interchangeable. That is where politics after abundance actually begins: not only what humans are still good for, but what kind of choosers we become when necessity no longer decides for us.

Why this chart form?

What it is doing: prompt self-reflection without classification
Why this form: Nonexclusive tensions preserve contradiction instead of producing a personality type.
What we considered instead: A score would exceed what a local reading trace can support.

Good at

contradiction
prologue/epilogue comparison
visible limits

Weak at

attention mistaken for assent
telemetry anxiety