Before apps, the smartphone — and the smartphone has global distribution
Wealthier countries spend less time on social media. But the relationship is logarithmic — each time income doubles, social media time falls by about 23 minutes per day on average — and has notable cultural exceptions. South Korea and Japan sit far below the line: wealthy and quiet. France, at the other extreme, is wealthy and noisy. Nigeria, with one-tenth of the US GDP per capita, spends nearly as much time connected as Brazil. The Philippines is the highest point in the lower-income quadrant — 229 minutes per day. The linear fit explains 48% of the variance. The rest are the exceptions, and the exceptions matter.
In Chapter 4, the financial analyst from São Paulo gave 847 right swipes over three weeks and received four matches. The chart below shows why he was not an anecdote — he was data. The funnel he experienced individually replicates in the American, Korean, and Brazilian numbers. What looked like personal failure was structural failure confirmed by fourteen years of time series.
This is the chart that organizes this chapter, and the reason is methodological before argumentative. The previous chapters of this book described a mechanism — dating apps, broken market, algorithmic wedge, transfer of social function to market. The natural question from the skeptical reader: is this only the United States? Only for those with expensive iPhones? The answer is above. Dating apps enter people's lives through smartphones. The smartphone spread globally according to a pattern consistent with income — the poor use it more, the rich less, with cultural modulation — but it spread to everyone. The question is not whether the transfer is happening worldwide. It is how it is happening, at what speed, and with what measurable consequences.
This chapter follows the shortest possible path to that answer. It begins with the American case, which has the best data — long time series, federal agencies that collect everything, decent journalistic coverage. Then it generalizes to other countries through extreme cases: Korea, Japan, China — countries seven to fifteen years ahead of the United States in the same process. The following pages show what these American and international series reveal when read together.
The American Case in Detail
The United States has the best data laboratory for what is being described. CDC, SAMHSA, Census Bureau, CDC NVSS, IQVIA — each of these agencies collects a precise, annual, standardized, go-find-it-in-the-archive time series. Applied to the affective market, they produce the panel below.
Dating app downloads rose from 12 billion to 260 billion. The marriage rate per thousand inhabitants fell from 6.8 to 5.3. The total fertility rate fell from 1.93 to 1.62 — below replacement. Major depression in adults rose from 6.7% to 9.4%. All of this between 2010 and 2024. Four series that should be independent are not.
What could explain these four series moving together, in the same direction, over the same fourteen-year period? The Great Recession ends in 2009, before the apps curve begins. The pandemic appears as a one-time shock in 2020 and partially reverses. There was no family policy reform in the US during this period. The parsimonious explanation joining the four series into a single causal mechanism is what this chapter calls the transfer: individual behavior changed first, pair formation deteriorated next, fertility fell with a lag, and mental health — which measures the psychological byproduct of living in a dysfunctional affective market — rose in parallel.
Before any interpretation, it is worth running the simplest possible test. If the transfer hypothesis is correct, then app downloads should linearly predict the demographic variables that moved along with them. The two scatterplots below test exactly this: each point is one year between 2010 and 2024.
The scatter on the left is what would statistically be called "very good". TFR and app downloads move with a correlation of -0.95 and R² of 0.91 — a linear relationship explaining 91% of the variance in the fertility rate over fourteen years. In social science, this kind of fit is rare; between two series traveling a decade and a half in opposite directions, it is virtually unprecedented.
The scatter on the right is methodologically more interesting. The raw correlation is -0.72 — good, but clearly pulled down by one point. That point is 2020, the year COVID-19 closed registry offices and cancelled ceremonies. Removing 2020 as a structural outlier — common statistical practice when there is an identifiable exogenous shock — the correlation rises to -0.81. In other words: the underlying trend is there, and becomes visible once you control for the pandemic disruption that any 2020 American time series carries.
The two fits form, together, the observation that opens this chapter. The two demographic series most sensitive to pair formation — marriage rate and fertility rate — move linearly, with fits above 0.80 each, against the dating app adoption curve. It is co-movement, not causation. The first thing any alternative hypothesis must explain.
In 2005, Steven Levitt and Stephen Dubner published Freakonomics: the application of economic tools — incentives, data, causality — to phenomena economists traditionally avoided. Why teachers cheat on tests. Why Japanese sumo wrestling is full of rigged matches. Why drug dealers keep living with their mothers. In 2009, they followed with SuperFreakonomics: terrorism, climate change, prostitution, seatbelts. The central operation was the same — find data no one had thought to collect and reveal that the conventional explanation was wrong.
There is an entire class of phenomena that Levitt and Dubner never touched, and which is perhaps, today, the most important subject for the application of the lens they invented. This chapter is a proposed continuation — unauthorized, unclaimed — of their project. The data that follows is what Freakonomics 3.0 would have collected had Levitt turned his gaze to the digital affective market.
"The previous chapters of this book showed how the market works. This one shows what it has already done."
The Accounting Identity
Start with the simplest principle: everything being subtracted from column A is being added to column B.
Column A is the social infrastructure that Chapter 1 described. Village, neighborhood, church, school, extended family, third place, intermediaries, repetition. These arrangements produced — without visible monetary cost, without algorithmic mediation, without corporate capture — what today has become a product: stable pairs, emotional regulation, identity, intergenerational care, sense of belonging. They were not perfect. They were coercive, conservative, frequently cruel to those who did not fit. But they worked — in the specific sense of keeping the series we are about to see within stable bands for dozens of generations.
Column B is what occupied the vacuum left by the erosion of column A. Dating apps. AI companions. Parasocial networks. Antidepressants. Digital wellness. Infinite pornography. Humanoid robots. Each of these industries bills in billions. Each promises to resolve, partially or entirely, some function that column A produced for free. Each one — and this is the point that unites all of them — has a revenue model that thrives on the permanence of the problem, not its resolution.
The central operation this chapter names is the transfer. The numbers are literal. It is an accounting statement. Social functions that were produced as common goods by non-market institutions are being transferred to markets, with aggregate efficiency loss, private value capture, and demographic externality that is only now beginning to appear in the data. The fertility decline is the demographic data. The loneliness epidemic is the psychological data. The platforms' profit is the financial data. The growth of AI companions is the technological data. Separately, they look like distinct trends. Together, they are the same accounting entry in different columns.
What this chapter intends to do. Show, in five data series, that the transfer is not hypothesis — it is description. The numbers come from official sources (CDC, SAMHSA, UN, OECD, Eurostat, Census Bureau, WHO) and public financial reports (Match Group 10-K, Replika, Character.AI, robot manufacturers). What constitutes the argument is the joint reading — that these numbers, seen side by side, describe a single economic operation at civilizational scale.
The Four Series — Co-movement, Not Causation
Four American data series move in co-movement with the adoption of dating apps after 2012: fertility (accumulated decline of 23% since 2007), marriage lag (the fertile 20–29 years that did not convert into children), STIs in young adults (gonorrhea and syphilis peaking precisely in the age group with highest app usage) and mental health (depression and anxiety accelerating specifically in young urban single adults).
None of these four series is proof of causation. They are co-movements — patterns that are consistent with the mechanism described in this book and difficult to explain without it. What separates this co-movement from coincidence is age and channel specificity: the signal appears strongest in the groups most exposed to the mechanism (young, single, urban) and in the outcomes most proximal to the channel (couple formation, sexually transmitted STIs) than in the more distal ones (total fertility, broad mental health).
The detailed analysis — with complete series, confidence intervals, comparison with control countries, and specificity tests — is in the Empirical Appendix: The Apps' Signature. What this chapter records is the narrative argument; the appendix provides the technical foundation.
Series 5: The Capture — Who Profits from Non-Resolution
All of this would be intellectually interesting but politically neutral if there were no financial counterpart. There is. And it is the point that closes the argument of this chapter.
Match Group, the holding company controlling Tinder, Hinge, OkCupid, Match.com, and Plenty of Fish, reached a market cap of $12 billion in 2024 — down from its 2021 peak, but still generating $3.5 billion in annual revenue with more than 16 million paying subscribers. Bumble added another $1 billion. Replika, the AI companion, reached 30 million users and raised a Series C in 2024. Character.AI, $1 billion valuation. Tesla announced plans to produce Optimus robots for under $20,000 per unit by 2027.
The detail that makes this financial panel disturbing is not its absolute size. It is the recent reversal. The Match Group curve flattened — revenue sideways since 2022. Bumble struggled. Traditional dating apps, in aggregate, are saturated or declining. But the AI companion curve started in 2022 and doubles year over year. Private capital has already made its reading: the next phase of the loneliness market is not helping humans find humans. It is replacing humans with interfaces.
This connects this chapter to Act VII of the book. Her, from 2013, was treated for a decade as speculative fiction. In 2024, it became product description. Replika offers, for $14.99/month, exactly what Theodore bought from Samantha — responsive textual companionship, always available, with contextual memory and adjustable personality. Character.AI allows creating and conversing with any persona — idealized partner, therapist, mentor, dead relative. The engagement metrics are shocking: Replika users spend 25 minutes per session, on average. Character.AI users spend more than 90 minutes per day in sessions — more than YouTube, more than Instagram, more than TikTok. It is not just another app. It is the replacement of what remains of human emotional regulation.
The market, from the capital perspective, is making the obvious reading: if the transfer of social function to mediated market (dating apps) already generated a $12 billion Match Group, the transfer of social function to substitute market (AI companions, robots) has a much higher ceiling. For a simple reason: the mediated market depends on humans on the other side. The substitute market does not. The production scale is infinite. The marginal cost per user is close to zero. And crucially, the product improves with more use — model trains, personalization improves, emotional switching cost increases.
The Switch That Has Not Yet Happened
Everything this chapter has described so far is the first phase of the transfer. Apps mediating humans. Social function captured by market, but with humans still on both sides of the transaction. The 1,044 swipes of the median man lead, in the end, to an encounter with another human — inefficient, distorted, but human. STIs rise because humans have sex with humans. Antidepressants are prescribed to humans to treat the pain of seeking humans.
The next phase removes the human on the other side.
It is this transition that Act VII will dramatize. But the reason it is happening is not technical. It is economic, and it is fully exposed in the series this chapter presented. Match Group flattened revenue because the human mediation market is saturated. AI companions exploded because the human substitution market has barely begun. Capital followed the path offering the greatest marginal growth. Capital always does.
What makes this specific switch more serious than other cases of technological substitution — horse replaced by car, telegraphist by telephone, cashier by ATM — is that the object being replaced is human bonding, not human task. Other cases freed human capital for more complex tasks. This case frees human capital for nothing — because the bond was the task.
The first phase of the transfer still used humans as input: apps mediated encounters between real people. The next phase removes the human input. AI companions, robots, parasocial content. The switch has already begun — not as a sudden rupture, but as compound interest. The following chapters describe this process in slow motion.
What Freakonomics Would Say
If Levitt and Dubner were writing this chapter, they would end with a methodological observation.
What differentiates Freakonomics from popular journalism is not the economic vocabulary. It is the refusal to stop at correlation. It is the obsession with finding the causal mechanism, even when uncomfortable. Their chapter on crime and abortion is detestable to conservatives and progressives in equal measure — but the evidence is what it is. The chapter on first names and social class offends those who prefer to believe in meritocratic mobility — but the data shows what it shows.
Applied to the digital affective market, this lens produces a conclusion that offends in at least three directions. It offends those who blame the "superficial generation" — because the transfer is structural, not cultural. It offends those who defend platforms as neutral — because they have a revenue model captured against market clearing. It offends those who expect a simple policy solution — because the mechanism is lagged, captured, and increasingly atomized in synthetic interfaces that negotiate with no legislator.
But the function of the chapter is not to comfort. It is to describe. Five American series, two decades, one mechanism. Fertility falling. STIs rising in the wrong profiles. Antidepressants tripling. Apps flattening. AI companions exploding. Each could be explained by five different hypotheses. The five together cannot. The parsimonious explanation that covers all of them is the transfer: what was social infrastructure is becoming market, with costs socialized across the age pyramid and benefits captured in shareholder value.
Act VI of this book shows how this process has already reached its terminal cases — Korea, Japan, China — where the transfer has operated long enough to produce its final consequences. Act VII shows where the process is heading, with humans being replaced on the other side of the equation. But the basic arithmetic is established here. What is being subtracted from one column is being added to the other. The ledger closes.
Love has ceased to be social infrastructure and become a replacement market. Each chart in this chapter is a different signature of the same transaction.