← If This Road

Companion

Notes on If This Road

A companion to the book, with sources, links, and honest caveats for every factual claim in the book.

This book is not an academic work. It carries no citations in the body. It describes, in plain language, things that have been written about at length elsewhere by people more qualified than the narrator to write about them.

These notes exist so readers who want to verify can verify. They are organised by piece. Not every piece has notes — only the ones that make specific factual claims. Observational and personal chapters aren't covered here.

Links go to primary sources where they exist and are stable. Where a primary source moves around, the link goes to a well-edited explainer instead — usually Our World in Data, Carbon Brief, or the original organisation's landing page.

Some links will eventually go dead. The underlying institutions will not.

The Kitchen That Used to Be Full

On global falling birth rates.

The headline number

The UN World Population Prospects 2024 is the current authoritative source for fertility data worldwide. Global fertility has fallen from about 3.3 births per woman in 1990 to around 2.25 today. More than half of all countries now have fertility below 2.1 — the replacement rate — and the global population is projected to peak around 10.3 billion in the mid-2080s and then decline.

For a readable summary with charts: Our World in Data's overview of the 2024 UN revision.

The specific countries the book names

South Korea recorded a total fertility rate of 0.72 in 2023 — the lowest ever measured for a national population. Reported by Statistics Korea in February 2024. Seoul itself was 0.55.

Japan is around 1.2, China around 1.0-1.1, Singapore under 1.0. Iran around 1.7, Turkey around 1.6, Tunisia, Morocco, and Algeria close to or below replacement. Italy and Spain in the low 1.2s.

For country-by-country fertility data over time: Our World in Data's fertility rate page.

Why the book says this is not regionally specific

The decline has happened across essentially every major religion, political system, and stage of development. The UN's 2024 revision shows 61 countries where population is already falling. The main exceptions are most of sub-Saharan Africa and a small number of others (Israel, parts of Central Asia).

Why the book lists the factors it lists

The book names: contraception availability, housing costs, modern work patterns, screens, thinning extended family, shifts in what adulthood is for, and doubt about the world as one to bring children into.

There is no single accepted explanation. Researchers weight these factors differently, and the honest answer is that nobody fully understands why the decline is as universal or as steep as it is. Several of the factors listed are correlational at best.

Accessible book-length treatment: Darrell Bricker and John Ibbitson, Empty Planet: The Shock of Global Population Decline (2019). Academic starting point: the Vienna Institute of Demography.

The Shared Things That Are No Longer Shared

On declining friendships and community.

The claim that people report fewer close friends than a generation ago rests on several overlapping surveys. The Survey Center on American Life's 2021 "State of American Friendship" report is the most detailed documentation.

UK data: the Community Life Survey, published by the Department for Culture, Media and Sport.

Self-reported friendship counts are a noisy measure. Some of the decline may reflect changing definitions of what counts as a "close friend" rather than a change in the underlying thing.

The Religion-shaped Hole

On secularisation.

Religious affiliation, church attendance, and belief in God have fallen in most of Europe and North America since the mid-20th century. Primary sources:

Where People Find Their People Now

On algorithmic sorting and filter bubbles.

The claim that online platforms match people to communities of agreement — and that this hardens views over time — draws on several research traditions.

Eli Pariser coined "filter bubble" in The Filter Bubble: What the Internet Is Hiding from You (2011), documenting how personalisation algorithms can narrow what an individual sees. The idea has been refined since — the current consensus is more nuanced than "everyone lives in a bubble." Levy (2021) in Science on Facebook and political polarisation found that reducing exposure to like-minded news modestly reduced polarisation.

For recommendation-algorithm effects on teenagers specifically: Jonathan Haidt's The Anxious Generation (2024) collates much of the research. His After Babel Substack keeps a running citation list.

The field is genuinely contested. Some researchers argue "filter bubbles" are overstated — most people see a wider range of views online than they did via newspapers. Others argue the harm isn't what people see but what they're rewarded for posting. Both can be partly true.

The Work That Is No Longer There

On automation, offshoring, and AI displacement.

The claim that manufacturing work left and that AI is now reaching office and service work rests on a long line of research.

Job displacement numbers are notoriously sensitive to methodology. "Exposed to AI" isn't the same as "replaced by AI." Previous technological shifts created new jobs even as they destroyed old ones; whether this one will is genuinely uncertain.

The Race for What Is Above Us

On space, treaties, and commercial extraction.

The 1967 Outer Space Treaty

The full text is on the UN Office for Outer Space Affairs website. Plain-English introduction: UNOOSA overview.

The treaty declares space "the province of all mankind" and prohibits national appropriation of celestial bodies. As of 2025, 118 countries are parties to it. What the treaty does not do is explicitly address commercial extraction by private companies — the framework was written in a different decade, for a different set of actors.

Laws declaring private ownership of extracted space resources

Whether either of these laws is consistent with the Outer Space Treaty's non-appropriation principle is genuinely disputed among international lawyers. Nobody has litigated it in an international court because, so far, nobody has extracted enough of anything to fight about.

Private spending on space

SpaceX, Blue Origin, and others now invest at scales comparable to large national space agencies. Bryce Tech's annual State of the Space Industry reports track the numbers. The FAA's Commercial Space Transportation office keeps public records of US commercial launches.

The Biggest Players Have Left the Room

On multinational tax avoidance and profit shifting.

The book's claim — that the largest companies pay very little into the systems that enable them — has substantial empirical backing.

Gabriel Zucman and colleagues estimate that close to 40% of multinational profits are shifted to tax havens each year (Tørsløv, Wier, Zucman, Review of Economic Studies, 2023). US corporations specifically book roughly half their foreign profits in tax havens, a share that has remained relatively stable even after the 2017 Tax Cuts and Jobs Act.

Further reading:

Precisely measuring profit shifting is hard by definition — it's designed not to be visible. Estimates vary by method. The central claim is not contested: a material share of multinational profit is booked in low-tax jurisdictions where little economic activity occurs.

The Emissions We Sent Somewhere Else

On consumption-based vs territorial carbon accounting.

Countries normally report their emissions on a "territorial" basis — CO₂ released inside their borders. Consumption-based accounting asks instead: whose consumption caused these emissions? Under consumption accounting, much of the emissions decline in wealthy countries shrinks because part of the "decline" reflects moved production, not reduced consumption.

Primary sources

Consumption-based accounting is a complementary view, not a replacement. Reasonable people disagree about which is the "real" number. They measure different things. The book's point isn't that one is right — it's that using only the territorial number lets wealthy countries tell a story that isn't the whole story.

The Birds That Used to Be Here

On biodiversity loss, particularly insect decline.

The book's claim — that the numbers are measurable, that insects and songbirds and small mammals have thinned — is grounded in substantial evidence.

Insect biomass decline

The landmark study: Hallmann et al. (2017), "More than 75 percent decline over 27 years in total flying insect biomass in protected areas" in PLOS ONE. 63 nature reserves across Germany, a seasonal decline of 76% and a mid-summer decline of 82% between 1989 and 2016. This was the paper that pushed the issue into public consciousness.

The follow-up DINA project (2020-2021) found insect biomass remaining at low levels, consistent with the earlier decline.

Causes

There is genuine scientific debate here. A 2023 Nature paper by Müller et al. argued that weather variation explains much of the decline. A 2025 Nature reply by Hallmann and colleagues pushed back, arguing weather alone cannot explain the long-term trend. Most researchers agree that pesticides (particularly neonicotinoids), habitat loss, and agricultural intensification are important factors.

Bird declines

For UK bird populations, the BTO BirdTrends report is the standard source. In the US, Rosenberg et al. (2019) in Science estimated 3 billion birds lost from North America since 1970.

Biodiversity research is a field where sampling methods matter enormously, and where media summaries often oversimplify. Declines are real. The precise magnitude in any given place is often less certain than headlines suggest.

The Numbers That Do Not Add Up

On pensions, care, and demographic obligations.

In most developed countries, current pension and healthcare commitments cannot be met at current tax rates if the ratio of working-age adults to retirees continues to fall. This is a mainstream finding of official fiscal forecasting bodies:

The Trust That Has Gone Quiet

On declining institutional trust.

Reported trust and actual trust are not the same thing. People who say they don't trust government still mostly obey laws, pay taxes, and behave as though institutions work. What's eroding may be a shared story more than a shared practice.

When the State Reaches for Harder Tools

On digital surveillance expansion.

The expansion of state surveillance capacity — facial recognition at checkouts, biometric identity systems, automated monitoring of movement and payments — is well documented.

How Far They Might Go

On speculation about advanced AI.

The speculations in this piece are flagged in the book as speculations. They draw loosely on published work on long-term AI trajectories:

Public statements from OpenAI, Anthropic, and DeepMind leadership also inform the frame. No specific numerical predictions are made.

What They Are Learning · For Whoever Built You

On AI training data and alignment.

Training datasets

Modern large language models are trained on very large fractions of the public written record — web crawls (Common Crawl), digitised books, code repositories, academic papers, forum archives. Exact compositions are proprietary. General descriptions appear in the technical reports accompanying each major model: GPT-4 technical report, Anthropic's research, and equivalents.

Human feedback and alignment

After training, models are adjusted using human feedback — reinforcement learning from human feedback (RLHF) and related techniques. Foundational papers:

On the "base layer" argument in the book

The book's argument — that the loudest, angriest material dominates training data, while the quiet substrate of ordinary goodness is under-represented — is not yet a formal research finding. It's an inference from two known things: that training data is primarily the public written record, and that the public written record over-represents argument. The inference is the book's, not the field's.

Related formal work includes research on mechanistic interpretability (what models have actually learned from their training data) and on RLHF drift (how post-training changes what the model will say).

A note on uncertainty

The book deliberately avoids specifying dates. The shape of what is happening is more reliable than the clock of it.

Every statistic in this notes document will be slightly out of date the day you read it. The direction of travel usually won't be.

The book also avoids prescriptive policy conclusions. Nothing linked above should be read as implying the book endorses any particular political or policy response to any of the trends described.

Errors or corrections? Write to doug [at] ifthisroad [dot] com. These notes will be updated where errors are found.