Systems Collapse

These are the systems that manipulate symbols and they are the systems slated for automation.

Thirteen systems, in the order they were named. Each one runs on literal glyphs letters, numbers, marks and each one deploys those glyphs as meaning, used as a weapon. Open a pillar to see both layers, the mechanism doing the work, and where the system stands.

The Staged Show

Tree Psych at the root

Psych branches down to Health, Media / Info, Community, and Energy. Health runs down to Food. Media / Info runs down to Education, Law, Entertainment, and Financial, and Law runs down to Governance. Community runs down to Housing and Labor.

Layer 1

Glyphs it runs on. The actual letters, numbers, and marks the system operates with.

Layer 2

How the glyphs are weaponized. The meaning those marks carry when they land on a person.

Mechanism

The named move. What the system actually does, in plain terms no authors required.

State

Where it stands. One line per system. Cracks on the pillar scale with how hollowed it is.

13 pillars · 1 beam · tap a pillar to open it

The Timeline

1955 – 2055 · the same thirteen, over time

The pillars above are the snapshot. This is the motion: for each system, the counts, prices, and credentials kept rising while outcomes, participation, and confidence fell. The pattern is strong for media, housing, higher education, finance, labor, and community; mixed or partly reversing for health outcomes, suicide, and overdoses in 2023–2024. Figures are a research snapshot (sources read October 23–24, 2026, via web index; not re-verified live). Contested values are flagged where they sit.

0% 20% 40% 60% 80% 1960 1980 2000 2020 Gallup average, 1980s–90s: 40–50% Trust in federal government (Pew) 73% in 1958 → 17% in 2025 Trust in mass media (Gallup) 68–72% in 1972–76 → 28% in 2025 Average confidence, 14 institutions (Gallup) 26% record low in 2023 · 28% in 2025
Trust lags function, then falls faster: the legitimacy declines are steepest after 2000 and again after 2020, decades after the cost and outcome divergences began. Sources: Pew Research Center; Gallup Confidence in Institutions and mass-media trust series. Note: the Gallup average-confidence composite (the “integrity index” shape the shaded 1980s–90s band and the dashed recent line) is a synthesized comparative shape drawn to show the range and direction, not measured data.

Inflection cluster · 1971–1983

The “before” picture is pre-1980. Fiat dollar (1971), the Butz cheap-calorie food era (1971–73), the HMO Act (1973), PATCO (1981), payment-by-code DRGs (1983), sentencing reform (1984), and the productivity–pay split that begins in 1979.

Inflection cluster · 1996–2010

Symbols scale digitally; functions localize and shrink. Telecom Act (1996), Glass-Steagall repeal (1999), test-count accountability under NCLB (2001), social platforms (2004–07), the financial crisis (2008), the ACA and Citizens United (2010).

2020 · accelerant, not origin

Almost every series was already moving for 10–40 years before COVID. The pandemic steepened the media, education, community, mental-health, and participation curves; 2023–24 then produced partial rebounds in health outcomes only.

Where it is not collapse

Life expectancy hit a record 79.0 in 2024 and overdose deaths fell 26.2% the largest annual drop recorded. Incarceration is down from its 2008 peak. Higher-ed confidence ticked up, 36% (2024) → 42% (2025). The chart keeps those reversals in, flagged, rather than smoothing them away.

Automation exposure the forecasters disagree, and the disagreement is the data

These figures measure different things exposure, displacement, technical potential so they are presented as a range, never as one number. The sector gradient is consistent across all of them: highest exposure in finance, law, administration, and media/content; lowest displacement in healthcare, education, and care work.

IMF, Jan 202440% of global jobs exposed; ~60% in advanced economiesOccupation-level AI exposure (ability overlap), not job loss.
Goldman Sachs, Apr 2023300M full-time jobs exposed globally; +7% global GDP over 10 yearsExposure; most exposed jobs are partly, not wholly, automatable.
Goldman Sachs update, Apr 2026 (as reported)Only ~2.5% of US employment at direct elimination riskElimination risk an order of magnitude below the “exposure” headlines.
WEF Future of Jobs 202592M displaced, 170M created by 2030; net +78M; 22% of jobs churnedEmployer-survey expectations, 1,000+ firms.
McKinsey Global Institute, Nov 202557% of US work hours technically automatableTechnical potential (44% AI agents, 13% robots), explicitly not an unemployment forecast.
MIT / Acemoglu, May 2024~5% of tasks cost-effectively automatable in 10 years; <0.66% TFP gainCost-effective deployment the skeptical bound.

Open questions read before quoting a number

  • Housing shortage size: 1.2M (NAHB) to 4.7M (Zillow) depending on definition vacancy-based vs. “missing households.” The timeline shows the Realtor.com series (3.80M in 2024 → 4.03M in 2025) as one method, not the settled figure.
  • Institutional investors in housing: nationally, owners of 1,000+ homes hold ~0.5% of the single-family stock (GAO / Urban Institute). The “Wall Street bought the housing stock” claim is not supported nationally; local concentration (e.g. Atlanta) is real.
  • “47-second attention span”: Gloria Mark’s mean from logging knowledge workers’ screens a sample, not a population constant. It is labeled as such in the Energy timeline.
  • Smartphones and youth mental health: the timing correlation is strong; the causal share is actively disputed in the peer literature. Not presented as settled anywhere on this page.
  • Obesity “decline”: 40.3% (2021–23) vs. ~41.9% (2017–20) may be stabilization, a GLP-1 effect, or survey noise. One cycle only.
  • No official 2050 forecast exists for mental-health prevalence, the legal system, or social capital; the forward markers on those rows are directional, from the forecasters named in each row.

The pattern under all thirteen

The glyph outlives the thing. A grade, a score, a code, a count the mark keeps circulating and deciding long after the underlying function has hollowed out. The system no longer needs the thing it was built to track. It only needs the symbol.

That is why these thirteen are the automation targets. A system that runs on symbols can be run by a machine that manipulates symbols. What gets automated is not the function the function is already gone in most of them. What gets automated is the mark-making itself.

What a normal person can do

Thirteen systems, one pattern underneath. Every move below counters a specific manipulation named in the pillar above it. No special skill is needed, no special tools, no account. Half the moves need nothing but paper and attention. The other half use a model as a mirror. Either way, the verdict stays with you.

01 / 13 Psychological

Counters: the psychological operation, the psych tax.

Without AI

  • Projection landing on unexamined shadow. When someone blames you for the exact thing they are doing, that is them, not you. Write it down: who, what, when. The written verdict is the storage the next mood cannot rewrite.
  • A mood that arrived with the message. Before you answer a charged message, ask whose mood is in your body right now. If the mood arrived with the message, it is not yours. Answer from your own state, or do not answer yet.

With AI

  • Language pre-loaded upstream of thought. Paste the exchange and ask the model to mirror the pattern back: who projected what, where the hook is, which words carry the verdict before you have voted. The model suggests. You decide.
  • The watcher kept offline. Ask the model to list the loaded words in a message and what each one smuggles in. Read the raw message first, then the model’s read. Your read comes before its read.

02 / 13 Financial

Counters: the decoupled symbol set, value as database rows.

Without AI

  • Value floating on belief plus enforcement. Price everything in hours, not dollars. How many hours of your work does this cost, and what did those hours buy last year against this year. That is the counter-ledger, priced in the territory.
  • A symbol with no anchor. Before you trust a number, a balance, a score, a rate, ask what it is anchored to. If the answer is "take it on faith," the anchor is the issuer.

With AI

  • The spread between issuance and adjustment. Hand the model a price, a wage, and a year, and ask it to convert the price into hours and show the math, line by line. Check the arithmetic yourself. The inputs are yours, the verdict is yours.
  • Rules rewritten by whoever holds the pen. Ask the model to read a contract or a terms page and list every place one side can rewrite the rules, in plain language. You send nothing until you have read that list.

03 / 13 Law

Counters: process as toll gate, the better ledger wins, statute as laundering, the faked delay.

Without AI

  • Outcomes going to the better record. Keep a dated ledger of every interaction: the date, the name of the person, what was said, what was promised. The better ledger wins.
  • Delay with no mechanism underneath. When a custodian stalls, ask for the mechanism in writing: what step is blocked, by what, until when. A real delay can be named, and its account stays the same when asked twice. A faked delay shifts with the audience.
  • The filing treated as the event. Put every complaint in writing, dated, and keep the copy. A verbal request does not exist on the operator’s ledger.

With AI

  • Escalation with a record. Ask the model to draft the notice or the demand letter from your ledger, in plain language, with the dates and the record cited. The model drafts. You send it.

04 / 13 Health

Counters: deferred maintenance as profit, managed symptoms billable indefinitely, the diagnosis as toll gate.

Without AI

  • Decay transferred onto the occupant. Keep your own body ledger: dated notes of symptoms, sleep, exercise, what changed and when. The chart can override your report only if you have no record of your own.
  • The body as the asset, the patient as the carrier. Exercise daily or every other day. It is the one lever that is not a symbol.
  • A resolved patient exits the system. When a plan manages the symptom forever and never names the source, ask for the source, in writing. Fix at the source, not downstream of it.

With AI

  • The pattern with no legal existence. Feed the model your body ledger and ask it to spot the pattern: what changed before the episode, what repeats, what the timeline shows. The model finds the shape. You verdict the cause.

05 / 13 Education

Counters: the good-enough-to-pass trap, stalling education.

Without AI

  • Passable output, zero learning. Never accept a passable answer as learning. After the model gives you the answer, close the screen and redo the work by hand, from memory. The friction is the learning.
  • Functions never built. Practice the three functions on purpose: abstraction (what is the general rule here), quantity (what are the numbers, what do they compare to), verdict (what is the single call, and what would change it).

With AI

  • Oracle atrophies, mirror sharpens. Use the model as a sparring partner, not an answer machine. Ask it to argue the other side, to find the hole in your reasoning, to test the verdict. You keep the verdict.
  • The trap at three scales. Ask the model to explain a topic, then explain it back in your own words with the screen closed. If you cannot, you did not learn it.

06 / 13 Media / Information

Counters: the authorless operation, the directed flood.

Without AI

  • Confession as the payload. Price every confession before you post it: what does this like, comment, or share give away, and who gets it. A confession given with its cost counted is a different substance.
  • Volume as a weapon. When the flood hits, slow down instead of speeding up. The flood works by making checking feel impossible. One checked claim breaks the wave.
  • The gradient that pleads innocent. Run the function check on any post that moves you fast: what got redefined in the first line, what conclusion does the redefinition carry for free, whose position improves if you believe it, what does it ask you to stop seeing.

With AI

  • Targeting trained on your confessions. Paste a viral post and ask the model to name the move, the redefinition, the free conclusion, the beneficiary, before you form a view. Read the raw post first, then the model’s read.

07 / 13 Housing

Counters: deferred maintenance as profit.

Without AI

  • Decay carried by the tenant. Photograph everything, dated: the stain, the wall, the unit, the notice. A dated photo converts private decay into a pattern.
  • The cosmetic fix. When the repair is paint over damp or a bent-fin clean, write down what was done and what was not. The cosmetic fix is the tell.
  • Channel theater. If the portal does not work, the phone loop never lands, or the vendor was never paid, write that down too. The denial instruments are part of the record.

With AI

  • The pattern with no negotiating existence. Hand the model your dated photos and notices and ask it to build the timeline: what was reported, when, what was done, what is still open. You keep the originals. The model builds the pattern.

08 / 13 Labor

Counters: face-saving jobs, temporal displacement.

Without AI

  • The role surviving the work. Run the face-saving test on your own role: if the job vanished tomorrow, what outcome would change? The answer tells you what the job actually is.
  • Hours eaten, watcher never online. Track your own hours for one week: work, commute, recovery. Read the total. If nothing is left, that is the mechanism, not a personal failure.
  • Exhaustion as compliance. Guard one unbooked hour a week: no input, no feed, no task. The watcher needs unbooked time to come online, and the script cannot eat an hour you have already claimed.

With AI

  • The audit of your own week. Ask the model to audit your week from your hour log: where did the hours go, what did they buy, what is left for the watcher. The numbers are yours. The model only adds them.

09 / 13 Food

Counters: the decoupled symbol set on labels, slop behind a presentable surface, private auditing only.

Without AI

  • The claim consumed as the property. Read the ingredient list, not the claim on the front. The claim is the symbol. The list is the referent.
  • Shit in, shit out. Run your own experiment: change one thing, write down how the body responds for two weeks, and trust the log over the label. The first week does not count. The body takes a while to readjust.
  • The presentable surface. Cook one meal from raw ingredients a day. The presentable surface cannot survive a kitchen.

With AI

  • Bad labels as the symbols in control. Photograph a label and ask the model to translate it into plain language: what is actually in the package, in order, with the numbers. Check the referent, not the claim.

10 / 13 Community

Counters: parasocial substitution, loyalty on a leaderboard, atomize the watchers.

Without AI

  • The count read as the relationship. Count the bonds, not the followers: who would show up, who answers, who reciprocates. The count is the substitute. The bond is the thing.
  • The seeing that never compounds. Move one conversation a week off the platform and into a room, a call, a walk. Shared infrastructure starts with two people in the same place.
  • Attention to reciprocators. Give your attention to the people who reciprocate, and cut the ones who only take. The feed inverts when the attention does.

With AI

  • One-way familiarity. Ask the model to audit your last week of platform time: how much went to people who reciprocate, how much to one-way feeds. The numbers show the substitution.

11 / 13 Governance

Counters: perform transparency, flood the zone.

Without AI

  • Consent-theater. Read the checkbox before you tick it: what exactly is being consented to, in plain words. A mark in a box is not agreement to terms nobody read.
  • The record made uncheckable by volume. Check the record against the territory: one claim, one checkable fact, one source that is not the system describing itself.
  • The flood. When the volume floods, slow down instead of speeding up. One checked claim breaks the wave.

With AI

  • The performance filed as the real thing. Paste a ballot measure, a bill summary, or a terms page and ask the model to list what changes, who gains, and what the plain-language version leaves out. Verify the key claim against the primary source yourself.

12 / 13 Energy

Counters: emotional extraction, strain as the cost line, one word with two meanings.

Without AI

  • Feeling prompted, then harvested. Name the feeling before you act on it: outrage, anxiety, devotion. A named feeling is a choice point. An unnamed one is a prompt.
  • The cost line carried by the person. At the end of the day, write down what the system gained from you (hours, posts, attention) and what you carry (tiredness, irritability). The strain is the receipt.
  • The highways. Stop feeding one highway for a week: the outrage feed, the loop, the late-night scroll. Note what returns. Starving the highway is the test that it was a highway.

With AI

  • Electrical words on a human state. Ask the model to read your day back to you from your own log: where did the feeling get prompted, what followed, what did it cost. The model mirrors. You verdict whether the feeling was yours.

13 / 13 Entertainment

Counters: attention capture, the synthetic flood, the siren song, collapse of the shared screen.

Without AI

  • Choosing replacing watching. Finish one thing before starting the next: one film, one season, one book. The flood wins when choosing replaces watching.
  • Stopping designed to feel like loss. Turn off autoplay. The switch makes stopping a choice instead of a failure.
  • The shared reference dissolved. Watch one thing with another person, at the same time, in the same room. The shared screen is the counter to a million private programs.

With AI

  • Manufactured feeling. Ask the model to list what the recommendation feed is optimizing for: watch time, not enjoyment. Compare it against what you actually liked this month. The gap is the capture.

You don’t fix a system by arguing with it. You stop feeding it.