Post-Labor Economics in 60 minutes
In a Nutshell
Post-labor economics describes a future where AI and automation eliminate human labor as a binding constraint on output, causing massive deflation, demonetization, and a declining labor share of income (from 66% to 56% over decades), threatening household income (70-80% of GDP) and government revenue. Solutions focus on shifting to transfers (e.g., UBI-like programs) and capital ownership (e.g., sovereign wealth funds like Alaska's, ESOPs, EOTs, DAOs), enabling $140k median household income (equivalent to $300k post-deflation) without wages. Humans will still be hired irrationally for embodied experiences, accountability, and social value, while trends compound over 4-5 decades amid an automation imperative.
These notes were generated by AI and may contain inaccuracies.
Post-labor economics started in the Human AI Empowerment Lab about 2 years ago. Carlos invited the speaker to speak to his students at Clemson. At the time, the speaker focused on AI safety, AI risk, and AI alignment. Students asked about AI's impact on their future prospects, what to study to maximize time at Clemson, and gave generic answers like learning how to learn or foundational knowledge.
After the talk, the speaker made a video called 'Post AGI Economics,' the most successful video with over half a million views, indicating strong interest. Spent the last 2 years studying, talking with professors and business leaders. This presentation is the current state of post-labor economics theory.
Simplest definition: an economic regime where human labor input is no longer a binding constraint on economic output. Speculative, based on models, data, trends. For all human history, labor has been a binding constraint for building Rome, bridges, getting to the moon, nuclear reactors—dependent on human mind, body, hands.
Economic baseline definition of labor: compensated human activity to produce valuable transformation. Transformation: take information, person, or matter and make it more valuable. Example: tree from forest, cut down, mill into timber, lumber, build house. University: take uneducated person, make more intelligent, educated, informed.
First principles physics: expenditure of energy to process information or manipulate matter yielding an economic state change. People pay for the outcome, not human input. Categories of human capabilities: cognition (plan, think, solve problems), dexterity (fine manipulation, mechanics, surgeons), strength (already automated with heavy equipment, power tools), social emotional (authenticity, empathy, trust, accountability).
Labor substitution: when capital or equipment replaces human labor entirely. Litmus test: better, faster, cheaper, safer—then economically rational, irrational to use humans. Example: tractor (over a century old) ended hand farming in modern economies; now a hobby like gardening.
Jobs are collections of discrete tasks, not monolithic roles. Technology unbundles tasks, repackages them to humans or machines. Automation: labor-saving technology—any machine doing what human used to do, starting with printing press over five centuries ago, enhancing output 150-400 times (less time per volume). Example: J.K. Rowling couldn't sell millions of hand-written Harry Potter books.
Automation not 20-50-200% efficient, but orders of magnitude: hundreds, thousands, millions times cheaper. Electric lights: candles cost 200-1,000 times more per day than today, produce less light. AI and robotics are latest automation, encroaching on human labor for a long time.
Printing press banned by Ottoman Empire for Muslims over two centuries to protect scribes' jobs, contributing to its fall by restraining technological advance.
GPT: Generative Pre-Trained Transformer (ChatGPT), but OG GPT is General Purpose Technology—fundamentally changes how humanity works: steam, electricity, wheel, metallurgy. AI satisfies GPT criteria:
- Pervasiveness: spreads across sectors (healthcare, law, government; construction via project planning, robots).
- Continuous improvement: rapidly evolving, cheaper over time, new abilities.
- Knock-on effects/spillover: innovation spillover, network effects.
Build intuition like electricity: started with shocks, then incandescent bulbs (Edison 1,000 iterations), electromagnets, motors, diodes, resistors, integrated circuits, AI. Electricity king of GPTs; AI downstream. Still discovering electricity uses after century; same for AI.
Value stream: raw material (experience, knowledge, stone, clay) through economic transactions to useful product. Campus bricks: clay, kilns, construction. AI/LLMs starting to transform value streams. Examples: chatbots for personal help, therapy, class work, family—changed life for most. Speaker used AI for hundreds of hours research on chronic illness, enabling in-person talks.
Healthcare value stream changing (not FDA approved, but self-treatment). Consulting with doctors, lawyers, engineers: AI will change everything. People pay for outcome, not process (later discussion on paying for process).
Encourage jotting questions; time for conversation.
Optimistic but realistic. Rapid innovation causes deflation (costs go down) and demonetization (transactions never happen). Technology deflationary: does more with less effort; money proxy for effort (unit of account, medium of exchange, store of value).
Haber-Bosch process: nitrogen in soil without guano; Earth's carrying capacity without technology 2-4 billion (half to three-quarters of room would starve). Lowers agriculture costs; productivity per acre up (fertilizer, breeding, genetics, weed control).
Demonetization: AI saves money (hundreds, thousands, tens of thousands). Speaker saved hundreds of thousands on medical experts, time, self-taught PhD in gut health.
Microeconomic decisions: use cheapest for satisfactory output (speaker chose AI over Johns Hopkins). Less demand for others' services → less revenue → fewer hires → less wages → less aggregate demand → layoffs → more automation.
Economy: firms provide goods/services, hire labor, pay wages to households, households demand goods/services. Capitalism requires supply/demand of goods/services; labor not required (strength, dexterity, cognition, social/emotional automatable).
Household income drives 70-80% GDP, wages 82% household income. If automation makes hiring irrational, household income drops → spending drops → consumer confidence drops → jobs dry up → tax base dries up.
Three sources:
- Wages: salary, hourly for time/energy.
- Transfers: government redistribution (SNAP, healthcare, housing, Social Security, veteran benefits; includes public university, school, transit).
- Capital: returns on ownership (dividends, equity, rent).
Most rely on wages. Keystone: if wages go away, replace with transfers/capital.
Secular trends (structural, not business cycles like Great Recession—speaker lost first tech job, lived in car). Living through next industrial revolution.
- Productivity-pay decoupling: since '70s, productivity rises, hourly compensation flatlines; automation a driver.
- Decline of labor's share of income: when all income received, income is received
Income is received by households from labor and by businesses. Historically, labor share of income was fixed at 64-66% for decades or centuries, considered an unspoken law of economics. Over the last five decades, it dropped from 66% to 56% in America and other developed economies, shifting trillions of dollars from households to capital and companies.
In 1950, 8-12% of household income was transfers; today it's ~20% in America and 30%+ in European nations. These trends indicate an economy becoming more automated and less labor-dependent, as businesses avoid hiring when tasks can be automated.
Three buckets: wages, transfers, capital. Universal Basic Income (UBI) involves government writing checks to everyone monthly, like pandemic stimulus checks. Other examples: guaranteed income, negative income tax—all government transfers.
Household income drives 70-80% of GDP, with 82% from wages. 80-85% of US federal revenue comes from income and payroll taxes. If wages vanish due to automation, government revenue collapses, challenging the assumption of the indispensability of labor underlying capitalist systems, policies, and treatises.
Replacing revenue requires wealth taxes, value-added taxes, corporate taxes. If 80% of household income disappears, living on 20% would be catastrophic; governments face the same in total automation.
Capital is currently the smallest bucket. Some families live on rent, dividends, bonds, but not enough.
Public Capital
- Alaska Permanent Fund (1970s, Republican governor): Funded by oil revenues into an endowment fund (like university endowments or retirement funds). Pays ~$1,700/year per family; critical for low-income ($30k/year) households.
- Norway Sovereign Wealth Fund (~$2 trillion, >$300k per citizen): Oil-funded, dividends fund government.
Public capital can distribute directly to families (Alaska) or fund government (Norway).
- ESOPs (Employee Stock Ownership Plans, popular in US): Employees own company stock; ESOP households have twice the retirement funds, reducing poverty and food insecurity.
- EOTs (Employee Ownership Trusts, popular in UK): Retiring owners pass businesses to employees.
- Cooperatives, trusts.
- DAOs (Decentralized Autonomous Organizations): Crypto/DeFi self-running companies; AI may enhance viability; slowly becoming legal.
Simulations of 140+ programs boosting transfers/capital yield median household income of $140k/year. With automation-driven deflation/demonetization, effective spending power reaches ~$300k/year without working.
Goal of post-labor economics: Keep government, households, and capitalism functioning if labor vanishes.
Not socialism/communism/Marxism (can explain why if interested).
Humans irrationally value embodied human experiences due to biology (e.g., prefer live concerts/sports over holograms/animatronics/recordings, despite cheaper alternatives).
Personal anecdote: Drove 5 hours to DC concert (band linked to meeting wife); valued for human provenance.
Audience uses AI for learning but prefers in-person from speaker/Carlos for greater value/retention. Economic actions cause transformation; interactions memorable vs. ChatGPT.
Other dimensions:
- Comfort with robot police/judge/president/governor? Humans corruptible; algorithms too (who writes them? Can't fire YouTube algorithm—unelected).
- Legal need for throat to choke: Human accountability (e.g., fire CEOs).
Human presence/provenance/accountability persists due to neurology, evolution, legal systems.
Trends (4-5 decades) will compound. Automation is beneficial (speaker automated 2 petabytes at data center to avoid headcount growth). Micro: Rational to automate (businesses, individuals use AI over hiring therapists/professors/doctors). Macro: Systemic impacts when everyone does.
~35 minutes for questions/discussion.
As AI/automation companies accumulate capital (data centers, robots, customers) without hiring, what compels sharing?
Presently little (corporate tax ~15%). Fordism: Henry Ford paid $5/day (high then) so employees could buy cars. Current economy demand-constrained: Supply investments (data centers/robots) outpace demand; people lack money for goods despite potential higher GDP/wealth.
Long-term rational for companies/billionaires to cooperate for circulation.
- Windfall trust: Voluntary one-time % to sovereign funds (unlikely, e.g., Apple/Microsoft won't give $20B).
- Prefer self-selling, implemented solutions: Sovereign wealth funds (national/state/municipal), EOTs, co-ops, DAOs.
- California: 2% wealth tax on billionaires + 40% exit tax (tax flight debate).
Already demand-constrained; need self-selling package.
Trump baby bonds (from Piketty's Capital in the 21st Century): Government buys $5k bond per newborn, matures to ~$20k at 18; mediates capitalism's hydrological cycle.
Design constraints: Politically viable, measurable, market/capitalist-friendly.
Not emperor-for-a-day redesign (fun in sci-fi, but ignores path dependency—starting point limits destinations). Communism/socialism abolishes private property for government ownership (tough sell in capitalist America; not viable voting bloc).
Instead: Proven, self-selling global policies (US, Europe, South America, India, Asia).
- Alaska Permanent Fund: Republican-created, not socialist.
Build incrementally to post-labor world. Capitalist-friendly: In 20-30 years, most income from capital ownership; everyone as investor (allocate returns manually/autopilot). Plenty to do.
Anthropic (others): Pay employees who automate their jobs.
Upper echelons business theory strategy. Companies rationally lay off (neoliberalism post-1980 Thatcher/Reagan: market labor flexibility—fire anytime for efficiency). Pre-1980 embedded liberalism (1930s-70s): Ensure jobs, even ghost/BS jobs.
Automation imperative in competitive markets (Anthropic vs. OpenAI/Meta).
Align incentives: Pay to automate out of job (reduces fear).
Speaker's Cisco story: Automated expensive snapshot cleanup—script matched engineer usernames to emails and automatically emailed
Engineers had a snapshot taking up space. Asked them to delete it. Took weeks to figure out a script. Ran it and freed many terabytes, tens of thousands of dollars in storage. Manual work would have been more than full-time. Set it to run automatically every morning. If an engineer at Cisco had a snapshot too old, it deleted it. Weeks of work avoided. Repeated with other tasks, became a one-man shop. Slightly ahead of the curve on automation.
Vibe coding does this. Instead of weeks figuring out Python, PowerShell, APIs by hand, ask Codex. Plug in code: "I need to delete these things. Do it for me." Takes 30 seconds, then test. Automation flywheel picking up.
In post-labor world, all humans lived with AI present. Children grow up, have kids. Automation is all humans know. Humans go to war. Hypothetical: big war, all data centers down. Can humans revert to pre-automation?
Hunger is powerful motivator. Short answer: yes. Longer: painful. Managed data centers. Amazon data centers bombed in probably illegal invasion of Iran. Glad not on call for that outage.
Outlining sci-fi term zero infrastructure: bomb every bit of civilization's infrastructure to send back to Stone Age. Very difficult. Bigger problems like nuclear fallout.
Carrington events, big solar flares, viral cyberattacks shut things down briefly. Amount of energy, kinetic or electromagnetic, to reset civilization would be visible, big event.
If dependent on automation long time, fear losing air conditioning. Grew up in South, dependent, vaguely know how it works, not how to fix.
Philosophical question: what do we do realizing dependence on technology? Haber-Bosch process for nitrogen: if reinvent agriculture, how many starve? Who knows agriculture at scale? Know germination, fertilizer, pests, but not at scale.
Species dependent on technology and division of labor long time. Risk? Sure. Minor possibility. Optimistic, but pessimists, doomers, preppers plan for worst-case. Diversity of human thought advantage. People prepare for worst-case.
Personal philosophy: avoid civil war from 48% unemployment, no solution. Hope work prevents it. If resonates, think how to store information to reboot civilization. Plenty working on that. Makes great story.
Separate imagination from empirical work. Early cognitive architectures with agents like OpenCog. One of lead first with LLMs. Wrote science fiction alongside: throw off training wheels, use imagination. What could happen?
Cases where AI oversteps: workflow deletes entire inbox.
When technology goes sideways, deletes email, databases: Tuesday in technology. User weakest link. PEBKAC (problem exists between keyboard and chair), ID10T error (idiot), layer eight issue (OSI model, meat using the thing).
For autonomous automation, need interlocks, cognitive architectures, safety checks. Not fully developed.
German dude retires selling tech company, beaches, then hyperfocuses on fully autonomous AI agent. Uses LLM, tools, writes code. Goes viral, surpasses Linux on GitHub popularity. Joins OpenAI, didn't need money, has tens of millions.
Market incentive enormous. Anthropic (Claude) charges more for OpenCog use, signals they'll do their own.
Every cybersecurity or UX issue: hive mind learning. Capitalism creates complex adaptive system. Find failure mode, vulnerability, everyone learns, fixes, moves on.
Setback? Yes, more if deletes databases or leaks data. Aggregate: builds trust. Seen failure mode, know how/why, new design spec. Short-term setback.
Hypothetical: Palantir (AI, military integrated) uses OpenAI/Anthropic. AI says nuclear attack, US scrambles fighters, DEFCON, ready counterattack. Litigated publicly, "not okay." Pressure politicians, next election anti-AI.
AI unpopular, divisive. Could slow politically. Technically, nothing stopping except energy, algorithmic requirements. Economic incentive insane.
Studied systems theory? Complex adaptive system: stock market. Investors, regulators, companies have beliefs, information asymmetry. GameStop: hyperstition stock price up.
Humanity and technology always complex adaptive system. Create technology, changes us, we react, technology reshapes. Knock-on effects make final shape hard to forecast.
iPhone obvious in hindsight. Japanese tech executive in '90s wrote book on mobile tech to work from, described smartphone. 12-13 years ahead, ignored. Japan more technologically sophisticated than US in '80s.
Can't predict final shape. If just figured electricity, "metal think with that" not intuitive to all.
Thank you, everybody. Big data.
Keep David Shapiro in your library
Save the videos and channels worth coming back to, and find them again in one place.
