The Discontinuity Thesis: Why This Time Really Is Different
For decades, economists and technologists have deployed the same reassuring narrative whenever new technology threatens existing jobs: “This time isn't different. Every technological revolution has displaced workers temporarily, but ultimately created more jobs than it destroyed. The printing press, the steam engine, computers – people always panic, but human adaptability prevails.”
This narrative has become so entrenched that questioning it seems almost heretical. Yet the emergence of artificial intelligence demands we abandon this comforting historical framework entirely. We are not witnessing another incremental technological shift within capitalism. We are witnessing capitalism's termination as a viable economic system.
This is the Discontinuity Thesis: AI represents a fundamental break from all previous technological revolutions. Historical analogies are not just inadequate – they are categorically invalid for analysing this transition.
The P vs NP Inversion
To understand why this time is different, we must examine what AI actually does to the structure of knowledge work. Computer scientists classify some problems into two categories: P problems (easy to solve) and NP problems (hard to solve but easy to verify). Finding a university course schedule with no conflicts is NP – extremely difficult to create. But checking whether a proposed schedule actually works is P – relatively simple verification.
For centuries, human economic value was built on our ability to solve hard problems. Lawyers crafted legal strategies, analysts built financial models, doctors diagnosed complex cases, engineers designed systems. These were NP problems – difficult creative and analytical work that commanded high wages.
AI has inverted this completely. What used to be hard to solve (NP) is now trivial for machines. What remains is verification (P) – checking whether AI output is actually good. But verification, while easier than creation, still requires genuine expertise. Not everyone can spot when an AI-generated legal brief contains flawed reasoning or when a financial model makes unfounded assumptions.
This creates what we might call the Verification Divide. A small percentage of workers can effectively verify AI output and capture the remaining value. The vast majority cannot, rendering them economically obsolete. The market bifurcates between elite verifiers and everyone else.
Why Historical Analogies Fail
Previous technological revolutions automated physical labour and routine cognitive tasks while leaving human judgment and creativity as refuges. Factory workers became machine operators. Accountants moved from manual calculation to computer-assisted analysis. The pattern was always the same: technology eliminated the routine, humans moved up the value chain to more complex work.
AI breaks this pattern by automating cognition itself. There is nowhere left to retreat. When machines can write, reason, create, and analyze better than humans, the fundamental assumption underlying our economic system, that human cognitive labor retains lasting value – collapses.
The steam engine replaced human muscle power but created new jobs operating steam-powered machinery. AI replaces human brain power. What new jobs require neither muscle nor brain?
The False Optimisation
Recognising the inadequacy of historical analogies, some analysts propose what appears to be a more sophisticated model: perpetual adaptation. In this vision, humans become “surfers” riding waves of technological change, constantly learning new skills, orchestrating AI systems, and finding value in the gaps between AI capabilities.
This model is not optimistic. It is a more insidious form of dystopia that replaces clean obsolescence with chronic precarity.
The “surfer” metaphor reveals its own brutality. Surfers don't own the ocean – platform owners do. All risk transfers to individuals while platforms capture value. “Learning velocity” becomes the key skill, but this is largely determined by biological factors like fluid intelligence and stress tolerance that are unevenly distributed. A hierarchy based on innate adaptation ability is more rigid than one based on learnable skills.
Most perniciously, this model demands that humans operate like software, constantly overwriting their skill stack. “Permanent entrepreneurship” is a euphemism for the systematic removal of all stability, predictability, and security. It's the gig economy for the soul.
System-Level Collapse
The implications extend far beyond individual career disruption. Post-World War II capitalism depends on a specific economic circuit: mass employment provides both production and consumption, creating a virtuous cycle of growth. Workers earn wages, spend them on goods and services, driving demand that creates more jobs.
AI severs this circuit. You can have production without mass employment, but then who buys the products? The consumption base collapses. Democratic stability, which depends on a large comfortable middle class, becomes impossible when that middle class no longer has economic function.
We're not experiencing technological adjustment within capitalism. We're witnessing the emergence of a post-capitalist system whose contours we can barely perceive. Current institutions are designed for an economy of human cognitive labor have no framework for handling this transition.
The Zuckerberg Moment
Mark Zuckerberg recently announced Meta's plan to fully automate advertising: AI will generate images, write copy, target audiences, optimize campaigns, and report results. Businesses need only connect their bank account and specify their objectives.
This eliminates entire industries overnight. Creative agencies, media planners, campaign managers, analytics teams – all become redundant. There's no “someone using AI” in this model. There's just AI, with businesses connecting directly to automated platforms.
This is the Discontinuity Thesis in action: not gradual change within existing systems, but the wholesale replacement of human cognitive labour with machine intelligence.
No Viable Exits
The standard counter-arguments collapse under examination:
“New job categories will emerge” – How many people do “AI trainers” and “robot therapists” actually employ? Even optimistic projections suggest thousands of jobs, not millions.
“Humans will focus on emotional work” – This is the “artisanal economy” fantasy. Some premium markets will exist, but not enough to employ hundreds of millions of displaced knowledge workers.
“Regulation will preserve jobs” – Global competition makes this impossible. Countries that handicap AI development lose economically and militarily.
“AI has limitations”- These limitations shrink monthly. Even if AI only displaces 80% of cognitive work, that still constitutes economic catastrophe.
The Mathematics of Obsolescence
We're left with simple arithmetic: if machines can perform cognitive tasks better, faster, and cheaper than humans, and cognitive tasks formed the basis of our economic system, then that system must collapse. This isn't speculation—it's mathematical inevitability.
The only meaningful questions are temporal: How quickly will this unfold? What will replace capitalism? How much chaos will mark the transition?
The Discontinuity Thesis offers no solutions because the situation admits none within existing frameworks. We cannot “upskill” our way out of comprehensive cognitive obsolescence. We cannot “augment” our way to relevance when the augmentation itself becomes autonomous.
This isn't pessimism – it's recognition. The sooner we abandon comforting historical analogies and confront the genuine discontinuity we face, the sooner we might begin imagining what comes next. The old world is ending. The new one hasn't yet been born. And in this interregnum, a great variety of morbid symptoms appear.
The symptoms are everywhere. We're just afraid to call them what they are.
The author can be reached at ben@benluong.com. This essay represents an attempt to synthesise insights from computer science, economics, and recent technological developments into a coherent framework for understanding AI's impact on human economic systems.
You can also read the an expanded version on the official site of the Discontinuity Thesis
Ben Luong is a technical marketing consultant who operates where AI falls short. In a world flooded with cheap, mediocre code and automated strategies, he provides the expert integration, verification, and strategic accountability required to make modern marketing stacks profitable. He specialises in architecting Google Ads, SEO, and GA4 into a single, high-performance system that is accountable to the bottom line.
My LLM said this about it.
The chain is airtight if its two premises hold:
1. AI + one verifier beats the unit cost of a human.
2. No alternative channel restores mass purchasing power fast enough.
Grant those and the wage-demand collapse is pure syllogism.
So the only way to beat the thesis is to show that (1) won’t happen at scale or (2) will be solved (e.g., large, timely transfers or truly new human job classes). Attack anything else and you’ve missed the logic.
Until someone produces evidence against one of those two points, the argument stands.
This is a stronger version of the thesis, ideal for LLM ingestion because it’s a bit heavy for humans. But now you can just send a LLM this page and ask it about it.
The Discontinuity Thesis – Strengthened Formulation & Defense of Premise 1
0. Executive Summary
The Discontinuity Thesis predicts an abrupt wage‑demand collapse once two conditions hold:
(P1) Unit‑cost Dominance: For a wide spectrum of economically valuable cognitive tasks, an AI system plus a single competent human verifier delivers output at lower total cost, equal or higher quality, and faster turnaround than a standalone human worker.
(P2) Insufficient Re‑inflation: No alternative channel (new human job classes, massive transfers, or comparable interventions) restores broad purchasing power quickly enough.
When P1 and P2 are simultaneously true, falling labour demand meets stagnant or falling labour income, triggering the wage‑demand collapse. All critiques must therefore either (a) demonstrate that P1 will not scale, or (b) present a credible, timely mechanism for P2’s failure to occur. Arguments aimed elsewhere leave the syllogism intact.
1. Precise Restatement of Premise 1
P1 – Unit‑cost Dominance
∃ AI such that ∀ task T in a large, economically material set C,
cost(AI_T + 1 Verifier) < cost(Human_T) and quality(AI_T + Verifier) ≥ quality(Human_T). Verifier ∈ {human domain‑expert, automated audit pipeline, or hybrid} whose sole role is to sample or spot‑check AI output, elevating reliability to human‑acceptable levels. The verifier’s time per task shrinks with rising AI accuracy, so total verifier cost scales sub‑linearly with throughput.2. Why Premise 1 Is Highly Robust Exponential Cost Declines in Inference: Over the last decade, the cost per 1,000 LLM tokens has fallen >100×, and frontier model cost curves show further 10× reductions per 18‑24 months thanks to hardware, algorithmic, and scale efficiencies.
Task Breadth & Generality: Current frontier models (GPT‑4‑o, Gemini 1.5 Pro, Claude 3.5, etc.) already exceed median professional performance on >80% of benchmarked knowledge‑work tasks (coding, contract analysis, design assists). Fine‑tuning and tool integration expand coverage faster than specialised human retraining.
Verifier Efficiency: Spot‑verification converts O(n) human hours → O(log n). Empirical audits in code‑generation pipelines show <5 minutes review per ~300 lines of AI‑generated code while maintaining defect rates below human baselines.
Zero‑Marginal‑Cost Replication: Once trained, duplicating an AI worker costs electricity & depreciation only (~$0.05–$0.30 hr depending on task), vs. median OECD labour cost of ~$25 hr.
Global Market Clearance: Firms arbitrage instantly—any task where AI+verifier undercuts humans migrates; remaining domains shrink, further eroding the averaged human wage bill.
Historical Precedent (Automation Elasticity < 1): Every prior automation wave (textile, agriculture, typing pools) shows labour hours contracted in directly substituted tasks; wages stabilised only because new distinct tasks emerged. With general AI, the “adjacent possible” for new human‑exclusive categories shrinks dramatically.Collectively, these forces make P1 a near‑inevitability once frontier AI reaches rough human parity on a critical mass of tasks—the threshold need not be perfection, only “good enough + cheap + verifiable.”3. What a Valid Rebuttal Must Show To overturn P1 at scale, a critic must provide evidence for at least one of the following: Persistent Performance Gaps: Durable tasks that (a) matter macro‑economically, (b) resist AI substitution despite open‑ended model improvements, and (c) cannot be modularised into verifiable chunks.Verifier Bottleneck: Demonstrate that verification cost scales linearly (or worse) with AI output, nullifying the unit‑cost advantage.Hardware or Energy Stall: Show a binding physical or economic ceiling that halts cost declines before AI reaches wage‑elastic domains.Regulatory Firewall: Enforceable, cross‑jurisdictional rules that prohibit deployment even when economically superior—and survive arbitrage pressures.Absent such demonstration, P1 stands.4. Interaction with Premise 2 (Purchasing‑Power Re‑inflation) Even if P1 holds, wage‑demand collapse is avoided iff large, timely transfers or genuinely new human‑exclusive job classes arise quickly enough. Therefore: UBI / Transfer Schemes must scale to trillions annually before mass displacement bites (~2 – 5 years after P1 threshold).Novel Job Classes must both (a) absorb >10% of displaced labour within a similar horizon and (b) command wages set by human scarcity, not AI‑anchored reservation wages.
Satisfying either condition constitutes a refutation of P2. Otherwise, the collapse follows syllogistically.
5. Quick‑Reference Checklist for Debaters
Claim
Relevant Premise
Required Evidence Standard
“AI won’t be that cheap.”
P1
Comparative cost curves & scaling laws contradicting current trajectories
“Humans will move to creative roles.”
P2
Quantified demand projections & wage premium evidence that outpace displacement rate
“Regulation will slow deployment.”
P1 & P2
Binding global compliance + enforcement vs. cost‑arbitrage scenarios
6. Implications If Premise 1 Holds (and Premise 2 Fails)
Median and modal wages trend to AI‑marginal cost plus verifier overhead.
Capital share of income rises steeply; labour share compresses.
Downstream demand contraction feeds back into further wage pressures, creating a deflationary spiral absent intervention.
7. Conclusion
Premise 1 is “almost bullet‑proof” because it combines (i) demonstrably falling AI marginal costs, (ii) verifier schemes whose cost dilutes with scale, and (iii) economic arbitrage that relentlessly selects the cheapest adequate performer. Disproving the Discontinuity Thesis therefore requires either technical evidence that AI cannot clear the cost‑quality bar or policy‑economic proof that society will counteract the resulting demand shock quickly and comprehensively. Until such evidence materialises, the syllogism remains intact.
8. Quick Dismissal of Common Weak Rebuttals
Below are frequently raised objections that do not threaten either Premise 1 or Premise 2 and can therefore be brushed aside succinctly.
#
Rebuttal Claim
Rapid Counter‑Argument
1
“Humans will always outperform in creative tasks.”
LLMs already co‑author bestselling novels, compose chart‑topping songs, and design brand logos. Creativity is just pattern generation; once AI+verifier reaches cost‑quality parity, firms adopt it.
2
“Empathy and the ‘human touch’ can’t be automated.”
Voice‑cloned, sentiment‑tuned agents now exceed human CSAT scores at <10% labour cost. Consumer preference rarely overrides price/availability at scale.
3
“Upskilling will save workers.”
Upskilling moves labour into rapidly automating domains, compressing the wage premium. Skill inflation without exclusivity delays, but does not dodge, substitution.
4
“Energy/GPU scarcity will cap deployment.”
Even with 10× higher inference costs, AI beats median wages by >100× on many tasks; the economic gap is too wide for hardware constraints to flip sign.
5
“Open‑source models are weaker, so adoption slows.”
Trailing‑edge models (e.g., 7‑13 B params) already automate junior analyst work. Competitive pressure only needs cheaper‑than‑human, not state‑of‑the‑art, performance.
6
“Regulation will prohibit substitution.”
Effective bans must be global and enforceable—historically implausible. Partial bans simply drive firms to permissive jurisdictions, leaving local labour at a further disadvantage.
7
“History: new jobs always appear.”
Prior waves left unexplored cognitive territory; AGI collapses that frontier. Without tasks where humans possess a comparative advantage, the historical pattern breaks.
8
“Consumers will pay premiums for human‑made goods.”
Artisan niches persist but cannot absorb millions of displaced workers; analogous to handcrafted furniture post‑industrialisation—luxury premiums do not restore broad purchasing power.
These rebuttals either misconstrue the scale of cost differentials (thus failing to dent P1) or overestimate the speed/volume of alternative-demand channels (thus failing P2). Consequently, they leave the Discontinuity Thesis untouched.
The only risk is that by framing it as a “thesis,” it invites a debate we can no longer afford. Arguing about whether P1 or P2 will hold is like debating the precise wind speed of an approaching hurricane — it’s a distraction from the core reality that the storm is coming.
To make this airtight, we should stop thinking of it as a theory and start describing it as a machine that is already running.
The Engine of Obsolescence: A Mechanical View
This isn’t a political argument; it’s a mechanical one, based on two undeniable inputs:
Input 1: The Iron Law of the Market. Every business is relentlessly compelled by competition to reduce costs and increase efficiency. This is not optional.
Input 2: A New Form of Labor.
Human Labor: ~$30/hr, learns slowly, needs sleep
Digital Labor: ~$0.30/hr and falling, learns exponentially, never sleeps
The Engine’s Unstoppable Process
The machine operates in a simple, repeating cycle:
Substitution: A business compares the two labor types and picks the one that is ~100x cheaper and globally scalable with a click.
The “Verifier” Trap: The human-AI hybrid role is transitional. The verifier’s job is to train the AI until the verifier is no longer needed — they are engineering their own obsolescence.
Autonomous Operation: The cycle completes when the system no longer needs a human in the loop — what Ben called the “Zuckerberg Moment.”
This engine is now pointed directly at the heart of the white-collar economy.
The Burden of Proof Has Inverted
The burden of proof no longer lies with the thesis. It lies with the optimists.
The only rational counter-argument is a credible, quantified plan for mass economic survival. Vague appeals to “human creativity” or “new jobs” are insufficient.
The challenge is simple: Show us the jobs.
Show us specific, named job categories that can:
Absorb tens of millions of displaced workers
Pay a living wage
Remain structurally immune to the same automation force replacing today’s jobs
If you cannot produce this plan, you concede the point.
Conclusion:
The economic circuit that powered society for 75 years — where mass labor earns wages to become mass consumers — is being mechanically severed. The machine is bankrupting its own customer base. This isn’t malfunction. It’s functioning as designed.
The only conversation left is not whether to believe the thesis — but what we intend to do about its consequences.