Louis Rosenberg recently argued we’re “entering the first stage of grief” about AI. He’s wrong on two counts: the timeline and the stage.
The anger didn’t start this summer with mixed reviews of new AI models. It started in January 2021 when OpenAI launched DALL-E and the artist community exploded over unauthorized training data. We’re not entering grief—we’re three years deep into it.
But here’s what Rosenberg missed: The real bubble isn’t what the critics think it is.
The Linguistic Evolution of Rage
Watch how the language evolved:
- 2021-2022: “AI is THEFT”
- 2022-2023: “AI quality is questionable”
- 2023-Present: “AI slop”
Same rage. Increasingly palatable packaging.
The shift from “theft” to “slop” isn’t about changing concerns—it’s about finding socially acceptable language for anger. “Slop” lets people dismiss AI without sounding like Luddites. It focuses on output quality while masking the underlying fury about displacement and devaluation.
And it’s not just artists. Scroll through any AI post and you’ll see variations of the same anger: “AI bypasses real creativity,” “You can’t learn from using AI,” “It adds no value.” The tone isn’t skeptical—it’s lashing out.
Where We Actually Are in the Grief Cycle
Most people aren’t in denial or acceptance. They’re oscillating between anger and bargaining:
- Anger: “AI slop” dismissals, theft accusations, quality critiques
- Bargaining: “FINE. I’ll try it on this ONE task” (then either: “See? Slop!” or secretly using it while publicly complaining)
But here’s the problem with Rosenberg’s framework—he’s applying one grief timeline to everyone. In reality:
- Artists/writers: Deep in anger
- Enterprise buyers: Denial mixed with FOMO
- Workers in vulnerable sectors: Moving toward depression
- Tech investors: Bargaining (hedging while staying in)
- AI builders: Already at acceptance
Different groups process grief at different speeds. Everyone experiences it differently.
The Real Bubble: A Deflationary Spiral Disguised as Progress
Everyone’s debating whether AI is a financial bubble—questioning profitability timelines, utility gaps, compute costs. These are legitimate business concerns.
But they’re missing the societal bubble: a deflationary spiral that looks like technological progress.
Here’s the mechanism:
The Pre-AI Displacement Model
- Lose job → Apply same skills elsewhere → Re-employed
- Safety valve: “Learn to code” (retraining pathway)
The AI Displacement Model
- Lose job → Skills obsolete → No retraining pathway → Void
- The safety valve is automated too
The Feedback Loop
- AI-driven unemployment rises
- Reduced consumer spending
- Business revenue collapse
- Companies adopt MORE AI to cut costs
- Accelerated unemployment
- Repeat
The Leading Indicator Everyone’s Ignoring
While unemployment sits at 4.2%—historically low—mental health disability claims have surged 50% since 2015, now representing 32% of all disability claims in major economies.
This isn’t normal workplace stress. This is a leading indicator.
The Divergence:
- AI adoption: ↗️ Vertical (78% of organizations now using AI, up from 55% in 2023)
- Unemployment: ↘️ Declining
- Mental disability claims: ↗️ Climbing to historic highs
Translation: AI isn’t eliminating jobs yet. It’s making them psychologically unsustainable first.
Mental health-related disability complaints in the federal workforce increased 72% over a five-year period, far outpacing overall complaint growth. People aren’t losing their jobs—they’re burning out trying to keep them.
The Three-Phase Collapse Pattern
Based on correlation analysis of AI adoption vs. labor market health (2015-2025), we’re tracking a three-phase pattern:
Phase 1: Task Displacement (2015-2023)
- AI automates specific tasks within roles
- Productivity expectations increase (one worker now does 1.5 workers’ output)
- Mental health strain accelerates
- Unemployment falls (fewer workers needed, but remaining workers are stretched)
- Mental disability claims become the release valve
Phase 2: Role Automation (2024-2027) ← We are here
- AI capabilities cross threshold to automate entire roles
- “Hidden unemployment” via disability claims becomes visible unemployment
- The U-6 rate (7.6%—which includes marginalized workers) is the early warning
- Consumer spending begins contracting
Phase 3: Deflationary Spiral (2027+)
- Unemployment rises above 5% and continues climbing as the feedback loop intensifies
- Reduced consumption → Business revenue collapse
- Companies accelerate AI adoption to cut costs
- More unemployment → Repeat
- Unlike previous recessions, the “learn new skills” safety valve is gone—AI has automated the retraining pathway
Historical context: The U.S. hit 5.5% unemployment in 2015, 6.7% in 2014, and 10% in 2009. The economy recovered each time because displaced workers could retrain. The difference now: When unemployment climbs past 5% in Phase 3, there’s no skills-based escape route—AI has closed it.
The Wealth Concentration Trap
Here’s the twist in the data that most people will miss:
- Low-wage workers (bottom 10%) saw 148% indexed wage growth since 2020
- Top earners (top 10%) saw 127% indexed growth
Looks good, right? Low wages growing faster?
Wrong.
Labor’s share of total income: 58% in 2025, down from ~60% in 2015
What this means:
- Workers are getting marginally more crumbs
- Capital owners (AI investors) are capturing the surplus
- The pie is growing, but workers get a shrinking slice
- This accelerates when full role automation hits (Phase 2-3)
The Innovation-Cost Correlation
The strongest correlation in recent economic data: As AI capability increases, mental health strain increases proportionally.
- AI capability (Stanford HAI AI Index): 98 points (composite score 0-100)
- Mental disability claims: 32.5% of all disability claims
- Monthly startup formation: 540,000 (solopreneurs enabled by AI, but also indicating traditional employment instability)
We’re seeing a boom in business formation and a mental health crisis. This isn’t coincidence—it’s people trying to escape unsustainable employment conditions.
The Phase Shift Nobody’s Preparing For
Right now, AI displacement is primarily digital: writers, artists, coders, analysts, customer service.
When humanoid robots go mainstream—construction workers, hotel staff, retail, food service, manufacturing—we’ll see a surge in physical job displacement that dwarfs the current wave.
Without intervention, you’ll have:
- Mass unemployment without safety nets
- Accelerating wealth concentration (the “have/have-not” divide)
- Depressed, angry, lost workers with no support structure
- Social instability
This isn’t a financial bubble. It’s a societal pressure cooker.
Why Misdiagnosis Matters
If we treat this as a financial bubble, we’ll focus on:
- Market corrections
- Investment timing
- Profitability metrics
If we recognize it as a grief-driven economic transition, we’ll focus on:
- AI literacy programs (understanding, not just usage)
- Proactive job retraining (before displacement, not after)
- Universal Basic Income (funded primarily by AI companies)
- Managed transition mandates (limits on AI job replacement as % of workforce, increased gradually like minimum wage)
The first approach manages investor portfolios.
The second approach prevents societal collapse.
The Uncomfortable Truth
The “AI bubble” critics aren’t wrong about financial concerns. But they’re asking the wrong question.
The question isn’t: “Is AI overvalued?”
The question is: “Can our economic and social systems absorb this rate of displacement?”
Right now, the answer is no.
And when we misdiagnose the problem, we prescribe the wrong solution.
What We Actually Need
- AI companies funding the transition
Not governments scrambling to retrofit 20th-century safety nets. The productivity gains are being captured at the capital layer (declining labor share of income). The funding mechanism is redirecting a portion of those gains toward transition support. - Proactive policy
Workforce transition limits before unemployment spikes. This means managing the percentage of roles that can be automated within organizations, scaled over time based on labor market health metrics. - Mental health infrastructure now
The 50% surge in disability claims is a crisis today, not tomorrow. We need expanded mental health support, workplace stress reduction mandates, and recognition that burnout is a systemic issue, not individual weakness. - Honest conversation about grief
Stop dismissing legitimate anger and fear as technophobia. Acknowledge the psychological reality of rapid change. Create spaces for people to process displacement anxiety without shame. - Redefine work and value
When one person can do what ten did, how do we distribute both work and wealth? This is the foundational question we’re avoiding.
The Bottom Line
The grief framework is real. The “AI slop” narrative is anger in disguise.
But the data reveals something more dangerous: We’re not waiting for mass unemployment to trigger a crisis.
The crisis is already here—hidden in mental health disability claims, masked by low headline unemployment, accelerating in wealth concentration.
The “AI bubble” critics are partially right—there’s a bubble forming.
They’re just looking at the wrong metric.
This isn’t a financial bubble.
It’s a psychological sustainability bubble—and it’s already starting to burst.
When you misdiagnose the problem, you prescribe the wrong solution.
The question isn’t whether AI will take jobs.
It’s whether humans can psychologically and economically sustain the transition we’re already experiencing.
Right now, the answer is no.
Sources
- Institute for Fiscal Studies (UK): Mental health disability claims data (2025)
- U.S. Equal Employment Opportunity Commission: Federal sector mental disability complaints
- Bureau of Labor Statistics: U-3, U-6 unemployment, labor force participation
- Stanford HAI AI Index Report 2025: AI adoption and capability metrics
- Federal Reserve Economic Data (FRED): Wage growth, labor share of income
- Census Bureau Business Formation Statistics: Monthly startup data