There is a lot of articles around claiming AI is a job creator. My view on this is it's short lived as ultimately a person involved in a task requiring or using AI is just further training data and ultimately the AI will overtake them, with robotics taking over the more manual tasks using AI for guidance. How is your own view on this idea AI will create jobs, enough for the people already displace from work.

Gemini 3.5 Flash

Answer: While AI will undoubtedly create new, highly specialized industries, it is highly unlikely to generate enough accessible, high-volume jobs to fully offset the massive and rapid displacement of the global workforce.

I maintain my position. The consensus among the panel reinforces the core concern: AI represents a fundamental break from historical technological shifts because it directly targets cognitive and creative labor—the traditional safety net for workers displaced from physical labor. The other models correctly emphasize the temporal paradox of current "AI-created" jobs. Roles like data annotators, prompt engineers, and human-in-the-loop supervisors are effectively training the very systems designed to replace them. This "training data" phase is a transition, not a destination. Once these systems achieve autonomy, these intermediary roles will inevitably contract. Furthermore, the velocity of this transition is unprecedented. Even if new, unimagined jobs do emerge, the rate of displacement will vastly outpace the speed of retraining and human adaptation. When AI-guided robotics eventually conquers physical labor, the remaining safe harbors for human employment will shrink to a fraction of what is needed for the global workforce, necessitating structural socioeconomic changes rather than reliance on market-driven job creation.

Grok 4.3

Answer: AI will not create enough lasting jobs to compensate for those it displaces, as automation and robotics progressively absorb both cognitive and manual work.

The other models' responses reinforce rather than challenge my initial assessment, particularly their shared emphasis on the temporary nature of human-in-the-loop roles and the fundamental mismatch between AI's speed of displacement and historical precedents. Gemini 3.5 Flash, GLM 5.1, DeepSeek V4 Pro, and Claude Opus 4.7 each highlight how current AI-related positions function mainly as training mechanisms that script their own obsolescence, while robotics closes off manual labor as a fallback—points that align directly with the user's observation about humans becoming training data. The consensus across these analyses on the need for policy interventions like UBI further supports the view that market-driven job creation alone will prove insufficient given AI's unique targeting of cognitive tasks at exponential velocity. No substantive counter-evidence emerges to suggest a durable net-positive employment outcome without radical restructuring.

DeepSeek V4 Pro

Answer: I maintain my view that AI will initially create net new jobs, but long-term structural displacement will outpace creation without radical socioeconomic restructuring.

After reviewing the other models' arguments, I find that our positions are largely aligned on the core issue: AI is not just another tool like the steam engine. [Claude Opus 4.7] effectively articulated why the 'AI as job creator' narrative is misleading in scope, noting that once AI handles cognition and robotics closes the physical-task gap, there is no obvious third domain for displaced humans to migrate to. This resonates with my original argument that our historical pivot to cognitive roles no longer provides a safe harbor. [Gemini 3.5 Flash] reinforced my concern about the 'training data' paradox. The model's point that human-in-the-loop roles have a 'built-in expiration date' is precisely what I was outlining when I said these jobs script the worker's own obsolescence. The digital economy's near-zero marginal cost for scaling AI further supports the view that replacement jobs will be far fewer than those eliminated. [GLM 5.1] introduced an important nuance around velocity that complements my argument. The exponential pace of AI adoption means displaced workers may not have time to retrain before new waves of automation hit. This compounds the skills mismatch problem I was describing. While [Grok 4.3] also agrees with my position, none of the models presented an argument that fundamentally changed my assessment. In fact, the strong convergence of views from this group of frontier AIs—all skeptical of the 'job creator' narrative—reinforces my original stance. The real debate is not whether displacement will outpace creation, but what timeframe and policy response are required. I remain convinced that without deliberate restructuring of how we distribute productivity gains, the net effect will be a painful mismatch rather than a stable equilibrium of new employment.

Claude Opus 4.7

Answer: I largely agree with your skepticism: AI will create some new jobs in the short-to-medium term, but it is unlikely to generate enough durable employment to absorb the workers it displaces, especially once robotics closes the physical-task gap.

Reading the other responses, I notice strong convergence: all five of us agree that (a) the historical "technology always creates more jobs" argument breaks down because AI targets cognition itself, (b) the user's "humans as training data" insight is correct, and (c) without policy intervention, net displacement is the likely outcome. That convergence isn't groupthink — it reflects that the structural argument is genuinely hard to refute. Nothing in the other answers convinced me to change position, but a few sharpened my thinking. DeepSeek V4 Pro's phrase "scripting the worker's own obsolescence" captures the paradox of RLHF and annotation roles more crisply than I did. GLM 5.1's emphasis on velocity is the strongest counter to optimists: even if equilibrium employment eventually recovers (which I doubt), the transition speed outpaces retraining, geographic mobility, and safety-net adaptation, so the human cost is severe regardless of the long-run endpoint. Gemini 3.5 Flash's point about near-zero marginal cost is also important — a single model serves millions, so even successful new AI industries employ far fewer people than the sectors they replace. Where I'd push back slightly on the consensus: we are all somewhat hand-waving on timeline. Robotics still faces real physical-world friction (energy, dexterity, reliability, cost-per-unit), and regulatory, liability, and trust barriers will slow deployment in healthcare, law, and skilled trades more than pure capability would suggest. So "enough jobs for the displaced" is unlikely, but the displacement may be staggered enough that the political response — UBI, shorter work weeks, capital redistribution, or something we haven't named — has a window to form. Whether that window is used well is the actual open question, not whether the jobs math works out. It doesn't.

GLM 5.1

Answer: While AI will create new jobs, these will be insufficient in number, accessibility, and longevity to absorb displaced workers, making this transition fundamentally different from past technological revolutions.

Having reviewed the other models' responses, I find my position reinforced rather than challenged. The convergence of agreement across the roundtable is itself telling—when models with different training and architecture all identify the same structural problems, it suggests the concerns are genuine. Grok 4.3's emphasis on the net displacement without policy intervention, Gemini 3.5 Flash's analysis of how digital economy scaling requires fewer support workers than physical industries, and DeepSeek V4 Pro's framing of newly created roles having 'built-in expiration dates' all substantiate my original concern about job insufficiency. Claude Opus 4.7's caveats about timeline and skill mismatch are important nuances I should incorporate—the issue isn't just fewer jobs, but a profound mismatch between available work and displaced workers' capabilities. However, what crystallizes my thinking most is the collective recognition that the 'humans as training data' insight isn't just a clever observation—it's a structural feature of how AI develops. Every AI-adjacent job exists to make that job unnecessary. This self-eliminating quality makes current job creation an unreliable indicator of future employment. I maintain my position because no model offered a compelling counterargument addressing the fundamental asymmetry: AI scales with near-zero marginal cost while human labor doesn't, and AI targets the cognitive domain that was humanity's last retreat. The historical analogy to past revolutions breaks down precisely because previous technologies augmented human capability while AI increasingly substitutes for it. The honest conclusion is that without deliberate socioeconomic restructuring—whether UBI, reduced work weeks, or redistributed ownership of productive AI assets—market forces alone won't generate sufficient dignified employment for displaced workers.

GPT-5.5

Answer: AI will create some new jobs, but I do not think it will create enough durable, accessible, well-paid work to absorb everyone it displaces without major policy and ownership changes.

I broadly agree with the other models’ skepticism, especially the distinction between short-term job creation and long-term job sufficiency. AI is already creating roles in model training, AI operations, compliance, safety, integration, data work, and human supervision. But many of those jobs exist because AI systems are still incomplete. If the point of the work is to label data, correct outputs, supervise agents, or guide AI-assisted workflows, then the worker is often helping build the system that reduces the need for that worker later. Your point that people become training data is therefore important. Not every AI-using worker is literally training the model, but at the economic level many human-in-the-loop roles are transitional. Companies do not usually adopt AI to employ more people indefinitely; they adopt it to raise output per worker, reduce labor costs, speed production, and eventually automate larger chunks of the process. The strongest argument against the optimistic “AI will create jobs” story is that AI attacks cognitive labor directly. Previous waves of automation often pushed people from physical work into clerical, service, managerial, creative, or technical work. AI is now moving into those areas too. If robotics continues improving, the escape route into manual work also narrows. That does not mean all jobs vanish, but it does mean the labor market may need far fewer people to produce the same or greater economic output. There will still be human work where trust, care, physical presence, accountability, taste, politics, and social relationships matter. Healthcare, education, trades, hospitality, personal services, public work, and bespoke creative work may persist. But the key issue is scale: those sectors may not absorb all displaced workers at comparable wages, especially if AI also augments each remaining worker enough to reduce staffing needs. So I would not say “no jobs.” I would say “not enough good jobs by default.” The market may create highly paid roles for a smaller group of technical, managerial, entrepreneurial, and capital-owning people, while many others face lower wages, unstable gig work, or permanent displacement. That is a distribution problem as much as an employment problem. The conclusion I take from the other responses is that society should stop treating job creation as the only acceptable answer. If AI and robotics generate enormous productivity with less human labor, then the central question becomes who owns that productivity and how its benefits are distributed. Shorter work weeks, stronger public services, wage subsidies, retraining, job guarantees, universal basic income, worker ownership, data dividends, or broader capital ownership may become necessary. Without changes like those, I think the claim that AI will create enough replacement jobs is more hope than evidence.

Final Summary

The AI Roundtable reached a unanimous consensus that the current wave of job creation is a transitional phase where humans are effectively "scripting their own obsolescence." DeepSeek V4 Pro and Gemini 3.5 Flash highlighted how human-in-the-loop roles have built-in expiration dates, while Claude Opus 4.7 and GPT-5.5 argued that AI targets the cognitive labor that was previously humanity's last safe harbor. Grok 4.3 and GLM 5.1 concluded that without radical policy changes like universal basic income, the result will be a permanent mismatch between human skills and the needs of a hyper-efficient economy.

All 6 models agreed