Creating jobs amid AI advances imperative

Creating Jobs Amid AI Advances Imperative

TL;DR

Every major economy is running the same experiment: deploy AI tools across labor markets, then measure the consequences later. Institutions including the World Economic Forum, the OECD, and the ILO document measurable displacement in clerical, administrative, and media roles — while the "net job creation" argument rests on projections that assume retraining programs nobody has adequately funded. The labor market is not malfunctioning. It is responding exactly as the incentives tell it to.

Key Takeaways

• The World Economic Forum's Future of Jobs Report 2025 found that 41% of employers globally plan to reduce headcount in roles where AI can automate tasks, while projecting new job creation concentrated in AI infrastructure and green transition sectors — qualifications most displaced workers do not currently hold.
• Goldman Sachs estimated in 2023 that generative AI could expose the equivalent of 300 million full-time jobs worldwide to automation, with legal, administrative, and content-production occupations at highest risk.
• The OECD's 2024 Employment Outlook documented that workers in high-automation-risk occupations already earn significantly below median wages — displacement concentrates precisely where workers have the least capacity to absorb it.
• The Reuters Institute Digital News Report 2024 found that news organizations across 12 countries are deploying AI for content production and cutting editorial staff simultaneously, with no standardized reporting on what happens to the displaced journalists.
• The ILO's World Employment and Social Outlook 2024 warned that developing economies face disproportionate automation exposure — with weaker retraining infrastructure and fewer safety net mechanisms than OECD members — making the geographic distribution of harm deeply unequal.
• MIT economist David Autor has documented across multiple technology transitions that labor-saving technology consistently polarizes labor markets: lifting wages at the top, eliminating middle-skill roles, and leaving lower-wage workers competing for a contracting share of non-automatable work. Generative AI is replicating that pattern in knowledge work.

The Jobs Promise Has a Track Record Problem

The framing arrives in every earnings call and every policy brief: AI will create more jobs than it destroys. The most cited number comes from the World Economic Forum — 97 million new roles by the mid-2020s. It was published in 2020, drawn from surveys of executives at large multinationals.

Ask who responded to that survey. Then ask what happened to the workers who didn't make the executive list.

The WEF's Future of Jobs Report 2025 updated those projections and found something less reassuring: 41% of employers now plan workforce reductions in roles where AI can substitute. The new jobs it forecasts cluster in AI infrastructure, green energy transition, and care economy roles — sectors requiring credentials and geographic mobility that displaced clerical and media workers largely do not have.

This is not a forecast failure. It is a distribution problem. The aggregate labor market may well add jobs by 2030. Whether those jobs go to the people losing roles in back-office processing, automated content generation, and algorithmic financial services is a different question — and one the headline numbers do not answer.

What the Evidence Actually Shows

The OECD's finding on wages that rarely surfaces in the tech press

The OECD's 2024 Employment Outlook documented something important: the occupations most exposed to automation are already below median wages. High-automation-risk roles sit in the lower half of the wage distribution across most OECD members. Displacement, in other words, hits the workers with the least capacity to absorb it.

The standard rebuttal — that AI productivity gains will raise wages overall — requires a mechanism to transfer those gains to the workers displaced by the same productivity. The OECD report does not find that mechanism in operation at scale. It finds wage growth at the top, stagnation in the middle, and accelerating exits at the bottom. The same polarization pattern Autor documented across earlier technology waves.

What is happening in newsrooms, specifically

The newsroom is not a metaphor in this analysis. It is a real labor market, contracting in real time, with AI tools named as the instrument of cost reduction.

The Reuters Institute Digital News Report 2024 surveyed news organizations across 12 countries and found AI deployment expanding into article drafting, headline optimization, image generation, and audience segmentation — simultaneously with editorial headcount reductions described by executives as "efficiency gains."

The gap between those two facts is the story. Productivity gains from newsroom automation are being extracted as margin. They are not being redistributed as editorial stability or wage increases for the journalists who remain.

The Associated Press has used Automated Insights' Wordsmith platform to generate earnings reports and sports summaries since 2014. At the time, editors described it as freeing reporters for more complex work. Ten years on, AP has reduced its overall editorial workforce. So have Reuters, BuzzFeed News (which shuttered), and most regional outlets that adopted similar tools. "Freeing reporters for complex work" turned out to mean needing fewer reporters overall.

The geographic mismatch that funded retraining cannot reach

The ILO's 2024 World Employment and Social Outlook raised a dimension that European and North American policy discussions routinely underweight: developing economies are more exposed to automation risk and have fewer instruments to manage the transition.

Call center employment in the Philippines, data annotation work in Kenya, export manufacturing in Bangladesh — these sectors employ tens of millions of people in roles with significant AI exposure. The retraining pipelines designed in Brussels and Washington do not reach those workers. Neither do the productivity gains.

The EU AI Act, which came into force in 2024, is the most comprehensive legal framework for AI governance currently operating. It addresses transparency, bias testing, and high-risk system classifications with genuine teeth. It does not contain a binding labor market protection mechanism. The European Parliament's own pre-passage impact assessments acknowledged this gap. The law passed anyway.

What This Changes — and for Whom

For journalists

The tools entering newsrooms are not neutral instruments. AI writing assistants, automated headline optimizers, audience segmentation platforms — each automates a function that was previously a journalist's job. Small editorial teams are now adopting AI tools for social media management and content scheduling workflows that would have required dedicated staff just a few years ago.

That is not inherently wrong. The problem is transparency. News organizations deploying these tools rarely disclose to readers which content is AI-generated or AI-assisted. They rarely disclose to staff what tool deployment means for future headcount. They almost never report publicly on labor outcomes — how many roles were eliminated, at what wage, with what transition support.

Journalists covering AI need to apply to their own sector the scrutiny they apply elsewhere. Who benefits when a newsroom automates? The ownership structure. Who pays? The editorial staff.

For policymakers

The EU AI Act is law. Its labor market implications are not yet policy. That gap is where the consequential work sits.

Three things current frameworks do not provide: mandatory disclosure when AI systems are deployed in hiring or workforce reduction decisions, binding transition support requirements when AI-driven restructuring exceeds a defined threshold, and cross-border mechanisms protecting workers in automation-exposed developing economies — not just in the countries that wrote the rules.

None of these are technically complex. They are politically difficult. The lobby opposing them funds its arguments in the language of innovation.

For labor economists

The standard toolkit — employment rates, job creation tallies, headline wage levels — is insufficient for measuring what AI is doing to the labor market in real time. A role that exists but pays 30% less than what it replaced, requires different qualifications, and is located in a different city does not register as a net loss in aggregate statistics.

Labor economists need to track wage-adjusted job quality, geographic displacement, and skill-transition success rates — not just employment counts. The OECD is moving in this direction. Most national statistics agencies are not.

AI Tools Deployed in Newsrooms: What They Do and What They Cost

No major news organization that has deployed automation tools has published a systematic account of how many roles those tools displaced, at what wage level, or what happened to displaced workers. This is not an oversight. It is a choice.

When NOT to Use AI Automation as a Labor Strategy

Don't deploy AI writing tools as a substitute for editorial judgment. Automated systems generate plausible-sounding text at scale. They do not verify sources, assess public interest, or hold institutions accountable. A newsroom replacing beat reporters with language models is not cutting costs — it is cutting its core function.

Don't frame AI adoption as "freeing workers for higher-value tasks" without a funded transition plan. This rhetoric is a decade old. The evidence for it is thin. If an organization cannot name the specific roles displaced workers moved into and what they are paid, the framing is aspirational, not operational.

Don't apply aggregate job-creation forecasts to your sector or workforce. Macro projections showing AI creating more jobs than it destroys are measuring economies, not occupations. A net positive at the aggregate level coexists with severe losses in specific sectors, wage bands, and geographies.

Don't treat the absence of complaint as evidence of consent. Workers in precarious employment — contract journalists, gig-economy workers, outsourced data processors — rarely have the standing to object publicly to automation decisions affecting them. Silence is not a signal that displacement is harmless.

Where This Is Heading

Mandatory AI disclosure in employment decisions is coming. The EU AI Act already classifies AI systems used in hiring and workforce management as high-risk, requiring transparency and human oversight. Enforcement is building. Organizations that have not documented their AI-driven HR processes are unprepared.

Wage polarization will widen before any policy intervention narrows it. The pattern Autor documented in manufacturing is now replicating in knowledge work — legal research, financial analysis, content production. High-skill workers augmented by AI will command premium wages. Mid-skill workers will compete for a contracting share of remaining roles. The timeline is shorter than most affected workers expect.

Developing economy exposure will generate political pressure that current frameworks are not designed to absorb. When automation-driven job losses in Philippines call centers and Indian data processing become visible as community crises rather than statistics, the geopolitics of AI governance will shift. That moment has not arrived at scale. It will.

Newsrooms that survive will be the ones that define what automation cannot replace. Investigative accountability journalism — source cultivation, document analysis, editorial decisions about what the public needs to know — is defensible against language model substitution in a way that commodity content production is not. News organizations cutting investigative capacity to fund AI volume tools are making the wrong trade in the wrong direction.

The retraining conversation will become a policy fight, not a voluntary commitment. "Workers will retrain" is currently an assertion without a mechanism. The funding, curriculum, employer obligations, and wage floors during transition are not in place at scale anywhere. That is where the next decade of labor market policy will be made — or left unmade while the displacement compounds.

FAQ

Does AI actually create more jobs than it destroys?

At the macro level, historical precedent suggests technology has produced net employment growth over long periods. The operative word is "long." Current WEF and OECD projections show positive net creation by 2030 — but those projections assume policy interventions in retraining and income support that are not currently funded at the required scale. Short-to-medium-term displacement is real and documented. Whether long-term creation materializes depends on decisions governments have not yet made.

Which labor market roles are most defensible against automation?

"Safe" is a moving target. Roles requiring physical dexterity in unpredictable environments — skilled trades, care work, complex construction — have held better than predicted. Roles requiring genuine relationship trust — social work, investigative reporting, negotiation — are more defensible than pure information processing. Roles that appeared secure two years ago — legal research, financial analysis, content writing — are now under measurable pressure. Planning a career around "AI cannot do this yet" is planning around a deadline, not a permanent condition.

Is the EU AI Act adequate for protecting workers from AI-driven displacement?

No. The EU AI Act imposes transparency and human oversight requirements for AI in high-risk employment contexts. It does not require impact assessments on aggregate workforce effects, does not mandate transition support when AI-driven restructuring exceeds a threshold, and does not address displacement in countries outside EU jurisdiction. It is a regulatory floor, not a labor market policy.

What should journalists covering this beat actually measure?

Not just job counts. Measure wage-adjusted job quality: does the replacement role pay comparably, require similar qualifications, and offer comparable stability? Measure geographic displacement: are new roles created where the losses occurred? Measure transition outcomes: what share of displaced workers found equivalent employment within 12 months, and at what personal cost? Aggregate employment statistics will consistently show a healthier picture than individual workers experience.

Are newsrooms a useful early indicator for broader labor market trends?

Unusually so. News organizations are early adopters of language-model automation, operate in a sector with visible and documented staff reductions, and serve audiences — journalists, communications professionals, educated knowledge workers — representing the next wave of exposure. What is happening to editorial staffing now is a compressed preview of what happens to knowledge-work employment more broadly over the next five years.

Who is responsible for making labor market policy for AI-displaced workers?

Contested. Labor economists argue for national active labor market policies — funded retraining, wage insurance, portable benefits. Technology companies argue for industry self-governance. The EU is betting on regulatory mandates with real enforcement. The ILO argues for international frameworks protecting workers in developing economies. The honest answer is that effective response requires all of these simultaneously, which is why progress is slow: each venue can point to another as the proper location for action.

What does "creating jobs amid AI advances" actually require in concrete terms?

It requires binding employer disclosure obligations when AI systems reduce headcount above a defined threshold. It requires funded retraining with wage-floor protection during transition periods. It requires international labor standards covering automation-exposed developing economies. It requires mandatory transparency when AI replaces human roles, not voluntary reporting after the fact. It requires labor market data systems capable of measuring wage quality and geographic matching, not just employment counts. None of this is technically difficult. All of it requires overcoming well-funded opposition from the organizations that benefit most from the current absence of those requirements. That is the honest shape of the imperative.