LinkedIn tests AI training labour marketplace
Microsoft-owned platform pays experts up to $150 an hour to label model outputs, public AI anxiety grows as the work gets industrialised
Images
ScreenshotImage Credits:Pew Research data, via Stanford
Image Credits:Pew Research data, via Stanford
ScreenshotImage Credits:Ipsos data, via Stanford
Image Credits:Ipsos data, via Stanford
ScreenshotImage Credits:Ipsos data, via Stanford
Image Credits:Ipsos data, via Stanford
Sarah Perez
techcrunch.com
LinkedIn is testing an “AI labor marketplace” that pays specialists up to $150 an hour to label and critique data for model training, a move that would place Microsoft’s social-networking unit in direct competition with firms such as Scale AI and Mercor.
According to Business Insider, LinkedIn has been running early pilots where contributors are hired to train AI on domain tasks including coding, nursing and finance. The pitch is familiar across the sector: language models improve when they are fed high-quality examples and when human reviewers score outputs for accuracy, safety and style. What is changing is who owns the funnel. LinkedIn already sits on verified work histories, professional credentials and a messaging layer that can route offers at scale; turning that graph into a paid labor pool shifts a cost center in AI development into a platform product.
The timing also intersects with a widening gap between the people building AI systems and the people paying their bills. TechCrunch reports that Stanford University’s annual AI index finds public sentiment turning more negative even as expert optimism remains high. The report cites Pew Research data showing only a small minority of Americans feel more excited than concerned about AI, while a majority of AI experts expect net positive effects on the economy and work. The same survey split appears in specific domains: experts rate AI’s likely impact on medical care far more positively than the general public does.
That divergence matters for the economics of AI because the visible costs are no longer abstract. Model training is increasingly constrained by scarce inputs—specialised human feedback, GPUs, electricity and data-centre capacity—while the benefits are unevenly distributed across firms and job categories. As companies compete for the same narrow band of high-skill annotators, wages rise; when platforms intermediate the market, they can standardise tasks, compress negotiating power and take a cut. LinkedIn’s entry suggests the industry expects this work to remain large and recurring, not a temporary bootstrap phase.
For LinkedIn, the bet is that its identity and credential signals can reduce fraud and improve quality in a market plagued by questionable qualifications and opaque vendor chains. For AI developers, it offers a single counterparty with reach and compliance infrastructure. For workers, it is another reminder that the “AI economy” includes not only engineers and executives but also a growing class of remote piecework—often well-paid at the top end, and increasingly routinised as the platforms learn which judgments can be turned into templates.
LinkedIn has not said when a broader rollout would occur. But the company is already posting the price of expert attention in public: up to $150 an hour, for the humans teaching machines what “good” looks like.