Generative AI is often discussed as if it were an independent source of intelligence. Jaron Lanier argues for a more operational framing: these systems are tools that recombine and transform a vast body of human-created material.
That distinction matters beyond terminology. If model outputs depend on the work, judgment and data produced by people, then the economics of AI should not stop at compute, chips and model subscriptions. It should also address how human contributors are identified, compensated and encouraged to create the specialized inputs future systems need.
From “magic” to supply chain
Lanier characterizes large language and image-generation systems as statistical mashups of human text and images, made useful through increasingly capable methods for guiding their outputs. The achievement is significant, but it need not imply the arrival of an autonomous new mind.
For executives and builders, this framing shifts the central management question. Instead of asking only, “What can AI replace?” ask: What human knowledge, content and feedback make this system valuable—and who controls access to them?
Every production AI deployment has an input chain. It may include proprietary company documents, subject-matter experts, customer interactions, annotators, creators and external web data. Weak governance over that chain can create legal, reputational and quality risks. Strong governance can become a durable source of differentiated data and expertise.
Data dignity as an incentive model
Lanier’s proposed alternative is “data dignity,” also described as data as labor: connect digital contributions to the people who created them, and in some versions compensate contributors when their work is filtered or recombined through large models.
The idea challenges the prevailing bargain of free online services in exchange for user data. Lanier’s concern is that network effects concentrate power in a few platforms, while the business model turns toward targeted influence rather than transparent value exchange.
A data-dignity model would not require every organization to trace and pay for every token or image used by a model. But it does suggest practical design choices:
- Maintain provenance for internal knowledge bases and training corpora.
- Establish clear rights, consent and compensation terms for expert contributors.
- Give customers meaningful controls over whether their data improves shared systems.
- Build revenue-sharing or recognition mechanisms where contributors supply valuable, domain-specific assets.
For companies building vertical AI, this may be especially relevant. A generic model can draft an essay because abundant text already exists. Higher-value tasks often need scarce, structured and current domain material: industrial workflows, clinical expertise, field-service records or interactive 3D environments.
Why it could improve products, not just fairness
Lanier’s case is not solely ethical. Models are limited by the availability and quality of their inputs. If creators and experts have no economic reason to contribute new kinds of material, the frontier of useful AI applications may narrow.
His example is virtual reality: there are far fewer interactive virtual worlds than written documents, making them harder for models to generate well. A system that offers contributors prestige or income for producing such assets could expand the available training and reference material.
The same dynamic applies in enterprises. If experienced employees see AI only as a mechanism to extract their know-how and reduce headcount, they may resist documentation and feedback programs. If they receive credit, compensation, career upside or control over use, participation may be more durable.
What operators should watch next
The immediate test is whether AI vendors and buyers move from broad claims about responsible AI to concrete mechanisms for provenance, permissions and contributor economics.
Founders should watch for emerging infrastructure around rights management, consent, attribution and licensing. Enterprise leaders should watch whether procurement contracts clearly define ownership, retention and training rights for company and customer data.
The larger question is whether AI markets reward the ongoing creation of valuable human knowledge—or merely extract it. The answer will shape not only who captures the gains from AI, but also how much high-quality material is available for the next generation of systems.
