There Is No AI: Philip Wadler Makes the Case for Data Dignity

In a recent blog post, computer scientist Philip Wadler (known for his work on Haskell and type theory) argues that what we call "AI" is not actually intelligent. He echoes Jaron Lanier's view that models like GPT-4 are essentially advanced mashups of human-created content, not autonomous minds. Instead of treating them as new forms of life, Wadler suggests we should view them as tools for enhanced collaboration — and restructure the economics to compensate the humans whose work powers them.

Key Arguments from the Post

• AI as mashup: Wadler describes GPT-4 as "something like a version of Wikipedia that includes much more data, mashed together using statistics." Image generators are similar to "online image search" with a system for combining pictures.
• No new mind: The innovation is in the "guided and constrained" mashup process that produces usable results — but it"s not the invention of a new mind. It "illuminates previously hidden concordances between human creations."
• Data dignity: Wadler proposes a concept he"s contributed to developing: "data dignity" (also called "data as labor" or "plurality research"). In this model, digital content would be linked to the humans who created it, and they could get paid when their work is used in AI models.

The Problem with "Free" Online Services

Wadler points out that the current arrangement — where users give data for free in exchange for free services — has "turned out to have a dark side." Network effects allow a few platforms to dominate, and because the experience must be free, the only business left is "hawking influence." Users are targeted by "stealthy and addictive algorithms" that make them "vain, irritable, and paranoid."

A Concrete Example: Tree-Trimmed robots

Wadler gives a practical scenario: if AI-driven tree-trimming robots replace human workers, those workers could be devalued. But with data dignity, new creative fields like "holographic topiary" could emerge, and the models could distribute income through collective organizations. Each new AI application could "inaugurate a new kind of creative work."

Why AI Companies Should Care

Wadler notes a non-altruistic reason for companies to adopt data dignity: models are only as good as their inputs. To expand AI into new frontiers (like generating interactive VR worlds), we need more examples of those worlds — and paying creators is a way to get them. It"s "much easier to get an LLM to write an essay than to generate a virtual-reality world" because there are few VR worlds in existence.

This isn"t a practical guide for developers, but a thought-provoking argument for how we might reshape the AI economy. If you"re building with AI and wondering about compensation models or the ethics of training data, Wadler"s post is worth a read.