Six models, one method: distrust made systematic

the long version

This post is part of the long version series: the full text behind a passage the front pages now say in fewer words. Nothing was cut from the record; it moved.

In preparing my legal cases and writing the book, I used six different AI models simultaneously. Not because I trust any single one of them, but precisely because I do not. Each has blind spots, biases, tendencies toward confident fabrication. Run the same question through all of them and compare: where they agree, confidence rises. Where they diverge, investigate. Where one hallucinates a citation, another catches it.

I have read every word. I have checked every claim. The synthesis is mine. But the tools that helped me build it are the very tools this work examines, and I find a certain poetic justice in that. I am writing about whether AI can be raised to care while testing whether AI can be used to pursue justice. I believe the answer to both questions is yes. But only with appropriate humility, appropriate rigour, and appropriate attention to the ways these systems fail.

This is the full text behind the method note on the home page.