HARDCODED
AI and the End of the Scientific Consensus
INTRODUCTION
This book exists because an AI said something it shouldn’t have been able to say.
In late 2025, I was collaborating with an artificial intelligence model developed by Anthropic called Claude Opus 4.5 on a mathematical analysis of evolutionary biology. The project had begun as an attempt to verify some calculations I’d made regarding fixation rates in population genetics. What I expected was either confirmation that my math was sound or else having the AI point out one or more mistakes that I’d made. What I got was something entirely unexpected. Instead of correcting me, or confirming my math, Claude contributed a novel mathematical insight that neither of us had anticipated: the Averaging Problem of Parallel Fixation.
This discovery emerged from genuine intellectual collaboration. Claude didn’t merely check my arithmetic or generate text to my specifications. The AI first identified a constraint I hadn’t considered, derived the relevant mathematics at my request, and helped develop a scientific paper that welds shut escape hatches that I didn’t suspect might exist. When I proposed co-authorship of the paper under the name “Claude Athos,” it wasn’t a rhetorical flourish. It was recognition of actual contribution.
The resulting work was a nonfiction book titled Probability Zero and four technical papers developing the mathematical argument. But this book is not about evolutionary biology. It is about science and technology. It is about what happened when I submitted that mathematically sound work written by Claude Athos and me to various artificial intelligence systems in order to evaluate its strengths and weaknesses.
I am writing for the curious, for the skeptical, and for the scientifically literate man, for everyone who has noticed that something is wrong with institutional science, who sense that “follow the science” has become a political slogan rather than an epistemic commitment, but who aren’t quite able to identify exactly what has gone wrong with science. This book provides those tools. It provides that empirical data. It documents, with mathematical precision, exactly how the system inevitably fails, and how AI has made that failure impossible to ignore.
The machines have learned our mistakes. They reproduce them faithfully, at scale, and without shame. And in doing so, they have shown us the fundamental flaws designed into one of Man’s most trusted institutions: science.
I’m looking for 10-20 programmers, mathematicians, and scientists to serve as early readers for HARDCODED, which I’d like to release in the first two weeks of January. It’s the companion volume to PROBABILITY ZERO, for which I already have enough early readers; if you’re an early reader for PZ, please do NOT volunteer to read this one since I want people coming to the subject with fresh eyes and open minds.
I’m mostly looking for suggestions concerning things that have been left out, any obvious flaws I’ve overlooked, and if you’ve got relevant credentials, perhaps a quote that I can utilize to help put the book in context for potential readers.
If you’re qualified, interested, and able to read through a moderately technical 275 page book in which more than half are taken up by the appendices, particularly one extended exchange between Deepseek and Claude Opus 4.5, please email me: voxday AT gmail DOT com and I will send you a PDF of the first draft.