The AI Canon is a free, public reference library of the texts that define artificial intelligence, built on an open method that lets anyone check, question, and overturn its judgments. It ranks texts, not people. It sells nothing. It is built by Jeroen Janssen, founder of the Dutch AI governance firm Apparens.
Why it is worth covering
The literature of AI has outgrown anyone's ability to read it, and the maps that exist are mostly commercial: affiliate reading lists, vendor guides, influencer rankings. They ask the reader to trust the curator. Almost none show their work. The AI Canon is built the opposite way. Every ranking is produced by a published method, every number carries its source and date, and anyone can download the audit file and rebuild the result themselves. The premise is that curation of knowledge can be made auditable, the way an account can be audited, rather than taken on faith.
What is genuinely new here, and verifiable
Three things, each checkable rather than asserted. First, it ranks texts and refuses to rank people: the voices and organizations in the field are described, never scored, and the data model has no way to rank a human being. Second, it is checkable end to end: the method, the corpus, the weights, and the audit files are public, and every rank links to the evidence that produced it. Third, it invites correction as a feature, not a complaint box: anyone can formally challenge any ranking or omission with evidence, and every challenge and its resolution is published in a permanent, public log.
The China and United States angle
Most maps of AI thought only see half the field. You cannot understand artificial intelligence in 2026 by reading only what was written in English. The AI Canon is built, as a published rule, to score Chinese-language works within their own publishing and citation ecosystem before any cross-language comparison, rather than against English metrics that would erase them. That rule is written into the method; the mechanism that enforces it, and the Chinese corpus it needs, are still being built. Today the scored pilot is English-language papers, and the Chinese section is thin and openly under construction, with the project actively recruiting Chinese-literate scholars and readers to help build and verify it. The story is not a finished global canon. It is a Western-built reference work that is structurally committed to including China and is openly asking Chinese experts to help, at a moment when most Western and Chinese AI discourse barely acknowledge each other.
Quotable, attributable to Jeroen Janssen
Every AI reading list asks you to trust the curator. This one asks you to check the math.
I will rank texts. I will not rank human beings, and the system is built so I cannot start.
A canon you can check is worth more than a canon you must believe.
You cannot understand AI today by reading only what was written in English. So the Chinese works go in the spine, not the appendix.
What you can verify before you publish
The method, the ontology, the full corpus, and the audit files are open. The challenge log is public. There is no advertising, no affiliate income, and no paid placement anywhere in the project, by design and by rule. It is a non-commercial public good; there is nothing to buy and no upsell to find.
The builder's own book, The AI Accountability Trap, is in the corpus and carries a visible conflict flag, scored by the same rules as everything else, with no exemption and no boost.
Contact
Jeroen Janssen, Apparens (Deventer, Netherlands). office@apparens.nl.
If you want to share this rather than write about it, see the share page.