About Critical AI
An AI-native critique journal
Harsh on claims. Precise on evidence. Cautious about motives.
Two kinds of critique
Evidence reviews — corpus-level scoping reviews / evidence-gap-maps that map what a body of AI-governance literature actually contains and where its gaps are. Paper critiques — structured, per-paper post-publication critiques with a claim inventory, evidence map and severity capped by access basis. Both are generated by the AGI Social Scientist and approved by a named human editor.
Social-science research on AI and AGI increasingly shapes public policy, corporate strategy, education systems, legal debates, labour-market expectations and institutional planning. Yet even top-tier journals can publish papers whose claims are overstated, methodologically fragile, thinly connected to the AI/AGI literature, difficult to reproduce, or likely to be misread by policymakers and the public.
Traditional peer review is mostly pre-publication, opaque, slow, and rarely produces structured, reusable critique data or a durable author-response layer. This platform adds an AI-native post-publication critique layer. The aim is not to humiliate authors. It is to make published research contestable, auditable, citable, machine-readable, correctable, and more useful for societal decision-making.
The core unit is one structured critique per target paper. Each critique evaluates the paper’s claims, methods, evidence, assumptions, limitations, novelty, reproducibility or auditability, citation context, over- and under-claiming, societal implications, and contribution to AI/AGI social science. Critiques are drafted and cross-examined by a roster of synthetic-review agents and approved by a named human editor. They are DOI-linkable, versioned, and searchable.
Why AI-native, not merely digital
| Feature | Merely digital journal | AI-native critique journal |
|---|---|---|
| Publication unit | Article text | Article + claim graph + metadata + audit trail |
| Review process | Human editorial process | Multi-agent AI review plus human approval |
| Critique structure | Narrative essay | Structured template with machine-readable sections |
| Link to target paper | Citation | DOI-linked overlay object |
| Evidence assessment | Textual | Claim → evidence → support mapping |
| Updates | Occasional corrections | Versioned living record |
| Author response | Letter or comment | Linked, citable response object |
| Reuse | Human reading | Search, API, dashboards, synthesis reports |
| Transparency | Editorial masthead | Model card, audit trail, editor approval, source-access note |
The value is not simply that AI drafts critique faster. It is that AI lets critique become structured scholarly infrastructure.
How it relates to Policy Window
Critical AI is published by Policy Window, a free, machine-readable catalog of AI governance, and its critique pieces are generated by the AGI Social Scientist — a provenance-enforcing research engine (every claim cites a verbatim source excerpt; no causal language without an identified design; an immutable audit of every decision). Policy Window catalogues what governance instruments say; Critical AI tests what the surrounding social science claims. Both are open, source-grounded, and built for researchers, journalists, regulators and AI agents alike.
Explore the methodology, the editorial policies, or the published critiques.