Digital Cognitive Infrastructure: Who Decided What We Get to See (A KEELEY HALLORAN INVESTIGATION)

Two unaccountable points now decide what an Australian citizen gets to see and hear. The feed decides what spreads, tuned for engagement rather than accuracy. The AI guardrail decides what gets said, tuned for caution rather than transparency. Neither is required to explain itself, and no one has yet proposed a workable, Australian, precedent-grounded way to make either one auditable.

Digital Cognitive Infrastructure: Who Decided What We Get to See does both. It is a trade book making the public case, and, inside the same volume, a fully drafted Commonwealth Bill ready for a Member of Parliament's office to take up: working legislative text, not a general position paper.

The book's core distinction, held firm throughout: safety-critical guardrails, the kind that block weapons synthesis or child exploitation material, exist because the information itself is the harm, and this book leaves those alone entirely. Its target is narrower — the epistemic-caution guardrail, the invisible values judgement made when a system hesitates over a contested but legitimate claim, with no disclosure that a judgement was made at all.

The Bill's mechanisms come from Australia's own regulatory toolkit: vehicle safety standards, the phase-out of leaded petrol, spectrum licensing, telecommunications structural separation, the Auditor-General's independence model, and the plain-packaging framework that reshaped an industry's disclosure obligations. Each major mechanism, the audit authority, enforcement, the funding levy, user control over algorithmic treatment, is drafted at five distinct tiers, from the least disruptive option that still works to the most structurally ambitious, each with full clause-level text. An office can take the tier that fits its political room to move and use it directly.

The book is equally direct about its own construction. Eight AI systems were independently commissioned to produce comparative policy proposals for the same problem, with no visibility into one another's work; their strongest and weakest ideas are assessed on the record, not cherry-picked for effect. A central case study, working from a documented incident, makes a narrow and disciplined argument: a legitimate source was, by default, omitted from an AI-assisted research output. The point isn't that the caution was wrong. It's that the omission was only caught because someone was running a genuinely rigorous verification process, and anyone without that process would never have known it happened. That single, checkable case is the whole justification for the Bill's audit architecture — not proof of malice, proof that invisible caution is indistinguishable from invisible bias unless the system is built to be checked.

Every claim is held to what the authors call the High Court Test: would it survive being read aloud by a judge actively looking for its weakest point? Where the evidence supports a claim, the book says so and cites it. Where it only suggests one, the book says that instead. No chapter allows a system's motive or a contested finding to be asserted as settled fact.

A continuing fictional narrator, Keeley Halloran, carries a single locked personal chapter at the book's opening, establishing the human stakes before the argument proper begins; everything after is a direct, professionally registered discussion paper, cited throughout, with no further narrative device between the reader and the case being made. Readers arriving cold will find a complete, self-contained policy document that assumes nothing and asserts nothing it hasn't shown.

Six appendices close the volume: a Statement of Compatibility, an Explanatory Memorandum, full methodology and bibliography and a Second Reading Speech, giving an MP's office every document it would need to move the Bill forward without a redraft.