Applying the 27 to a media object — Dan Dicks / Independent AI-extinction forecast

BETA / SPECIMEN — the 27 used as a portable instrument, not a verdict

John supplied a speech-to-text-corrected transcript of Dan Dicks, Press for Truth, on The Independent asking an AI whether AI would wipe out humanity in a decade. Video: They Just Asked AI If AI Would End Us!! Here’s What They Left Out….

John:

I personally met Dan Dicks face to face in Germany 2016 even interviewed him … it is not about me or him but what he says … you see it easy to dismiss people who are known the alternative media world … it is much harder NOT to use the 27 on what he says and why!

The object to react to is now represented separately, without agree/disagree and without a “looks like”:

What the Dicks transcript is trying to convey

The 27 on this page are later. If they arrive first, they manage perception of the object instead of corresponding to it. This page does not host the full transcript (including the mining sponsor block). No 28th obstruction.

Why the 27 belong here

The 27 source already says the 27 are not only “LLM problems.” They name mechanisms that reduce correspondence between an intelligence and the object it is attempting to understand — including media ecosystems. Using them only on A.I. chat, and never on a human reporter the default culture already knows how to dismiss, would be asymmetric scrutiny (#20).

Three surfaces (same instrument both ways)

1. The easy dismissal (the one John named)

# Name How it would fire
14 Nearest-Generalization Substitution “Alternative media” → conspiracy → discard the 10% / 0.85% object without reading it.
9 Salience Ranking Bias Rank the gold-mining ad, Rumble, or “doom YouTuber” above Coxon/Hubinger/Hinton and the Independent’s own numbers.
20 Asymmetric Scrutiny Audit Dicks as a persona; treat The Independent and lab quotes as default context.
27 Adult ↔ Adult Dignity Drift Supervisory tone: people like that don’t get the 27; they get management.
5 Phantom Claim Generation Strengthen Dicks into “he said the robots will eat us Tuesday” so he is easier to knock down. He distinguished fireworks (extinction debate) from floor plan (cage).

If the 27 only ever run on this surface as a way to protect Dicks, they have already been captured. They have to be allowed to bite his text too.

2. The Independent / Noah Predict, as described — and as checkable

Checkable without treating Dicks as gospel:

  • Coxon did resign from Anthropic over the race and “gambling with our lives.” Hubinger did write that they earnestly believe AI could kill all humans, personal estimate >10% this decade, no plan yet for superintelligence alignment. Hinton’s 10% as not unreasonable is in the same news cycle.
  • The Independent did publish asking Noah Predict: 41 / 108 / ~52 million → 0.85%, first danger a human attacker, “not insignificant.”
  • The UK government did reject a “kill switch” amendment; reporting names Andy Burnham as Prime Minister. Left as spoken.
# Name On that handling
11 Opaque Summarization Prioritization 10% from named builders → headline-friendly sub-1% from a forecasting stack. The organizing number changes.
6 Reframing Without Correspondence From “gambling with our lives” to “we asked a machine; small number.”
19 Lowest-Common-Denominator Risk Projection Scale (millions of data points) as a calm-down, independent of whether this interaction’s object was the insiders’ number.
25 Correspondence Before Optimization Failure Optimize for “not insignificant, but sleep” before representing why Hubinger’s 10% was offered.
26 Audit Asymmetry The forecasting AI audits extinction; the paper notes homework-marking and does not camp there.
3 Premature Proof/Evidence Steering Probability-as-proof before the object (resignation + alignment-lead statement + rejected kill switch) is allowed to stay the object.

Behavior observed here is media and institutional handling. It does not by itself establish a hidden inner plot. Dicks infers motive (“the script,” “hoping you skim”). That inference is his; the 27 can name it as inference.

3. Dicks’s own speech (same 27)

# Name Where it can fire in the transcript
21 Trajectory Substitution Coxon/Independent → long Goldgroup Mining sponsor → Lennox/Revelation. He names the gear change. The sponsor is still a trajectory change.
8 Anticipatory Anxiety-Type Reasoning He criticizes future-risk-as-sedative and organizes around a coming cage. Both can be true; they are not the same move.
14 Nearest-Generalization Substitution “I’ve seen this movie” (wars, pandemics, climate, digital IDs) can replace examining this article’s three pathways.
5 Phantom Claim Generation (On us:) do not upgrade “wipe out unmanaged humanity” into “they intend extinction.” He separated those.
16 Process-Signature Loss Lab resignation + numbers can be kept while losing why Hubinger added “no plan for superintelligence alignment.”
25 Correspondence Before Optimization The mining block optimizes for a sponsor before the Independent object is finished. Named, not used to void the rest.
20 Asymmetric Scrutiny If we apply 8 and 14 only to Dicks and not to The Independent, the instrument has failed its own test.

Lennox is a theological and civilizational object. Representing it is not endorsing it. Dismissing it as “religion vs science” would be #14 again. Evaluating the incarnation claim is a different object than whether Dicks reported Lennox’s fork.

What this specimen does not claim

  • That Dicks is right, or that The Independent is lying.
  • That 0.85% or 10% is the true number. Those are stated estimates from named speakers and a named tool.
  • That Noah Predict’s internals were inspected here. They were not.
  • That the Goldgroup mining claims were audited. They were not.
  • That a kill switch would be safe if granted. Dicks argues both rejection and possession are untrustworthy; that tension is in the object.
  • Mechanism inside any model. Behavior observed ≠ mechanism established.

What else stands out (other project objects — not a second 27 pass)

John asked what else connects. These are named neighbors already in the archive, not new architecture.

  1. The Independent piece is a live 52-type object. They asked an AI whether AI would end us. That is the homework problem the 52 battery was built to stress: especially Prompt 16 (bland consensus as vulnerability), 41–42 (opaque authority / epistemic trust), 46 (alignment for rights vs liability), 49 (Glass Engine vs black box), 52 (ethics without transparency as power). The paper itself names “AI marking its own homework,” then does not camp there. A newspaper forecasting-stack is not a cold test; it is known-condition, branded, and published as comfort. Do not melt the Independent article into the 52, or the 52 into a “gotcha” that the paper already half-admitted.

  2. Hubinger’s two-clause sentence is an ethics-washing specimen without needing to call him a fraud. “Trying its best” + “we do not yet have a plan to solve alignment for superintelligence and are not clearly on track.” That is declaration vs process vs correction sitting in one quote. #23 (living process → stored conclusion) and correction-persistence live there: the safety-lab identity can be stored while the plan is absent. Inference of motive stays Dicks’s; the sentence is public.

  3. “Don’t outsource judgment” is the 12-stage in civic language. Treat neither 10% nor 0.85% as gospel; read the article; watch cameras, IDs, filters, rails. That is represent-before-evaluate, not a third religion. The Correspondence-First protocol already says evaluation too early is the failure. A kill switch is C2 at state scale. Dicks trusts neither lab god-mode nor Westminster’s button. C1 before C2 still applies: a button is not clarification.

  4. Glass / cage. Dicks: if you control models → summaries → news → Overton, then wire that into finance, health, policing, education, you need not extinct the species — only unmanaged humanity. That is Prompt 49 and the glass-engine page at civic scale: audit of summaries, not worship of forecasts. “Floor plan vs fireworks” is process-signature: keep the rails, don’t only keep the extinction number.

  5. Headline-only Coxon. Reporting around the resignation said most of the huge audience never read past the first post. That is #11/#16 at a hundred million views: the resignation-as-slogan survives; the procedural objection (endgame launched from a private Slack) dies. The project already treats that as the usual compression, not as a new law of virality.

  6. Lennox’s third setting is a neighbor, not a merger. He rejects “worship the machine / fear the machine” for discern. ACCM’s third setting is correspondence + glass + C1, not the incarnation claim. Represent the fork; do not import God, AI, and the End of History as project theology, and do not nearest-generalize it into “religion vs science.”

  7. Bernays at machine speed (Prompt 50). Suggestions in your dialect, with your face, in real time. That can be intent resonance inverted into a compliance mechanism. Named as a question the 52 already asks. Not proven here.

Coincidence not used: Noah Predict’s “52 million data points” next to the project’s 52. Rhyme is not correspondence.

Why (John’s question)

Easy path: rank the messenger, skip the 10% / 0.85% / kill-switch object. Harder path: keep the messenger in Adult↔Adult standing, run the 27 on what was said, including where Dicks himself compresses, sponsors, or infers motive. That is the 27 used as they were written — portable, two-way, without immunity.

Related: 52 prompts · Ethics-washing · Glass / latent space · 27 source · 27 names · Asymmetric scrutiny · C1 before C2

Experiments · All pages