| paper | d4-conf-aaai-2012-0158 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | no |
The paper presents Algorithm 1 and experimental results, but no numbered theorem, lemma, corollary, or proposition asserting a computational result. The proposed worker-type optimization is an unanchored new research question, so no continuous mirror of a result in this paper survives.
fails bit a — no named computational result to mirror
There is no eligible named computational result to cover; Algorithm 1 and the experimental comparisons do not meet the computational-result requirement.
I can’t make a properly anchored positive case from this paper. It contains no numbered theorem, lemma, corollary, or proposition asserting a computational complexity or algorithmic result. Algorithm 1 describes adaptive polling, and the paper analyzes its model and reports experiments, but none is an eligible anchor. So there is no anchor here for a continuous problem-per-result case.
The closest plausible mirror would treat workers as a finite set of information-quality types, with \(\mu_t\) the fraction of the available crowd of type \(t\). Given the current estimates and a polling budget, choose a candidate pair to maximize expected information gain per cost across that worker distribution. This is a natural extension of the paper’s polling setup, especially since it notes that workers may differ in how informed they are. But the paper assumes workers are ex ante equally informed and makes no computational claim about this extension. It is therefore a possible research question, not a mirror anchored in a named result of this paper.
The strongest case against starts with the proponent’s concession: the paper has no named computational result to mirror. Algorithm 1 is a polling procedure, and the paper reports empirical comparisons; neither gives a complexity classification or algorithmic guarantee. Its likelihood and information-gain calculations do not supply one either. The proposed optimization over worker types would therefore be a new question, not a continuous version of a result in this paper.
The best version of that new question is plausible: suppose a large crowd consists of \(\tau\) known worker-quality types, with \(\mu_t\) the fraction of each type. One could choose which candidate pair to ask—and perhaps which type to recruit—to maximize expected posterior information under a budget. But the paper sends each question to a fresh worker and treats responses as random evidence. A type distribution can describe the response mixture, yet it does not make an individual answer divisible. Replacing many answers with fractional response counts would suppress the sampling uncertainty that the paper’s information-gain objective measures. Choosing types, imposing type-specific capacities or costs, and optimizing over those choices would add the central ingredients of a new active experimental-design problem; the paper studies none of them.
That extension could be worthwhile in a broader study of crowd-based experimental design. The negative case is weaker if stated as “no such model could ever be useful”: large crowds can plainly contain repeated worker types. Its stronger, narrower point is that this paper supplies no computational result for ChoCo to continuize, and the most credible repair changes the problem into adaptive statistical design. That makes it a poor anchor for the programme, even though it does not establish that every related continuous question is uninteresting.
The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.