Truth and Truthfulness: Experimental Tests of the Bayesian Truth Serum

真实性与真实性：贝叶斯真值血清的实验测试

• 批准号: 0519141
• 负责人: Drazen Prelec
• 金额: $ 26万
• 依托单位: Massachusetts Institute of Technology
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2005
• 资助国家: 美国
• 起止时间: 2005-09-15 至 2009-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0519141&HistoricalAwards=false
• 关键词: Truth; Truthfulness; Experimental; Tests; Bayesian

The proposal objective is to experimentally test and theoretically extend the "Bayesian Truth Serum", a new method for obtaining superior quality information from respondents in opinion research or expert judgment settings. The method rests on a mathematical scoring algorithm that gives respondents an incentive to provide truthful and carefully considered answers to questions, even though only the respondent knows which answer best matches to their truthful opinion. The method could be used to collect truthful expert forecasts of events in the far future (e.g., whether mankind will survive past year 2100), truthful expression of personal tastes or intentions (e.g., intentions to vote), and truthful self-reports of personal behaviors and characteristics (e.g., risky sexual practices).The main idea behind the scoring algorithm is to give high scores to answers that are more common than predicted, with predictions generated by the same respondents who answer the questions. It follows from Bayesian reasoning that truth-telling maximizes expected score even for a respondent who is sure that their truthful opinion represents a minority view. Furthermore, in situations where there is a single objectively correct answer to a question (i.e., a question dealing with objective knowledge), the scoring algorithm should be able to identify the objectively correct answer even if the majority of respondents endorse a different, incorrect answer.

该提案的目的是通过实验测试和理论扩展“贝叶斯真理血清”，一种新的方法，用于从民意研究或专家判断设置的受访者中获得上级质量信息。 该方法依赖于一个数学评分算法，使受访者提供真实和仔细考虑的问题的答案的激励，即使只有受访者知道哪个答案最符合他们的真实意见。 该方法可用于收集对遥远未来事件的真实专家预测（例如，人类是否能活过2100年），个人品味或意图的真实表达（例如，投票意向），以及个人行为和特征的真实自我报告（例如，评分算法背后的主要思想是给比预测更常见的答案打高分，预测由回答问题的同一个受访者产生。 根据贝叶斯推理，即使被调查者确信自己的真实观点代表少数人的观点，说实话也会使预期得分最大化。此外，在存在对问题的单个客观正确答案的情况下（即，处理客观知识的问题），评分算法应该能够识别客观正确的答案，即使大多数应答者赞同不同的、不正确的答案。