Research press release




人間集団に提示された特定の問題の最適解は、その集団のコンセンサスだと考えられることが多いが、その理由としては「集合知」を利用して解を導き出す方法である点が挙げられる。ところが、この方法によって必ずしも正しい答えが得られるとは限らない。多数決で得た答えや回答者が最も自信をもっている答えを採用すると、この問題のテーマに関する専門知識を有する少数派の人々の意見が覆い隠されてしまうからだ。例えば「ペンシルベニア州の州都はフィラデルフィアか」という問題を提示した場合に、一般投票で多数となった「はい」が不正解(正解はハリスバーグ)であることをDrazen Prelecの研究チームが明らかにした。



A strategy for determining the correct answer to a question posed to a group, when the most popular or most confident response fails under the same circumstances, is described this week in Nature.

The consensus of a group is often considered to be the best answer to a particular question because this approach takes advantage of the ‘wisdom of the crowd’. However, this method does not always produce the correct response because the most popular or confident solution can mask the minority opinion of a few individuals who may have specialized knowledge of the topic. For example, when a group was asked whether or not Philadelphia is the capital of Pennsylvania, Drazen Prelec and colleagues show that the popular vote endorses the incorrect answer (‘yes’ - the capital is actually Harrisburg).

The authors propose an alternative to this democratic vote that instead asks respondents to predict, in addition to casting their own vote, the distribution of others’ answers to the same question. By selecting the answer that is more popular than people predict (in this case ‘no’) the authors demonstrate that they can reliably identify the correct answer. They test this ‘surprisingly popular’ algorithm across a variety of scenarios in experiments involving the US state capitals and general knowledge, as well as by asking professional dermatologists to classify skin lesion images and asking art professionals and laypeople to estimate the value of various pieces of artwork.

In these studies, which involved different groups of 20-50 participants, Prelec and colleagues show that their algorithm reduced errors by approximately 21-35% relative to other common selection principles, such as choosing the simple majority or weighting votes by confidence. They note that this technique would be useful for questions about controversial topics, such as politics or the environment, where respondents might alter their vote predictions to align with their own views.

doi: 10.1038/nature21054

「Nature 関連誌注目のハイライト」は、ネイチャー広報部門が報道関係者向けに作成したリリースを翻訳したものです。より正確かつ詳細な情報が必要な場合には、必ず原著論文をご覧ください。

メールマガジンリストの「Nature 関連誌今週のハイライト」にチェックをいれていただきますと、毎週最新のNature 関連誌のハイライトを皆様にお届けいたします。