Skip to content
#explainable ai Open access

The Card People Pick Second Most Often Carries Exactly Zero Information ── Taking the rule "if p then q" with a prior of 0.5, the two cards that can falsify it (p and not-q) carry 0.311278 bits each, while the other two (q and not-p) carry exactly 0.000000000 ── the separator is whether a card can falsify the rule ── [Paper 699]

Sep 2026 · Zenodo (CERN European Organization for Nuclear Research)
Quantum Mechanics and Applications

Abstract

In the Wason selection task the second card people turn is the consequent of the rule itself. Whatever is on its back, it cannot contradict the rule. What each chosen card carries is counted here in bits. No new theorem or law is claimed. Scope of this paper (scope note): No new theorem or law is claimed──the Wason selection task, the low rate of correct answers and matching bias are all standard (Wason 1968, Evans and Over 2004). The experiment is not reproduced──the figure of 0.100000 is taken as a representative reported value; actual figures move widely with the task and the population. Content effects are not derived──that framing the rule as a social contract raises the rate is mentioned and not explained. No psychological mechanism is offered──why people match the words of the rule is not discussed. The assumptions are not claimed──the prior of 0.5, and the probability 0.5 that a p card shows q when the rule is false, are assumed; other values change 0.311278. The two zeros do not move. Relation to earlier papers: Paper 281 showed that a quantum state cannot be copied yet can be corrected, the syndrome carrying only the location of the error──there what to measure in order to obtain information was counted from the side of design. Here the same unit counts an observation of zero information being chosen. The same bits, working in opposite directions. What is added is computing the information of all four cards and showing two to be exactly zero, showing that zero not to depend on the assumptions, putting the ratio against chance at 0.600000, showing the commonly chosen pair to reach only 0.500000 of the correct pair, and placing the separator on whether a card can falsify. First, and this is the core. Two of the four carry exactly zero──the q card and the not-p card carry 0.000000000 bits, and only the two that can falsify carry 0.311278 (Section 2). Second, the zero is not an artefact of the prior──whatever prior is assumed, a card that cannot falsify still cannot, and its information stays at zero (Section 2). Third, people do worse than chance──picking two at random gives 0.166667, against a reported figure of about 0.100000, a ratio of 0.600000 (Section 3). Fourth, doing worse than chance is itself the evidence of a bias──random choice would give 0.166667; falling below it requires a leaning towards one particular wrong pair (Section 3). Fifth, that pair leaves exactly half the information on the table──it carries 0.311278 bits against the 0.622556 of the correct pair, a fraction of 0.500000 (Section 3). Sixth, the separator is whether a card can falsify the rule──not whether it is named in the rule; people substitute the second for the first (Section 4). The card people pick second most often carries exactly zero information. Of the four, q and not-p carry 0.000000000 bits, and only the two that can falsify, p and not-q, carry 0.311278──that zero is not an approximation and does not move with the prior. Picking two at random would be right 0.166667 of the time against a reported 0.100000, a ratio of 0.600000──doing worse than chance is itself evidence of a leaning towards one particular wrong pair. That pair carries 0.311278 bits, exactly 0.500000 of the 0.622556 carried by the correct pair──what is chosen are the cards showing words that appear in the rule, not the cards that can falsify it. The separator is whether a card can falsify the rule. Placed among the earlier papers──Paper 281 counted what to measure in order to obtain information; here the same unit counts an observation of zero information being chosen. To be honest──experiment, content effects and mechanism are not treated, and both the rate and the assumptions are representative. On the making of this work: The ideas and content of this work stem from the author's own considerations. Assistance from an AI (a large language model) was used for structuring, English translation, and checking the algebra. Any remaining errors or misinterpretations are solely the author's. Feedback and corrections are sincerely appreciated. Keywords: Wason selection task, falsification, information in bits, confirmation bias, matching bias, observation of zero value. ----- ウェイソンの選択課題で人が選ぶ二枚目の札は、規則の後件そのものである。だがその札は、何が出ても規則を否定できない。選んだ札が運ぶ情報を、ビットで数える。新しい定理も法則も主張しない。 本稿の射程(射程注記):新しい定理も法則も主張しない──ウェイソンの選択課題、正答率の低さ、照合の偏り(規則に出てくる語の札を選ぶ傾向)は、いずれも既知である(ウェイソン 1968、エヴァンス&オーヴァー 2004)。実験を再現しない──正答率 0.100000 は報告される代表値として置く。実際の値は課題と集団で大きく動く。内容効果を導出しない──社会契約の形にすると正答率が上がることは触れるにとどめ、その理由は論じない。心理機構を論じない──なぜ照合してしまうのかの説明は述べない。置き値を主張しない──事前確率 0.5 と、規則が偽のときに p の裏が q である確率 0.5 は前提であり、値が変われば 0.311278 も変わる。ただし零の二枚は値によらず零である。既刊との関係:論文281 は、量子の状態は複製できないのに訂正はでき、症候群が誤りの場所だけを運ぶことを示した──そこでは「何を測れば情報が取れるか」を設計の側から数えた。ここでは「情報が零の観測を人が選んでしまう」ことを、同じ単位で数える。設計ではなく選択の側である。加えたのは、四枚の札の情報量を出して二枚がちょうど零であることを示したこと、零が置き値によらないことを示したこと、正答率と偶然の比を 0.600000 と数にしたこと、よく選ばれる組が正解の 0.500000 しか取れないことを示したこと、分離子を「反証しうるか」に置いたことである。 第一に、これが本稿の芯である。四枚のうち二枚は、情報をちょうど零しか運ばない──q の札も非 p の札も 0.000000000 ビットで、反証しうる二枚だけが 0.311278 ビットを運ぶ(第2節)。 第二に、零になるのは、置き値のせいではない──事前確率をどう置いても、反証しえない札は反証しえないままで、情報は零のままである(第2節)。 第三に、人の正答率は、偶然より低い──無作為に二枚選べば 0.166667 のところ、報告される代表値は 0.100000 で、比は 0.600000 である(第3節)。 第四に、偶然より低いことは、偏りの証拠である──偶然に選んでいるなら 0.166667 になるはずで、それを下回るには、正解でない特定の組を選ぶ傾向が要る(第3節)。 第五に、その組が取りこぼす情報は、ちょうど半分である──よく選ばれる組は 0.311278 ビットで、正解の組の 0.622556 ビットの 0.500000 にとどまる(第3節)。 第六に、分離子は、その札が反証しうるかどうかである──規則に出てくるかどうかではない。人はその二つを取り違えている(第4節)。 人が二番目に多く選ぶ札は、情報がちょうど零である。四枚のうち q と非 p は 0.000000000 ビットしか運ばず、反証しうる p と非 q だけが 0.311278 ビットを運ぶ──零は近似ではなく、事前確率をどう置いても零のままである。無作為に二枚選べば正答率は 0.166667 のところ、報告される代表値は 0.100000、比は 0.600000 である──偶然より低いことは、正解でない特定の組へ寄る傾向の証拠である。よく選ばれる組は 0.311278 ビットで、正解の組の 0.622556 ビットのちょうど 0.500000 しか取れていない──選ばれているのは規則の文に出てくる語の札で、反証しうる札ではない。分離子は、その札が反証しうるかどうかである。既刊との位置──論文281 は「何を測れば情報が取れるか」を設計の側から数えた。ここでは情報が零の観測を選んでしまう側を数えた。正直に言えば──実験も内容効果も心理機構も扱わず、正答率も置き値も代表値である。 作成にあたって:本稿の着想と内容は、著者自身の考察に基づくものです。文章の構成整理や英訳、数式の確認には AI(大規模言語モデル)の助力を得ました。最終的な内容の解釈や誤りがあれば、それらはすべて著者の責に帰します。お気づきの点があれば、ご教示いただければ幸いです。 キーワード:ウェイソン選択課題、反証、情報量、確証バイアス、照合の偏り、零ビットの観測。

View source

Similar papers

#artificial intelligence Conference Open access Apr 2020

ECCOLA - a Method for Implementing Ethically Aligned AI Systems

The method, ECCOLA, is presented, which aims at making the high-level AI ethics principles more practical, making it possible for developers to more easily implement them in practice.

Ville Vakkuri, Kai-Kristian Kemell, P. Abrahamsson · 64 citations · ⚡6
#computer vision Review Apr 2024

AI-powered Code Review with LLMs: Early Results

The goal is to not only refine the accuracy of the LLM-based tool but also to underscore its potential in streamlining the software development lifecycle through proactive code improvement and education.

Z. Rasheed, Malik Abdul Sami, Muhammad Waseem et al. · 62 citations · ⚡3
#computer vision Open access Mar 2024

LLM-based agents for automating the enhancement of user story quality: An early report

The use of large language models to automatically improve the user story quality in Austrian Post Group IT agile teams is explored, with a reference model for an Autonomous LLM-based Agent System developed and implemented at the company.

Zheying Zhang, M. Rayhan, Tomas Herda et al. · 48 citations · ⚡4
#computer vision Review Mar 2024

System for systematic literature review using multiple AI agents: Concept and an empirical evaluation

This paper introduces a novel multi-AI-agent system designed to fully automate SLRs, and demonstrates how it substantially reduces the time and effort traditionally required for SLRs while maintaining comprehensiveness and precision.

Abdul Malik Sami, Z. Rasheed, Kai-Kristian Kemell et al. · 44 citations · ⚡2
#computer vision Feb 2024

Can Large Language Models Serve as Data Analysts? A Multi-Agent Assisted Approach for Qualitative Data Analysis

The proposed LLM-based multi-agent system automates qualitative data analysis process, creating opportunities for researchers and practitioners, and future improvements focus on enhancing multilingual performance and integrating continuous expert feedback.

Z. Rasheed, Muhammad Waseem, Aakash Ahmad et al. · 41 citations
#artificial intelligence Conference Open access Jun 2018

The Key Concepts of Ethics of Artificial Intelligence

It is suggested that the focus on finding keywords is the first step in guiding and providing direction for future research in the AI ethics field.

Ville Vakkuri, P. Abrahamsson · 39 citations · ⚡2

Related blog posts

Microsoft Research Blog Oct 7, 2026

Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses

Training AI agents with reinforcement learning can be challenging because their tools, context, and decision-making are managed by complex frameworks. Agent Lightning connects existing agents to RL training, making it easier to improve them without rebuilding them. The post Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses appeared first on Microsoft Research.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.