著者

犬童 健良

出版者

KANTO GAKUEN UNIVERSITY

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.45, pp.40-75, 2019 (Released:2019-03-31)

参考文献数

21

被引用文献数

1

---

リスク下の人間の意思決定をモデル化するために効用と確率に非線形バイアスをかけて加重和をとる数量的評価が使用されてきた.KahnemanとTverskyのプロスペクト理論では,心理学的実験で検証された参照点に依存する選好を表現できる価値関数と確率ウェイト関数が提案された.累積プロスペクト理論ではランク依存変換によってより複雑なクジを選ぶ場合に拡張された.またSchmidt, StarmerとSugdenの第3世代プロスペクト理論ではプロスペクトの価格評価が組み込まれた.一方,これらのモデルは理論的および実証的な観察に基づき批判されてもいる.人間のリスク行動を計算的なモデル化と理解に向けた準備的段階として,本研究はリスク下の人間行動の計算的モデル化と理解に向けた準備的段階として,Prologを用いて累積プロスペクト理論とシフトする参照点を記述するその微修正に対する論理プログラムを提供する.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.46, pp.31-121, 2020 (Released:2020-03-31)

参考文献数

79

---

Prologはコンピュータプログラムをホーン節として構築し,バックトラックメカニズムを備えた背理法アルゴリズムを適用する論理プログラミング言語である. この論説は,主に日本語で書かれた自然言語コンピュータープログラミングのためのPrologの質素で実用的なアプリケーションが示される. 2つの異なるアプローチが用いられた.第1のアプローチではPrologの述語をベースにした日本語論理プログラミンを提案する.日本語の単語で書かれた質問を助詞で区分することによって,システム/ユーザ定義述語やシェルコマンド呼び出しを本体に含むルールの頭部と単一化する.とくに助詞「は」は知識の参照と知識の主張に活用される.第2のアプローチでは,日本語で書かれた文書から節プログラムを直接抽出する.「相続税の計算」を説明したサンプル文書からの自動抽出がデモンストレーションされた.さらに,第1のアプローチは否定,三段論法,および意味論的概念を組み込んで修正された.既知/未知,旧い/新しい情報,総記と対比が,「XはY」や「XがY」の形の文に対するpyログラムに組み込まれた.総記の「が」と対比の「は」については,語のペアの代替的関係からなるネットワークのゲーム理論的モデルにおける安定性条件を用いて説明された.さらに,総記や対比の認知機能とルール節の導出との類似が論じられ,また非単調推論におけるサーカムスクリプションと関連付けられた.関連分野として,認知的-社会的な文脈におけるエージェンシー関係が論じられた.

著者

犬童 健良

出版者

日本認知科学会

雑誌

認知科学 (ISSN:13417924)

巻号頁・発行日

vol.23, no.2, pp.153-169, 2016-06-01 (Released:2016-12-01)

参考文献数

33

---

This study presents a model of human reasoning as a collection of micro-level in-ternal agents, and applies it especially to two problems: the four-card selection task (selecting cards to verify a conditional sentence) and the problem of three prisoners (a problem to estimate probability when an outcome proves impossible). In both the problems real people tend to report an intuitive answer which differs from one that is considered mathematically correct. In the presented approach, game theory is used to explain human reasoning, and connected to the frame systems and the society of mind proposed by Marvin Minsky. As it is assumed in a standard non-cooperative game the-ory, the internal agents do not have a language to communicate with other than game play. Also, it is assumed that the intelligence of agents is restricted only in terms of improving one’s expected payoff. At any step of reasoning, each agent microscopically responds (i.e., collecting labels) to a given collective action of other agents. The infor-mation response model, the best responses in standard game theory, is parametrized by the payoffs of a bimatrix game, so as to predict the observed patterns typically cho-sen/answered by real people at an equilibrium point of the game. Human reasoning is modeled as a path from the first (i.e., default) equilibrium point to the final equilibrium point. Moves between the two equilibrium points are represented using Lloyd Shap-ley’s labeling system and interpreted as Minsky’s frame system (or K-lines). It can be concluded that the proposed modeling can summarize and improve previous research on both the four-card selection task and the problem of three prisoners, which seems to have been inaccurately modeled in the literature.

著者

犬童 健良

出版者

一般社団法人 経営情報学会

雑誌

経営情報学会 全国研究発表大会要旨集 2003年度春季全国研究発表大会

巻号頁・発行日

pp.11, 2003 (Released:2003-08-01)

---

論理プログラミング言語Prologは知識処理の代表的な手段として知られる。Prologは他の手続き的言語や表計算では厄介な離散的領域でのモデリングに適している。また少々の数学的ユーティリティを作成することにより、動学的最適化を含む意思決定技法や情報システムのモデリングに幅広く適用できる。筆者は、決定木分析、最適探索、線形代数、協力ゲーム、信念関数、最短路アルゴリズム、関係代数などのProlog化を試み、実際にうまくいくことを確かめた。

著者

犬童 健良

出版者

一般社団法人 人工知能学会

雑誌

人工知能学会全国大会論文集 第17回全国大会(2003)

巻号頁・発行日

pp.248, 2003 (Released:2004-02-03)

---

Prologはゲーム理論の各種モデルと解概念、およびゲームプレイヤーの知識推論について比較的素直に表現できるプログラミング言語である。またゲーム理論や社会選択理論を操作的にモデル化することにより、限界合理性の下でのふるまいを理解するのにも役立つ。本論文では筆者が行ったモデリング事例をいくつか紹介する。

著者

犬童 健良

出版者

関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:13451588)

巻号頁・発行日

vol.41, pp.32-63, 2016-03-31

---

J. Edmonds and D. R. Fulkerson defined a blocking system as a pair of subset families, i.e., a clutter and a blocker, in which the bottleneck theorem holds true (Journal of Combinatorial Theory, 1990, 8(3), 299-306). Blocking systems have been studied in several combinatorial problems, such as those including graph, matroid, network flow, optimal assignment, and multi-person games. In this study, blocking systems are applied to cognitive modeling of a single decision maker. The domain of the blocking system corresponds to aspects in the multi-dimensional comparison of alternatives, and the models are discussed including choice under risk, consumer brand choice, and reasoning task.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.44, pp.19-43, 2018 (Released:2018-03-30)

---

Prospect Theory (PT), as proposed by Kahneman and Tversky (Econometrica, 47(2), 263-292, 1979), is employed to study reference-point dependent preferences in order to describe real human behavior in decisions under risk, where typically observed patterns of choice over given lotteries violate the assumption of maximizing expected utility, for example in Allais’ paradox. This theory assumes a probability weighting function (typically inverse S-shaped) and a value function (typically concave above and convex below the reference point) for at most two possible outcomes. It also assumes a single reference point as the monetary value for each lottery. Cumulative Prospect Theory (CPT) (Tversky and Kahneman, Journal of Risk and Uncertainty, 5(4), 297-323, 1992) and Third Generation Prospect Theory (PT3) (Schmidt, Starmer, and Sugden, Journal of Risk and Uncertainty, 36(3), 203-223, 2008) have been proposed as advanced versions of Prospect Theory. Both advanced theories generalize PT for finite outcomes by using rank-dependent transformation (i.e., the Choquet integral). However, PT and CPT do not explicitly explain how to specify the reference point for each choice problem of a decision maker and the descriptive power of CPT is severely restricted due to its satisfying stochastic dominance. PT3 allows stochastic reference points in order to predict the phenomenon of preference reversals, which the original PT and CPT cannot predict. In PT3, the problem of reference point selection is partially solved by assuming the lottery that is not chosen forms the reference point for each state satisfying a state-dependent version of stochastic dominance. However, this paper presents a computer simulation of CPT, which can predict both types of problems: a common ratio effect (CRE) and common consequence effect (CCE), which may have very different parameters from those empirically proposed by Tversky and Kahneman. It also shows that the prediction of CPT (and PT3) never qualitatively improves in the specified problem considered in this paper, even if shifts in the reference point and its stochastic indetermination are explicitly incorporated.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.47, pp.1-29, 2021 (Released:2021-03-31)

参考文献数

65

---

The probability weighting function is a quantitative model that captures people's responses to risk as a non-linear bias to probability. Various fields such as psychology, economics, and management science apply this function. However, the one-parameter functional form in Tversky and Kahneman’s cumulative prospect theory (CPT) is not capable of predicting the common consequence elimination case originally presented by Allais. Additionally, it cannot model the reference dependency in this problem. Therefore, we proposed a series of computational experiments of a modified version of CPT with a beta distribution; that is, an incomplete beta ratio and other functional forms, including generalized hyperbolic discounting, instead of the original probability weighting function used in CPT. The results reveal that the modified CPT with a beta distribution can reproduce a typical choice pattern in the Allais paradox and reference dependency under more natural inverse S-shaped curvature.

著者

犬童 健良

出版者

一般社団法人 経営情報学会

雑誌

経営情報学会 全国研究発表大会要旨集 2007年度春季全国研究発表大会

巻号頁・発行日

pp.55, 2007 (Released:2007-10-17)

---

非対称的優越性,妥協効果,および類似効果は,意思決定の規範的モデルに対するアノマリーとして知られる.本論文はこうした2属性3代替案の選択問題およびギャンブル選択における選好逆転現象における文脈依存性を説明するための認知的モデリングを提案した.これらに関して近年ニューラルネットを用いた計算的モデリング(DFT)が提案されたが,本論文は,対立する評価を統合する選好集計とみなし,とくに多数決ルールを用いたモデリングを試みている.また選好の転送可能性を解釈する認知的権利システムの概念を導入した.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.39, pp.53-80, 2014-03-31 (Released:2017-04-03)

---

Maurice Allais (Econometrica, 21, 503-546, 1953) proposed a choice problem designed to question comparison of paired gambles, in order to show that the Expected Utility Theory cannot predict choices by real people. Kahneman and Tversky (Econometrica, 47, 263-291, 1979) proposed Prospect Theory with a solution for this puzzle by using a probability weighting function. However, the shape of the weighting function of the decision-maker may differ between choice problems. Regret Theory, proposed by Bell (Operations Research, 30, 961-981, 1982) and Loomes and Sugden (The Economic Journal, 92, 805-824, 1982), used the notion of (anticipated) regret and rejoice to shed light on the role of cognitive attention on the difference in possible outcomes by not only the actual choice but also the counterfactual choice. In this paper, the Allais paradox is extended slightly, by adding a question on the evaluation of attention strength for each possible joint outcome (called a "cell" in this paper) on a pair of gambles, for each gamble to be chosen. The experimental results showed a persistent tendency for greater attention to regret than rejoice outcomes. Counterfactual outcomes draw more attention than the outcome of the gamble actually chosen. Further, regret cells, that is, joint outcomes of negatively valued differences, draw more attention from the participants who choose riskier options than participants who choose safer options, on average. In addition, by analyzing the monotonicity of the function from attention to choice, a consistent pattern that can explain the experimental results is derived. The monotonicity is argued to mean that the preference is stable for self-conscious decision makers. For the analysis of monotonicity, extraction of logical implications in attentional rank comparison, and so on, Prolog programs have been developed and used.

著者

犬童 健良

出版者

行動経済学会

雑誌

行動経済学 (ISSN:21853568)

巻号頁・発行日

vol.9, pp.81-84, 2017 (Released:2017-06-01)

参考文献数

15

---

本論文は,アレの背理における共通比効果(あるいは確実性効果)の要因を分析するため,クジ間の比較を交差的注目ネットワークモデルとしてモデル化し,先行研究(犬童,2013)の交差的注目の実証データを再解釈した.交差的注目とは,仮想的に選択されたクジの一つの可能な結果を条件として,選択されなかったクジの別の結果に注目することである.草的注目ネットワークでは,一対のクジ間の比較における認知的活動を,一方のクジの可能な結果を仮想的な参照点として,もう一方のクジの可能な結果がどれだけ気になるかの指標としてポテンシャル値が各枝に割当てられる.またこのネットワークはポテンシャルゲームとして解釈される.本論文では経路の追加が混雑を悪化させる交通現象のゲーム理論モデルであるブライスの背理が,アレの背理を引き起こす交差的注目ネットワークにおいて起きることが確認され,また参照点の設定との関係が論じられた.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.49, pp.1-42, 2023 (Released:2023-03-31)

参考文献数

68

---

In organizations, or societies, individuals are specialized as agents by embedding their decision-making rules in the structure of organizational design, and their activities are coordinated under the authority or other communication devices. The range of decisions and responsibilities assigned to agents is specified, and at the same time, is consciously or unconsciously limited. This study proposes a quasi-computational model of cognitive processes and knowledge in organizations in which certain pathological phenomenon of cognitive blocking may appear. Agents embedded in organizational context suffer from internal conflict between their wants and requirements, and sometimes ignore signals that imply the ethical issues or injustices. Combined with the elimination by tolerance, and linking to inverse resolution, the phenomenon of cognitive blocking is modeled as a version of the centipede game, an agent extension of Prolog. Tolerance indirectly controls the agent’s reasoning; the notion of “a feeling of responsibility for the effect of one’s actions on others” is modeled and is expected to be the basis for AI programs with self-concept, acceptance of authority, responsibility, and reliance.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.48, pp.28-62, 2022 (Released:2022-03-31)

参考文献数

85

---

Information avoidance is a dynamic feature of a decision maker’s information response in which certain message patterns are kept blocked through a feedback loop, as illustrated in the example of the Heikegani crab. On the other hand, its symptomatic potential is maintained, and under certain conditions, the block is lifted, thereby facilitating a positive information response. While focusing on information avoidance, this paper attempts to review the potential relationship between economic science in a broad sense, i.e., the rationality of decision makers and its epistemological foundations, as well as logic programming.

著者

犬童健良

雑誌

情報処理学会研究報告知能と複雑系(ICS)

巻号頁・発行日

vol.1996, no.105(1996-ICS-106), pp.37-44, 1996-10-30

---

本論文では,意志的主体は脳の代理人であるという仮説の下で,ゲーム理論にもとづくナッシュ遂行理論を用いたエージェントのメンタルモデルを提案した.これは知能の計算メタファーと限界合理性概念という,認知科学と意思決定科学それぞれにおける説明の難点を補うものである.

著者

犬童健良

雑誌

情報処理学会研究報告知能と複雑系(ICS)

巻号頁・発行日

vol.1994, no.67(1994-ICS-095), pp.87-88, 1994-07-21

---

著者の研究スタンスは、自己意識の認知的モデルを探究するにあたって、市場理論や非協力ゲーム理論(あるいはメカニズムデザイン論)とのアナロジーから出発し、心の社会における調整(ordinati)として捉えることであった[4,5].しかしこれら合理性に基づく協調のための一連の語り口は、じつは分権的調整を支える論理的な正当化の信念(stifying beli)にはなりえないことが最近のゲーム理論の文献[2]で論じられている.ここで協働(operati)は、むしろ調整のしくみを通じて各個人の能力の限界が克服されるときに興味ある現象であったことを思い出さずにはいられない.じっさい協働はC.?.Barnardから、後にH.A.Simonがその定式化を受け継いだ、限定された合理性(unded rationali)の概念や満足化原理が論じられたオリジナルの文脈であった.しかし、そもそも合理的でない人がどうやって自分に許された合理性のうち、協働のために放棄すべきかを定められるというのだろうか.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.40, pp.25-103, 2015 (Released:2017-04-03)

参考文献数

27

被引用文献数

1

---

This study proposes a risky decision model, which models a single decision maker as a collection of microlevel agents. In this model, each agent focuses on gambles comprising two branches on their basic aspects, i.e., possible monetary outcomes and probabilities. In addition, it considers each agent as a simple condition of difference at a branch pair of the two gambles in comparison for each basic aspect, and a tuple of such conditions for both gambles forms a coalition of agents that may be effective (promotional) or preventive (inhibitive) to choose. This model is a special case of noncompensatory multiattribute decision making and is described through voting functions, which was first introduced by K. Jacobs in Socio-Combinatorics (in O. Moeschlin and D. Pallaschke, eds., Game Theory and Related Topics, 1979, 309–335). We reanalyze the experimental data in the literature, especially that in E. A. Mellers et al. (Organizational Behavior and Human Decision Processes, 1990, 52, 331–369) to reconstitute the aspects and voting function of real people. This analysis revealed three main facts: First, a merit in probability is effective in changing a choice, but it makes the reasons for the choice unreliable. Second, a nonlinearity occurs if the voting functions for gain and loss are merged. Third, the priority heuristics proposed by E. Brandsttäter et al. (Psychological Review, 2006, 113, 409–432) can be considered as a search procedure for an effective aspect in agreement with the voting function. The appendices present the Prolog programs of voting functions and computational experiments.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.41, pp.32-63, 2016 (Released:2017-04-03)

参考文献数

37

---

J. Edmonds and D. R. Fulkerson defined a blocking system as a pair of subset families, i.e., a clutter and a blocker, in which the bottleneck theorem holds true (Journal of Combinatorial Theory, 1990, 8(3), 299-306). Blocking systems have been studied in several combinatorial problems, such as those including graph, matroid, network flow, optimal assignment, and multi-person games. In this study, blocking systems are applied to cognitive modeling of a single decision maker. The domain of the blocking system corresponds to aspects in the multi-dimensional comparison of alternatives, and the models are discussed including choice under risk, consumer brand choice, and reasoning task.

著者

犬童 健良

出版者

一般社団法人 経営情報学会

雑誌

経営情報学会 全国研究発表大会要旨集 2006年度秋季全国研究発表大会

巻号頁・発行日

pp.3, 2006 (Released:2007-08-10)

---

本論文は選好集計理論を限界合理性に対して応用することによって,(誤)表象をもつエージェントの認知的モデリングを提案した.選好集計理論は,社会的選択の分析に用いられたが,本論文では前診断的プロセス,つまり日常的推論やオフィスにおける断片的な情報の統合に適用する.また論争ジレンマを条件文推論の検証(選択課題)のための実用的スキーマとして再解釈することを通じて,権利システムとしての側面に光を当てる.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.42, pp.1, 2017 (Released:2017-03-31)

参考文献数

12

---

This paper studied cognitive modeling based on the cross attentional network for gamble comparison tasks developed by Indo (2013, 2014a, 2016b) by reanalyzing the empirical data and connecting it to the game theoretical model first studied in transportation network optimization. Cross attention potentials are rank values of reported strength concerning (i.e., markedness) each ordered pair of possible outcomes of two lotteries. By reanalyzing the experimental data of cross attention potential in the Allais paradox, this paper proves a relationship between shift of risk attitude to the orientation (i.e., spin) of cross attention network. This network model of cognitive cross attention can be translated into potential games, especially a congestion game, and then transformed into the Braess paradox, which predicts suboptimal flow caused by selfish behavior.

著者

犬童 健良

出版者

学校法人 関東学園大学

雑誌

関東学園大学経済学紀要 (ISSN:21878498)

巻号頁・発行日

vol.43, pp.12-33, 2017 (Released:2017-08-31)

参考文献数

22

---

Pythagorean expectation is a simple yet experimentally robust method that utilizes a single parameter for predicting the winning percentages of sports teams based on the score ratio. The score ratio can be defined as runs scored divided by runs allowed, and the Pythagorean probability can be identified as a parameter of logistic distribution. In this study, we aim to estimate the winning probability by incorporating actual sports data and to interpret the Pythagorean probability by using the generalized matching law of behavioral economics. In addition, the probabilistic models for runs scored and runs allowed, which can be considered as two independent Gumbel distributions, are estimated with maximum likelihood parameters from the actual data of a student baseball league. We compare four types of models: the Gumbel model, Poisson model, Pythagorean model, and histogram-based resampling model. For the Gumbel model, three types of methods are applied, namely maximum likelihood (ML), ordinary least squares (OLS), and EAP (expected a posteriori). Further, a series of Monte Carlo simulations shows that the histogram is best, and the Pythagorean is quite good; however, the Gumbel models slightly outweigh the Pythagorean model in terms of the mean square prediction error.