Strategic Decision s Survey Taxonomy a nd Future Direction s from Artificial Intelligence Perspective Foundation of A Representation Space for Machine Learning

2025-05-02 0 0 1.2MB 32 页 10玖币

侵权投诉

Strategic Decisions: Survey, Taxonomy, and Future Directions from Artificial

Intelligence Perspective

Foundation of A Representation Space for Machine Learning

Caesar Wu*

University of Melbourne caesar.wu@computer.org

Kotagiri Ramamohanarao

Institution of Engineers Australia rkotagiri@gmail.com

Rui Zhang

Tsinghua University rui.zhang@ieee.org

Pascal Bouvry

University of Luxembourg pascal.bouvry@uni.lu

Strategic Decision-Making is always challenging because it is inherently uncertain, ambiguous, risky, and complex. By

contrast to tactical and operational decisions, strategic decisions are decisive, pivotal, and often irreversible, which may result

in long-term and significant consequences. A strategic decision-making process usually involves many aspects of inquiry,

including sensory perception, deliberative thinking, inquiry-based analysis, meta-learning, and constant interaction with the

external world. Many unknowns, unpredictabilities, and organic/environmental constraints will shape every aspect of a

strategic decision. Traditionally, this task often relies on intuition, reflective thinking, visionary insights, approximate estimates,

and practical wisdom. With recent advances in artificial intelligence/machine learning (AI/ML) technologies, we can leverage

AI/ML to support strategic decision-making. However, there is still a substantial gap from an AI perspective due to inadequate

models, despite the tremendous progress made. We argue that creating a comprehensive taxonomy of decision frames as a

representation space is essential for AI because it could offer surprising insights that may be beyond anyone’s imaginary

boundary today. Strategic decision-making is the art of possibility. This study develops a systematic taxonomy of decision-

making frames that consists of 6 bases, 18 categorical, and 54 elementary frames. We use the inquiry method with Bloom’s

taxonomy approach to formulating the model. We aim to lay out the computational foundation that is possible to capture a

comprehensive landscape view of a strategic problem. Compared with many traditional models, this novel taxonomy covers

irrational, non-rational and rational frames capable of dealing with certainty, uncertainty, complexity, ambiguity, chaos, and

ignorance.

*This work is partially supported by The the Luxembourg National Research Fund (FNR) under grant C21/IS/16221483/CBD

Former Professor of the University of Melbourne

Visting Professor of Tsinghua University and former professor of the University of Melbourne

2 
CCS CONCEPTS • Insert your first CCS term here • Insert your second CCS term here • Insert your third CCS 
term here 
Additional Keywords and Phrases: Artificial Intelligence, Strategic Decision-Making, Decision Frames, Machine 
learning, Computation.  
ACM Reference Format: 
Caesar, Wu, Kotagiri Ramamohanarao, Rui Zhang, Pascal, Bouvry, 2021. Strategic Decision Frames: Taxonomy, Survey, 
Challenges,  and  Future  Direction  from  An  Artificial  Intelligence  Perspective,  this  document  explains  and  embodies  the 
submission format for authors using the word. 
1  INTRODUCTION 
Our life is full of choices. Our past decisions make who we are, and our current judgment will make whom 
we will become. This study primarily focuses on Strategic Decision Making (SDM) for an organization through 
careful assessments and deliberation. We pay special attention to Strategic Decision (SD) frames. A good SD 
will make an organization flourish. A bad one will lead an organization to catastrophe. Many SD makers have 
sought to understand how and why SDs are made in what context [1] [2]. Moreover, they also want to automate 
Decision-Making (DM) processes by taking advantage of new technologies, such as AI/ML [3] [4] . Historically, 
the  SDM  process  usually  requires  tremendous  resources  and  time.  However,  the  results  still  seem  to  be 
arbitrary. The final resolution often depends on the intuition and experiences of select individuals. 
With the current advance in AI/ML and other computational technologies, the analytic process of SDM has 
become much more powerful, foreseeable, reconfigurable, trustworthy, transparent, flexible, scalable, and cost-
effective.  Although  many  scholars  have  made  significant  progress  regarding  framing  and  knowledge 
representation [5] of ML [6] [7] and AI [8] for complex problem solving [9] [10] [11] [12] in practices [13] [14] [15], 
there is still a large gap in decision framing and modelling for SDM, which is “a series of associated knowledge 
representations or logic statements stored in our memory.”[16] Most previous studies often focused on rational 
reasoning for a particular application [17] [18]. However, rationality alone would not be able to solve all our 
problems, especially for an SD. We often make SDs based on our values, personal beliefs, and psychological 
emotions or passion. Clausewitz [19] summarized these elements (passion, probabilities, and reasons) in “the 
paradoxical trinity”. Simon [20] defined it as “bounded rationality”. Damasio [21] and Rolls [22] illustrated that 
many decisions primarily depend on our emotions rather than logic or reasoning alone from a neuroscience 
perspective. Minsky [23] argued that emotions are different ways of thinking. Therefore, this paper intends to 
conduct a comprehensive survey and articulate a taxonomy of DFs, including rational (Knowledge), irrational 
(Emotions), and non-rational (Data) decision frames (DFs) [24]. 
1.1  AI/ML In DM Process 
The study of DFs for SDM has a fundamental challenge. According to Schoemaker [25], the nature of our 
contemporary business environment is shifting from  certainty to chaos (See Figure 1). Strategic prediction 
becomes increasingly complex with growing uncertainty, ambiguity, and chaos. Niels Bohr stated, “It is difficult 
to make predictions, especially about the future.”. Schoemaker argued that traditional tools are not enough to 
manage states of chaos because the world has become much more complex. There are many possibilities and 
variances  in  the  wide  knowledge  spectrum,  which  demand  generating  multiple  DFs,  reviewing  deep 

assumptions, and exploring different unknown territories. Schoemaker offered a set of novel solutions in

contrast to the traditional toolkit and emphasized “Systems dynamic modelling” to increase our capability to

explore, exploit, and test our multiple hypotheses with different DFs. The ultimate goal of developing various

DFs from the environment to organic (left to right) and abstraction to reductionism (top to bottom) is to enhance

the learning capability and to cope with the shift from certainty to chaos.

Figure 1: A Wide Knowledge Spectrum Underpinned by DFs with Various Decision Tools (More Details of DFs in Section 4)

A lexical definition of learning capability is a person’s ability to comprehend and understand the world and

profit from one’s experiences by taking multiple DFs, which is an essential part of human intelligence. However,

we may become overwhelmed when we face many DFs. Each frame could also have numerous assumptions

that are continuously updated due to interactions with the external world. By leveraging AI/ML, we can tell a

machine what we want (output or the final strategic goal) rather than what to do (rules) because we often do

not know what the optimal solution (decision rules) could reach the final goal. We let the machine find the

optimal solution for us. We consider the capability of AI/ML as part of our intelligent faculty for an SDM process.

Figure 2 illustrates how to use Decision Framing and AI/ML processes for SDM. In other words, we create

multiple frames with different hypotheses to feed AI/ML. We let the machine find a set of the optimal decision

rules with a given or desired SDM and dataset rather than given decision rules and datasets for an SD. The

logical process is reversed when compared to Good Old-Fashioned Artificial Intelligence (GOFAI).

The essential proposition of adopting AI/ML is that machines will help us review a valuation landscape in the

representation space (or model), which is often very challenging to define explicitly. The subsequent issue is

how to determine a representation space that is broad enough to include all possible decision rules. On the

other hand, it may also be desirable to set the representation space small and precise enough for a machine to

learn with less time and resources. There are numerous ways to determine this representation space. With all

these goals in mind, we survey comprehensively and propose a two-dimensional decision taxonomy to enable

the machine to search for possible rules on our behalf.

Certainty Risks Uncertainty Ambiguity Chaos /

Ignorance

Higher Risk with Higher Reward

Lower Risk with Lower Reward

Traditional Tools____

•Expected Utility Theory

•Monte Carlo simulation

•Portfolio Optimization

•Insurance Programs

•Extrapolative Forecasting

•Net Present Value

•Decision Trees

•Bayesian Statistics

Novel Tools

•Influence Diagram

•Scenario Planning

•Real Options Theory

•Developing Intuition

•Agent-based Modelling

•Dynamic Systems Theory

•System Thinking

•Learning Culture

Reductionism

Manageable Unmanageable

Abstraction

1 Base 1 Base 1 Base

3 Categorical 3 Categorical 3 Categorical

9 Elementary 9 Elementary 9 Elementary

Environment Organic

A Wide Knowledge Spectrum

1.2 Primary Contributions

By exploring and exploiting various DFs in a two-dimensional space, this paper makes the following

contributions from an AI/ML perspective:

1. We provide a comprehensive survey of decision frames for various domains, including decision framing

bias, corporation planning, AI and robotics.

2. The study presents a novel taxonomy of SDM frames that consist of a total of 54 DFs. These decision

frames are derived from rational, irrational and non-rational domains.

3. The taxonomy lays the groundwork for us to deploy five different ML algorithms' tributes (based on their

origins) [26] in the learning space or model.

4. The uniqueness of this taxonomy is that it combines Bloom’s classification principles [27] with the logic of

reductionism and abstract reasoning.

5. In contrast to the previous dichotomy of subjective and objective classification, this study focuses on the

organism and its environment of different DFs.

6. This study sets a stepping stone for improving SDM capability by leveraging AI/ML.

Figure 2: Processes of Decision Framing and AI/ML for SDM

1.3 Scope of This Study

The rest of the paper is organized as follows: Section 2 is a simple introduction to the method of how we

classify SDM and DFs. Section 3 first articulates the concept of SDM and then classifies some related terms

into six groups under the umbrella of what it means to be “strategic”. Then, we argue why we want to focus

directly on DF instead of SDM. Section 4 provides an extensive survey and details of the novel taxonomy of

Represent

Representational space

Evaluate

Loss function on data

Select

Optimizer

Foundation of Machine learning

Decision

Frames

(or Rules)

Computer Science

Artificial Intelligence (AI) = Subfield of Computer Science

Machine Learning (ML) = Subfield of AI

GOFAI

Rules

Dataset

Outputs

Rules

Dataset

Outputs

ML (Neural Net) Representational space: for Possible Rules

Certainty Risks Uncertainty Ambiguity Chaos /

Ignorance

Strategic Decision-Making

Multiple Frames Different Hypothesis

Decision Framing

Reverse logic

5 
DFs.  The  taxonomy  consists of  three  layers  based  on  reductionism  logic  and  abstract thinking.  Section  5 
discusses the implications of the new taxonomy. Section 6 highlights future challenges, conclusions, and our 
view of the future direction for SDM research. 
2  RESEARCH METHOD 
We can adopt different research methods to study DF, not limited to statistical, observational, case-study, 
quantitative, qualitative, experimental or nonexperimental, etc. We classify these methods into four categories: 
observation, quest, inquiry, and empirical design [28]. Among them, the most compelling category is the inquiry 
method (See Figure 9 in Appendix) because the characteristic of the survey and taxonomy is exploratory, 
explanatory, and descriptive. The exploratory survey aims to learn more about a topic than previous researchers 
have done. The descriptive study aims to answer why something (e.g., a problem) is the way it is. Explanatory 
classification  is  to  answer  why  and  how  questions.  In  addition,  we  also  combine  Bloom’s  taxonomy  [27] 
approach for various  inquiries because  Bloom’s  method  is  a  top-down classification approach that fits our 
purpose. We intend to figure out how things stand for their principles first and then go from there to infer how 
the theories can be applied in practice. 
3  STRATEGIC DECISION-MAKING, CLASSIFICATION, FRAMING, AND INTELLIGENCE 
3.1  Strategic and Representation Model 
SDM is often very challenging to define because “strategic” has almost become anyone’s means to an end. 
We can easily list at least 81 terms that may fall under the umbrella of "strategic". Although " strategy " originates 
from warfare, it is now part of our everyday vocabulary. Murray [1] indicated that "the concept of "strategy" has 
proven notoriously difficult to define.", and many theorists failed to clarify the essence of the meaning because 
"theories all too often aim at fixed values, but in war and strategy, most things are uncertain and variable.". 
Although a strategy can be vague, it does not mean it is undefinable. The common definition is that executing 
a strategy usually has long-term and profound impacts beyond the ordinary and fragmental. It is often contrasted 
to tactical and operational decisions, which are short-term focused and isolated.  
Historically, "strategic" is derived from "strategy". The lexical meaning of strategy is a plan of action designed 
to achieve an enduring or overall goal rather than isolated objectives. The origin of strategy is drawn from the 
Greek word "stragegia", which stands for generalship. Therefore, it also represents the art of planning and 
directing overall military operations in a war. 
Practically, we can find that many business terms are associated with “strategic”. One of the primary terms 
is “strategic management”. Traditionally, strategic management [29] often uses case studies to develop future 
business strategies. It is compelling for a particular or static environment. However, it does not fit into a complex 
and dynamic situation because we constantly need to alter our current view and update our representation 
model in our memory or a database. Minsky called the model a frame that is “a data structure for representing 
a stereotyped situation…,” “We can think of a frame as a network of nodes and relations.”. [30] 
3.2  Why Decision Frames 
Minsky’s definition gives us some clues on how to create “a data structure”, which simplifies and summarises 
a large quantity of information so that a user can make sense of it. Consequently, we must pay attention to 

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

StrategicDecisions:Survey,Taxonomy,andFutureDirectionsfromArtificialIntelligencePerspectiveFoundationofARepresentationSpaceforMachineLearningCaesarWu*UniversityofMelbournecaesar.wu@computer.orgKotagiriRamamohanarao1InstitutionofEngineersAustraliarkotagiri@gmail.comRuiZhang2TsinghuaUniversityrui.zhan...

展开>> 收起<<

Strategic Decision s Survey Taxonomy a nd Future Direction s from Artificial Intelligence Perspective Foundation of A Representation Space for Machine Learning.pdf

共32页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Strategic Decision s Survey Taxonomy a nd Future Direction s from Artificial Intelligence Perspective Foundation of A Representation Space for Machine Learning

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: