Chapter 5 MAIO

What are the 4 main learning goals of Lecture 5 on AI and Decision Making?

Understanding the characteristics of managerial decision-making, the role of AI in decision-making, Human-AI collaborative structures, and the barriers to AI decision-making along with their mitigation strategies.

What is the definition of decision-making?

The action or process of thinking through possible options and selecting one, where a decision is a choice made from available alternatives.

List the 6 steps of the traditional decision-making process.

1. Recognize the need for a decision, 2. Generate multiple alternatives, 3. Analyze the alternatives, 4. Select the best option, 5. Implement the alternative, 6. Evaluate its effectiveness.

What are Mintzberg's three overarching categories of managerial roles?

Interpersonal roles, Informational roles, and Decisional roles.

Name the 4 specific decisional roles of a manager according to Mintzberg.

Entrepreneur, Disturbance-Handler, Resource-Allocator, and Negotiator.

List 6 key barriers to effective human strategic decision-making.

Bounded Rationality, Escalation of Commitment, Time Constraints, Uncertainty, Personal Biases, and Conflict.

What is the definition of organizational structures in the context of decision-making?

Mechanisms that aggregate individual decisions into a group-level decision.

What two classical organizational decision designs were modeled by Sah and Stiglitz (1986)?

A polyarchy and a hierarchy.

How do a polyarchy and a hierarchy fundamentally differ in graph notation?

In a polyarchy, an option rejected by the first agent is passed to the second agent for review; in a hierarchy, an option accepted by the first agent must be passed to the second agent for final approval.

What necessary step must a manager take when making decisions under uncertainty?

The decision-maker must predict the outcome of each available option and choose the one most likely to yield the best result.

What are the two common forms of predictions used by managers that AI can automate or augment?

Estimation (deciding on the value of a variable) and Screening (accepting or rejecting a proposal).

What are the three components of the mutual augmentation framework for intelligence?

Human Intelligence (human's tacit knowledge), Artificial Intelligence (machine's tacit knowledge), and Hybrid Intelligence (human-augmented AI + augmented human intelligence).

According to the performance framework, what are the three structural classifications of decisions?

Type "A" decisions (Full Automation), Type "B" decisions (Still with Human), and Type "C" decisions (Human-AI Collaboration).

How do AI and humans differ regarding the specificity of the decision search space?

AI requires a well-specified decision search space with specific objective functions, whereas humans can accommodate a loosely defined search space.

How do AI and humans differ regarding interpretability according to Shrestha et al. (2019)?

AI has low interpretability due to complex functional forms (black box), while human decisions are explainable but vulnerable to retrospective sense-making.

How do AI and humans compare regarding alternative set size and processing speed?

AI accommodates large alternative sets and is fast with limited trade-offs, whereas humans have limited capacity to uniformly evaluate large sets and experience a high speed-accuracy trade-off.

Why are AI decision outcomes highly replicable compared to human decisions?

AI uses standardized computational procedures, whereas human replicability is vulnerable to inter- and intra-individual factors like experience, attention, context, and emotion.

What are the 4 primary structures for Human-AI collaborative decision-making?

Full Human to AI Delegation, Hybrid 1 (AI to Human Sequential), Hybrid 2 (Human to AI Sequential), and Aggregated Human-AI Decision Making.

Give 3 business examples of Full Human to AI Delegation.

Dynamic Pricing (e.g., Uber), Matching in Platforms, and High-Frequency Trading.

What are the architectural trade-offs of Full Human to AI Delegation?

High specificity, low interpretability, large alternative set, fast decision-making speed, and high replicability.

What is Hybrid 1: AI to Human Sequential decision-making?

A structure where AI handles the initial phase (evaluating a large set of alternatives), but a human makes the final, highly interpretable decision.

Give 3 business examples of Hybrid 1 (AI to Human Sequential) decision-making.

AI in Recruitment, Credit and Loan Assessment, and Drug Discovery (e.g., Novartis).

What is Hybrid 2: Human to AI Sequential decision-making?

A structure where a human handles the initial phase (evaluating a small alternative set) and inputs data into an AI model, which then makes the final decision.

Give an example of Hybrid 2 (Human to AI Sequential) decision-making.

Sport Analytics (e.g., Moneyball peak performance and injury forecasting).

What is Aggregated Human-AI Decision Making?

A structure where the exact same set of alternatives is evaluated simultaneously by both humans and AI to reach a combined decision.

Give a real-world example of Aggregated Human-AI Decision Making.

AI getting a seat in the corporate boardroom to support directors.

What is the structural bottleneck for decision speed in all Hybrid and Aggregated systems?

Human decision-making acts as a processing bottleneck, causing the overall speed to be slow.

What are the 4 primary barriers to deploying AI in organizational decision-making?

1. Lack of Causality, 2. Black Box Nature, 3. Algorithmic Aversion, 4. Organizational Inertia.

Why does Judea Pearl argue that current machine learning is stuck in a rut?

He asserts that modern AI is mostly about curve fitting and correlation rather than true intelligence, which requires understanding cause and effect ("why").

What are the three levels on Judea Pearl's Ladder of Causation?

1. Association (Seeing/Observing), 2. Intervention (Doing/Intervening), and 3. Counterfactuals (Imagining/Retrospection).

What is the practical definition of causality?

X causes Y if and only if changing X leads to a change in Y, while keeping everything else constant.

What approach serves as a mitigation strategy for AI's lack of causality?

Causal ML (Causal Machine Learning), which enables managers to explore what-if questions and predict the outcome of interventions.

What is the core trade-off involved in the Black Box nature of AI?

The Explainability vs. Performance tradeoff: highly accurate models (like Deep Neural Networks) have low interpretability, while highly interpretable models (like Linear Regression) have lower accuracy.

What are the 3 technical mitigation strategies for the Black Box nature of AI?

Deep Explanation (modified deep learning), Interpretable Models (structured/causal models), and Model Induction (inferring an explainable model from a black box).

What is Algorithmic Aversion?

The behavior of discounting or avoiding algorithmic decisions with respect to human decisions, even when the algorithm is a superior forecaster.

According to Mahmud et al. (2022), what are 4 factors that influence algorithm aversion?

Lack of trust, general negative perceptions, preconceived expectations of near-perfection, and a moral obligation to follow one's own decisions.

How can organizations mitigate Algorithmic Aversion according to Dietvorst et al. (2016)?

By giving people some control over the system (allowing them to slightly modify predictions), increasing transparency, and fostering experience.

What is Organizational Inertia in the context of AI adoption?

The structural resistance or dampening of change within an organization when adopting new technology, which can lead to overall project failure.

What are the 3 requirements to mitigate Organizational Inertia during AI adoption?

1. Evaluation frameworks showing clear benefits/costs, 2. A diverse transformation workforce (mix of inside managers, AI experts, staff, and consultants), and 3. New incentives that reward AI exploration.