Phd Defence of Carla Schmitt

Introduction to Carla Smith's Thesis Defense

Overview of the Research

• Carla Smith introduces her PhD dissertation titled "How to be Strategic: Causal Models and Data as Strategic Resources," focusing on strategic decisions in firms, which differ from everyday decisions due to their complexity and significance for success.

Characteristics of Strategic Decisions

• Strategic decisions are characterized by novelty, uncertainty, and the need for long-term predictions. They require understanding interdependencies with other decisions made by various actors.

Example of a Strategic Decision

• An example provided is entering a new geographic market with an existing product, highlighting its novelty and uncertainty regarding market factors that influence success.

Complexity in Decision-Making

The Chess Analogy

• Carla uses a chess analogy to illustrate the complexity of strategic decision-making, emphasizing unpredictability and interconnectedness among players (decision-makers).

Motivation for Research

• The research aims to explore how strategic decision-makers can navigate this complexity effectively.

Key Aspects of the Research

Data-Based Methods and Causality

• The research examines two main aspects: data-based methods (like machine learning and AI for analyzing data) and causality—understanding cause-and-effect relationships in decision-making.

Importance of Causal Knowledge

• Chapter 2 concludes that strategic decisions cannot rely solely on data-driven approaches; they must incorporate causal questions relevant to the specific context.

Challenges in Current Practices

Mismatch Between Methods and Questions

• There is a noted mismatch between machine learning methods based on correlations and the causal questions necessary for effective strategic decision-making. This gap highlights the need for better integration of causal knowledge into data analysis practices.

Causal Analysis in Strategic Decision-Making

Challenges of Machine Learning in Strategy

• The first challenge discussed is explainability; predictions made by algorithms often lack clarity on how they were generated.

• The second challenge is robustness; models trained in one context may not perform accurately when applied to different settings, such as new geographic markets.

• The third issue is bias; algorithms can learn from imperfect data, leading to biased decisions based on flawed predictions.

Structural Causal Modeling

• Chapter 3 introduces structural causal modeling, a framework that combines existing data and methods to address the challenges of explainability, robustness, and bias for better strategic decision-making.

• A key conclusion is that the meaning derived from data relies heavily on assumptions about the causal mechanisms behind it, emphasizing the importance of reasoning about these mechanisms.

Decision-Making Without Data

• Chapter 4 addresses scenarios where decision-makers lack data, creating uncertainty and reliance on past events for future predictions becomes impossible.

• In high uncertainty situations, decision-makers form causal theories to strategize actions aimed at value creation and success for their firms.

Visualizing Causal Theories

• Using causal diagrams—visual representations of variables connected by arrows indicating causality—can enhance inference from these theories and improve strategic decisions.

Competition Among Firms

• Chapter 5 shifts focus to competition between firms adopting data-driven strategies. It explores how this affects competitive dynamics when firms personalize products based on customer preferences.

Product Personalization Example

• An example provided is Netflix's algorithm which customizes user landing pages based on viewing history and preferences. This personalization requires extensive customer data collection.

Strategic Resource Implications

• Data emerges as a strategic resource; firms with superior data can better tailor products to individual preferences, enhancing competitiveness.

Game Theoretic Model Insights

• A game-theoretic model was used to analyze competition among firms with horizontally differentiated products. Key findings indicate that:

• Data-driven methods intensify competition among firms while potentially harming profits.

• Customers benefit from personalized strategies despite potential profit losses for companies.

Broader Conclusions from Research

• Overall findings highlight that causality plays a crucial role in strategy formulation and influences decision-making processes across various contexts.

• Causal models are identified as valuable resources for strategic decisions both when sufficient data exists (Chapter 3), and under conditions of high uncertainty (Chapter 4).

Final Thoughts

• The research suggests that understanding causality in decision-making extends beyond firm strategy into other domains, offering broader implications for various fields.

Understanding Causality and AI Literacy

Importance of Causality in High-Stakes Domains

• The speaker emphasizes the need for education on causality, particularly in complex and high-stakes areas like policy-making and medical decisions.

• Understanding the distinction between causation and correlation is crucial for interpreting AI algorithms effectively in everyday life.

Addressing Bias and Fairness in AI

• Professor Maher commends the candidate's work on AI literacy, highlighting its relevance to current discussions about bias and fairness in machine learning.

• The professor questions how a causal perspective can help identify and mitigate discrimination within AI-supported decision-making processes.

Causal Modeling Approach to Mitigating Bias

• The candidate explains that a causal approach allows for better understanding of how protected attributes (e.g., gender, race) influence outcomes, which is essential for fair predictions.

• Awareness of biases linked to specific variables can lead to corrective measures post-hoc, even if complete removal of bias isn't feasible.

Synthetic Data as a Solution

• The candidate discusses synthetic data's potential to address biases, especially in sensitive domains where historical data may be lacking.

• There are opportunities to use synthetic data responsibly; however, challenges remain regarding bias introduction through this method.

Ethical Considerations in Data Augmentation

• A follow-up question raises ethical concerns about sourcing additional data inputs for creating responsible synthetic datasets.

• The discussion highlights the importance of defining what constitutes fair data practices amidst inherent biases present in all datasets.

Understanding Fairness and Causal Machine Learning

Defining Fairness in Decision-Making

• Fairness can be defined as equal treatment or equal opportunities for all individuals, particularly when considering disadvantaged groups. This raises ethical questions for decision-makers.

Transitioning from Traditional to Causal Machine Learning

• The discussion transitions to the role of Professor Bmans, who emphasizes the importance of causal machine learning (CML) within strategic decision-making contexts.

Approaches in Causal Machine Learning

• The dissertation presents a variety of methodologies including interviews, surveys, experiments, topic modeling, and formal analysis, showcasing a comprehensive approach to causal machine learning.

Shifting Paradigms in AI

• The conversation highlights a shift from traditional machine learning focused on prediction towards integrating causal reasoning into AI systems. This evolution is crucial for enhancing strategic decision-making capabilities.

Agentic Systems and Their Implications

• Current advancements in AI involve large language models that utilize retrieval augmented generation (RAG), marking a transition from mere prediction machines to reasoning engines capable of supporting complex decisions.

The Role of Causality in Strategic Decision-Making

Hybrid Approaches in Management Decisions

• A question arises regarding the potential benefits of combining traditional ML with agentic systems. It prompts an exploration of which types of decisions might still benefit from causal ML versus those where agentic systems excel.

Evaluating Causality in Agentic Systems

• The discussion probes whether causality is inherently part of agentic systems trained on human knowledge or if further development is needed to make these systems truly causal agents.

Causal Knowledge vs. Probabilistic Models

Importance of Causal Knowledge

• While generative AI can make powerful predictions and solve complex tasks, the speaker argues that strategic knowledge must remain causal to effectively apply it across changing contexts and actions.

Limitations of Current AI Models

• Generative AI operates on probabilistic models that learn data associations but lack true understanding of causality. Human judgment remains essential for interpreting predictions within a causal framework.

Future Directions: World Models and Causality

Speculation on Future Developments

• Emerging concepts like world models may bring us closer to integrating causality into AI systems; however, causality is fundamentally viewed as a human domain requiring human interpretation and judgment.

Discussion on Causality and Decision-Making

The Role of Causal Reasoning in Analysis

• The speaker suggests that while causal reasoning can enhance our ability to think and analyze data, it cannot fully replace human judgment.

• Professor Kamufo is introduced as the next speaker, emphasizing the importance of understanding causality in strategic decision-making.

Understanding Causality in Strategic Decisions

• Professor Kamufo highlights the significance of causal reasoning in bridging prediction and decision-making, stressing that understanding requires explanation and causality.

• He raises a critical point about forming priors without sufficient data, suggesting that defining future state spaces may be more challenging than merely understanding existing problems.

Causality as a Framework for Beliefs

• The discussion shifts to how causal reasoning can function as an equivalent to data under conditions of sparse information, aiding in structuring beliefs about outcomes.

• Emphasizing the need for causal assumptions, he notes that organizing beliefs around cause-and-effect relationships is essential for strategic actions.

Uncertainty and Data Collection

• The speaker acknowledges living in a world filled with uncertainty where beliefs must be relied upon; Bayesian updating is mentioned as a method for refining these beliefs based on observations.

• He explains that causality provides a structured way to represent uncertainty and identify necessary data collection efforts.

Creating New State Spaces through Causality

• Professor Kamufo argues that understanding stable mechanisms across different contexts allows strategists to apply knowledge effectively and innovate new solutions.

Behavioral Biases in Decision-Making

Impact of Behavioral Biases on Decision-Making

• Professor Turniser commends the candidate's research quality and connects his question to Chapter 4 regarding structural causal modeling experiments.

• He points out an implicit assumption within the research: decision-makers behave rationally despite acknowledging biases affecting their judgments.

Variability Due to Probabilistic Judgments

• The candidate responds by noting variability among respondents' answers during probabilistic tasks, attributing this difficulty to inherent biases in processing probabilities.

Understanding Causal Diagrams and Their Impact on Decision Making

The Role of Causal Diagrams in Addressing Biases

• The first experiment demonstrates that causal diagrams can help mitigate biases in responses, particularly among individuals without a strategic background.

• Even participants lacking statistical education or knowledge of causal identification show positive effects from using causal models, indicating their utility in understanding complex theories.

• Participants were not informed about advanced concepts like backdoor criteria, yet the presence of a causal diagram still aids in reducing biases.

• In subsequent experiments, allowing participants to draw their own causal models emphasizes the importance of accurately representing these diagrams for effective treatment outcomes.

• The discussion touches on whether rational behavior is assumed; the speaker clarifies that they do not assume rationality per se.

Educational Gaps and Strategic Decision-Making

• Professor Wilms praises the candidate's thesis and highlights a critical gap between necessary causal knowledge for decision-making and the predictive nature of machine learning methods used in organizations.

• There is an identified mismatch between data scientists' awareness of causality and managers' understanding within organizations, which hampers effective decision-making.

• Two perspectives are proposed: developing methods for data scientists to perform causal analysis and educating future managers about this mismatch to enhance organizational discussions.

• Emphasizing causality as part of AI literacy could bridge gaps between roles within organizations, facilitating better communication and understanding.

• The candidate agrees on the need for educational resources but stresses that time constraints hinder deeper engagement with causal methods.

Importance of Education in Causal Analysis

• While methods exist for analyzing causality, there is a significant lack of resources available to those who recognize its importance due to time-consuming nature.

• Education is deemed crucial for building a broader knowledge base regarding how to interpret causal results and understand underlying assumptions necessary for valid conclusions.

• There's a noted disparity in education around causal inference across disciplines; economics has more robust training compared to fields like computer science or data science programs.

• Integrating education on causality into university curricula is essential for preparing students across various fields to engage effectively with AI technologies.

Causal Analysis and Decision-Making in Organizations

The Role of Causal Models

• The importance of education and organizational process adaptation is highlighted for effective causal analysis.

• Complexity in causal models can lead to diminishing returns, necessitating organizations to simplify decision-making processes into smaller, manageable problems.

• Future research should explore team-based approaches to improve modeling accuracy when combining results from various analyses.

Complex Decision-Making Challenges

• A challenge is posed regarding the complexity of decisions like Netflix choices versus privacy fatigue influenced by EU regulations.

• The discussion shifts to higher education, questioning how personalization affects autonomy and creativity compared to simpler choices like streaming services.

Personalization vs. Autonomy

• Personalization in learning may narrow choices, potentially hindering student autonomy and creativity; this contrasts with the engagement seen in simpler platforms like Netflix.

• Psychological mechanisms play a crucial role in understanding how personalization impacts user engagement and co-creation experiences.

Strategic Choices in Education

• The complexity of educational settings requires consideration of psychological factors alongside data privacy concerns when personalizing study programs.

• Chapter five focuses on strategic pricing decisions driven by data collection needs, emphasizing competitive pressures faced by firms.

Data Collection and Competition

• Firms must navigate fierce competition for data while addressing the "cold start problem" inherent in machine learning applications.

• Understanding customer behavior through causal analysis can help firms cluster customers effectively for better service delivery.

Personalization Mechanisms

• Differentiating between mass customization and personalization is essential; both have strategic implications for students' educational choices.

• Students’ willingness to share data for personalized programs reflects broader assumptions about value exchange between users and companies.

Data-Driven Personalization: Impacts on Utility and Profit

The Value of Data in Personalization

• The discussion begins with the assumption that data may lead to disutility, which could be integrated into models assessing customer utility from personalized products.

• Customers derive value from personalized programs, experiencing higher utility compared to standard offerings, particularly in educational contexts.

• Universities must attract students initially to gather preferences, potentially lowering tuition fees as a reward for sharing data.

Implications for Firms and Profitability

• Professor Bummans raises concerns about how data-driven personalization can harm firm profits while benefiting customers and society.

• Despite profitability seen in companies like Netflix, the model suggests firms often monetize customer data through advertisements rather than solely through personalization.

Advertising Monetization Considerations

• Questions arise regarding whether ad monetization was considered in developing the model and if it should be included for a comprehensive understanding of profitability.

• Managers are advised to consider adding ad monetization strategies or explore alternative approaches beyond just relying on data sharing.

Challenges of Data Sharing

• Concerns are raised about the feasibility of data sharing due to competitive dynamics where firms operate under a "winner-takes-all" logic.

Policy Recommendations Regarding Privacy Regulations

• The speaker notes that advertising impacts customer attention and overall product attractiveness, suggesting this should be factored into pricing models.

• The findings indicate that firms face a prisoner's dilemma; they engage in costly activities to mitigate profit loss despite potential negative outcomes from competition.

Revising Perspectives on Data Sharing

• Policymakers are encouraged to reconsider the narrative that data sharing is inherently detrimental to consumers based on these findings.

• A historical perspective is provided indicating that strict privacy regulations may have been introduced without considering their impact on consumer benefits derived from personalization.

Heterogeneity in Market Dynamics

Impact of Existing Firms on Market Entry

• Discussion on how heterogeneity between firms can affect market dynamics, particularly the ability of new firms to enter successfully.

• Concerns raised about existing players with large data resources gaining a competitive advantage, potentially prohibiting new entrants from competing effectively.

Role of General AI in Decision Making

• Inquiry into the role of general AI and agentic AI systems in enhancing machine learning and causal reasoning for strategic decision-making.

Design Principles for Agentic Systems

• Emphasis on design principles that support decision-makers by facilitating understanding of causal relationships.

• Mention of the creative capabilities of generative AI, which can yield unexpected results that challenge existing theories.

Thesis Defense Conclusion

Assessment and Degree Conferment

• Announcement that the degree committee will deliberate on the quality of Carla Schmidt's thesis after her defense.

• Positive assessment from the degree committee leading to the conferment of a doctoral degree upon Carla Schmidt.

Commitment to Scientific Integrity

• Carla's affirmation to uphold principles of scientific integrity during her academic career as part of her doctoral conferral process.

Reflections on PhD Journey

Personal Anecdotes and Experiences

• Carla reflects on her journey, starting with an email from June 2019 that marked the beginning of her collaboration with supervisors.

Challenges and Learning Moments

• Description of initial challenges faced during qualitative research projects, highlighting moments where they had to adapt their methodologies.

Collaboration Insights

• Humorous exchange with her daughter about what it means to be a PhD candidate, emphasizing perceptions around intelligence associated with academic achievement.

• Recollection of learning experiences while managing research projects, including navigating qualitative versus quantitative research methods.

Research Methodology Development

• Discussion about engaging with online platforms for data collection and early sampling experiences that shaped their research approach.

PhD Journey and Impact of Dr. Khales Mitten

Reflections on Decision-Making and Research Perspective

• The discussion highlights the challenges of decision-making under resource constraints, emphasizing the importance of perspective in research, even when sample changes occur.

Constructive Criticism and Academic Identity

• Dr. Khales is described as a "constructive rebel," known for critical questioning and not accepting theories at face value, showcasing her unique approach to academia.

Diverse Academic Engagements

• Beyond her PhD work, Dr. Khales supervised an honors project on Danish ghetto laws, demonstrating her ability to balance multiple academic interests effectively.

Competence in Literature Analysis

• Her exceptional skill in hermeneutics allowed her to connect literature with ongoing conversations, surpassing peers in this analytical capability.

Contributions to Community and Events

• Dr. Khales significantly impacted the academic community through social events and initiating causal data science meetings that expanded globally.

Entrepreneurial Ventures Post-PhD

• Before completing her PhD, she ventured into entrepreneurship with LM Her Acoustic, indicating a proactive approach towards career development.

Acknowledgments of Support Systems

• Gratitude is expressed towards Martin for his unwavering support throughout the PhD journey, highlighting the importance of mentorship.

Family Influence and Personal Growth

• Recognition is given to Dr. Khales' parents for their role in shaping her character; it emphasizes familial influence on personal development during academic pursuits.

Collaborative Experiences Abroad

• Appreciation is shown for international collaborations at Pokuni University which enriched Dr. Khales' research experience beyond local confines.

Lasting Relationships Formed During PhD

• The essence of a PhD extends beyond thesis production; it fosters lasting friendships that enhance life experiences outside academia.

Overcoming Challenges During COVID Era

• Reflecting on initial challenges faced during the pandemic illustrates resilience; despite difficulties like remote learning defaults (e.g., "you're on mute"), success was achieved over time.

Conclusion: Celebrating Achievements

• The speech culminates in congratulating Dr. Khales Mitten on her achievements while expressing hope for future endeavors and continued contributions to academia.