Recommender Systems: Legal and Ethical Issues

Contents
Chapter 1: Introduction: Understanding and Regulating AI-Powered Recommender Systems
References
Part I: Fairness and Transparency
Chapter 2: Recommender Systems and Discrimination
2.1 Introduction
2.2 Reasons for Discriminating Recommendations
2.2.1 Lack of Diversity in Training Data
2.2.2 (Unconscious) Bias in Training Data
2.2.3 Modelling Algorithm
2.2.4 Interim Conclusion and Thoughts
2.3 Legal Frame
2.3.1 Agreement – Data Protection Law
2.3.2 Information – Unfair Competition Law
2.3.3 General Anti-discrimination Law
2.3.4 Interim Conclusion
2.4 Outlook
2.4.1 Extreme Solutions
2.4.2 Further Development of the Information Approach
2.4.3 Monitoring and Audit Obligations
2.4.4 Interim Conclusion and Thoughts
2.5 Conclusions
References
Chapter 3: From Algorithmic Transparency to Algorithmic Choice: European Perspectives on Recommender Systems and Platform Regulation
3.1 Introduction
3.2 Recommender Governance in the EU Platform Economy
3.2.1 Mapping the Regulatory Landscape
3.2.2 Layers of Terminology in EU Law: “Rankings” and “Recommender Systems”
3.3 Five Axes of Algorithmic Transparency: A Comparative Analysis
3.3.1 Purpose of Transparency
3.3.2 Audiences of Disclosure
3.3.3 Addressees of the Duty to Disclose
3.3.4 Content of the Disclosure
3.3.5 Modalities of Disclosure
3.4 The Digital Services Act: From Algorithmic Transparency to Algorithmic Choice?
3.4.1 Extension of Transparency Rules
3.4.2 User Control Over Ranking Criteria
3.5 Third Party Recommender Systems: Towards a Market for “RecommenderTech”
3.6 Conclusion
References
Chapter 4: Black Hole Instead of Black Box?: The Double Opaqueness of Recommender Systems on Gaming Platforms and Its Legal Implications
4.1 Introduction
4.2 The Black Box-Problem of AI Applications
4.2.1 Transparency and Explainability: An Introduction
4.2.2 Efficiency vs. Explainability of Machine Learning
4.2.3 Background of the Transparency Requirement
4.2.4 Criticism
4.2.5 In Terms of Recommender Systems
4.3 The Black Hole-Problem of Gaming Platforms
4.3.1 Types of Recommender Systems
4.3.1.1 Content-Based Filtering Methods
4.3.1.2 Collaborative Filtering Methods
4.3.1.3 Hybrid Filtering Methods
4.3.2 Black Hole Phenomenon
4.4 Legal Bases and Consequences
4.4.1 Legal Acts
4.4.2 Digital Services Act
4.4.2.1 Problem Description
4.4.2.2 Regulatory Content Related to Recommender Systems
4.4.3 Artificial Intelligence Act
4.4.3.1 Purpose of the Draft Act
4.4.3.2 Regulatory Content Related to Recommender Systems
4.4.4 Dealing with Legal Requirements
4.4.4.1 User-Oriented Transparency
4.4.4.2 Government Oversight
4.4.4.3 Combination of the Two Approaches with Additional Experts
4.5 Implementation of the Proposed Solutions
4.5.1 Standardization
4.5.2 Control Mechanisms
4.6 Conclusion
References
Chapter 5: Digital Labor as a Structural Fairness Issue in Recommender Systems
5.1 Introduction: Multisided (Un)Fairness in Recommender Systems
5.2 Digital Labor as a Structural Issue in Recommender Systems
5.3 Fairness Issues from Value Distribution to Work Conditions and Laborers’ Awareness
5.4 Addressing the Problem
5.5 Conclusion
References
Part II: Manipulation and Personal Autonomy
Chapter 6: Recommender Systems, Manipulation and Private Autonomy: How European Civil Law Regulates and Should Regulate Recommender Systems for the Benefit of Private Autonomy
6.1 Introduction
6.2 Autonomy and Influence in Private Law
6.3 Recommender Systems and Their Influence
6.4 Manipulation
6.5 Recommender Systems and Manipulation
6.5.1 Recommendations in General
6.5.2 Labelled Recommendations
6.5.3 Unrelated Recommendations
6.5.3.1 In General
6.5.3.2 Targeted Recommendations
6.5.3.2.1 In General
6.5.3.2.2 Exploiting Emotions
6.5.3.2.3 Addressing Fears Through (Allegedly) Harm-Alleviating Offers
6.5.4 Interim Conclusion: Recommender Systems, Manipulation and Private Autonomy
6.6 Regulation Regarding Recommender Systems
6.6.1 Unexpected Recommendation Criteria
6.6.2 Targeted Recommendations Exploiting Emotions or Addressing Fears
6.6.3 Regulative Measures to Take Regarding Recommender Systems
6.7 Conclusion
References
Chapter 7: Reasoning with Recommender Systems? Practical Reasoning, Digital Nudging, and Autonomy
7.1 Introduction
7.2 Practical Reasoning, Choices, and Recommendations
7.3 Recommender Systems and Digital Nudging
7.4 Autonomy in Practical Reasoning with Recommender Systems
7.5 Conclusion
References
Chapter 8: Recommending Ourselves to Death: Values in the Age of Algorithms
8.1 Introduction
8.2 Distorting Forces
8.2.1 Past Evaluative Standards
8.2.2 Reducing to Computable Information
8.2.3 Proxies for ‘Good’
8.2.4 Black Boxed
8.3 Changing Human Values
8.4 Same Problem with Humans?
8.5 Conclusion
References
Part III: Designing and Evaluating Recommender Systems
Chapter 9: Ethical and Legal Analysis of Machine Learning Based Systems: A Scenario Analysis of a Food Recommender System
9.1 Introduction
9.2 An Example Application: FoodApp- the Application for Meal Delivery
9.3 Current Approaches to Ethical Analysis of Recommender Systems
9.4 Ethical Analysis
9.5 Legal Considerations
9.5.1 Data Protection Law
9.5.2 General Principles and Lawfulness of Processing Personal Data
9.5.3 Lawfulness
9.5.4 Purpose Limitation and Access to Data
9.5.5 Data Minimization and Storage Limitation
9.5.6 Accuracy, Security and Impact Assessment
9.6 Results of the Combined Ethical and Legal Analysis Approach
9.7 Conclusion and Outlook
References
Chapter 10: Factors Influencing Trust and Use of Recommendation AI: A Case Study of Diet Improvement AI in Japan
10.1 Society 5.0 and Recommendation AI in Japan
10.2 Model for Ensuring Trustworthiness of AI Services
10.3 Components of a Trustworthy AI Model
10.3.1 AI Intervention
10.3.2 Data Management
10.3.3 Purpose of Use
10.4 Verification of Trustworthy AI Model: A Case Study of AI for Dietary Habit Improvement Recommendations
10.4.1 Subjects
10.4.2 Verification 1: AI Intervention
10.4.3 Verification 2: Data Management
10.4.4 Verification 3: Purpose of Use
10.4.5 Method
10.4.6 Results
10.4.6.1 AI Intervention
10.4.6.2 Data Management
10.4.6.3 Purpose of Use in Terms of Service Agreements
10.5 Necessary Elements for Trusted AI
References
Chapter 11: Ethics of E-Learning Recommender Systems: Epistemic Positioning and Ideological Orientation
11.1 Introduction
11.2 Methods of Recommender Systems
11.3 Recommender Systems in e-Learning
11.3.1 Filtering Techniques: What Implications on Social and Epistemic Open-Mindedness?
11.3.2 Model Selection: A Risk of Thinking Homogenization?
11.3.3 Assessment Methods: What Do They Value?
11.4 Problem Statement
11.5 Some Proposals
11.5.1 Knowledge-Based Recommendations
11.5.2 A Learner Model Coming from Cognitive and Educational Sciences
11.5.3 A Teaching Model Based on Empiric Analyses
11.5.4 Explainable Recommendations
11.6 Discussion and Conclusion
References