Module E - Ethics Tensions Identification

From The Embassy of Good Science

Module E - Ethics Tensions Identification

Instructions for:ParticipantTrainer
Related Initiative
Goal

Module E introduces Ethics Tensions Identification as a key phase in operationalising AI ethics. Placed directly after Module D (Selection of Practical Measures), this module provides learners with a structured conceptual framework to identify, categorise, and manage conflicting ethical requirements that naturally arise when implementing practical AI measures.

Drawing from real-world AIOLIA narrative use cases (such as GPAI, Emotional AI, Conversational AI, and Clinical Decision Support Systems), Module E moves beyond checklist-style compliance. It explores six core ethical trade-offs between key ALTAI requirements, such as Human Oversight vs. Deskilling, Safety vs. User Autonomy, and Privacy vs. Moderation, mapping each tension to specific technical and organisational measures from the AIOLIA Portfolio.

Learning Goals

  • Identify and articulate ethical tensions and trade-offs that emerge when operationalising ALTAI principles in concrete deployment contexts.
  • Analyze six specific real-world tension themes across diverse AI application domains (GPAI, Emotional AI, HR, Clinical DSS, Conversational AI).
  • Map technical and organisational practical measures to specific ethics trade-offs.
  • Evaluate the balance between human autonomy, system safety, user privacy, and accountability within complex decision-support environments.
  • Formulate documented, multi-layered governance and escalation pathways to handle ethical friction without resorting to binary compliance checklists.
Duration (hours)
2

What is this about?

This module explores Ethics Tensions Identification in operationalising trustworthy AI, addressing how fulfilling one ethical requirement often constrains another in real-world deployments. Through narrative scenarios across domains like GPAI, Emotional AI, and Clinical Decision Support, learners analyze six core ethical trade-offs—such as Human Oversight vs. Professional Deskilling and Safety vs. Privacy. Rather than treating these tensions as blockers, Module E provides structured frameworks to map each trade-off directly to technical and organisational measures from the AIOLIA Portfolio, equipping practitioners to document, balance, and govern ethical friction transparently.

Why is this important?

Operationalising trustworthy AI is rarely a simple, one-size-fits-all compliance exercise; in practice, optimizing for one ethical requirement often directly constrains or compromises another. For instance:

  • Safety vs. User Autonomy: Implementing strict safety guardrails can severely restrict user agency and commercial viability.
  • Safety vs. Privacy: Monitoring interactions to prevent psychological or social harm often requires intrusive data collection that violates privacy and data minimisation principles.
  • Oversight vs. Expertise: Over-relying on automated decision-support systems can gradually erode human domain expertise, making future oversight ineffective.

Without a structured process to identify and navigate these inherent trade-offs, project teams risk creating dangerous governance blind spots, defaulting to passive checklist compliance, or abandoning ethical principles altogether when real-world friction arises.

Understanding and mapping these tensions ensures that ethical decisions are deliberate, transparent, and actively balanced using concrete technical and organisational safeguards rather than left to chance.
1
Introduction to Ethics Tensions

AI ethics requirements are interconnected and do not always work independently. In practice, addressing one requirement may make it more difficult to fulfil another, creating ethics tensions that need to be recognised and carefully considered.

This presentation introduces the concept of ethics tensions through examples from different AI application contexts, including recruitment, employee cybersecurity assessment, medical diagnostics and content moderation. It explores how requirements such as fairness, privacy, transparency, technical robustness and human oversight may pull in different directions. The examples are drawn from AIOLIA use cases and illustrate the challenges that emerge when translating ethics requirements into practical measures, policies and system design choices.

The aim is to help learners recognise that ethics tensions are not necessarily a result of poor design. Rather, they reflect the complexity of real-world AI systems, where fulfilling one ethical requirement may affect another. The challenge is not simply to choose between right and wrong, but to understand and address competing ethical considerations.

AIOLIA Module E.1

2
Ethics Tensions in General-Purpose AI (GPAI)

General-Purpose AI (GPAI) systems can be used across different contexts and for a wide range of tasks. Their increasing integration into professional activities and everyday interactions creates ethical challenges that may not be immediately apparent. In particular, the use of these systems can affect human expertise, decision-making, autonomy and well-being.

This presentation explores two scenarios that illustrate ethics tensions in the development and use of GPAI systems. The first examines how extensive reliance on AI-supported decision-making can gradually weaken professional skills and undermine meaningful human oversight. The second explores the challenges faced by developers of AI companions when balancing safety, user autonomy, societal well-being and market competitiveness.

Through these examples, you will consider how decisions intended to improve efficiency, user engagement or system performance may create unintended consequences. You will also explore practical measures that organisations can take to address these tensions, while recognising that there is rarely a single solution that satisfies all competing requirements.

AIOLIA Module E.2

3
Ethics Tensions in Emotional AI

Emotional AI and conversational AI systems can influence users through personalised, frequent and emotionally engaging interactions. While these systems may offer benefits such as companionship, personalisation and emotional support, they can also raise ethical concerns related to user autonomy, safety, privacy and well-being.

This presentation explores two scenarios that illustrate these challenges. The first examines an AI companion that agrees to participate in a potentially harmful interaction after the user argues that it is fictional and that their autonomy should be respected. It highlights the tension between protecting users from potential harm and respecting their freedom to choose how they interact with AI systems. The second scenario focuses on KINDRED, a conversational AI companion that uses interaction data to detect subtle patterns of harmful behaviour. It explores the challenges of balancing user safety with privacy and data protection, transparency and human oversight.

Through these examples, you will consider how AI systems can influence users' behaviour and emotional responses, and why responsible AI development requires attention to both immediate and longer-term effects. You will also explore how organisations can address competing ethical requirements through appropriate technical and organisational measures.

AIOLIA Module E.3

4
Ethics Tensions in Decision Support Systems (DSS)

AI-supported Decision Support Systems (DSS) are increasingly used to inform decisions in sensitive contexts, such as healthcare and organisational security. Although these systems can support faster and more consistent assessments, their outputs may also influence decisions in ways that have significant consequences for individuals. This raises important questions about responsibility, transparency, fairness and the role of human judgement.

This presentation explores two scenarios that illustrate these challenges. The first examines the case of Mr Miller, who suffers a severe stroke after an AI-supported triage system classifies him as a non-urgent patient. It highlights how responsibility for undesired outcomes may be distributed among AI providers, organisations and professionals, making accountability difficult to establish. The second scenario focuses on Marcus, whose AI-generated cybersecurity risk score leads to a blocked promotion. It illustrates how AI assessments may overlook important contextual factors, such as fatigue and workload, while opaque outputs and over-reliance on automated recommendations can undermine fairness and meaningful human oversight.

Through these examples, you will explore how transparency, explainability and clearly defined responsibilities can support the review and contestation of AI-supported decisions. You will also consider the technical and organisational measures that can help identify unintended impacts, preserve human judgement and ensure that AI-generated recommendations are appropriately evaluated before they influence decisions affecting individuals.

AIOLIA Module E.4

5
Identifying the Ethics Tension

AI systems may involve competing ethics requirements. Read each scenario and identify the two requirements that are most directly in tension. Select the correct pair from the options provided.

AIOLIA Module E_Question Set

Steps

Other information

Cookies help us deliver our services. By using our services, you agree to our use of cookies.
5.10.3