#350 How to Make Hard Choices in AI with Atay Kozlovski, Researcher at the University of Zurich

2026-03-09 · Source: DataFramed · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, extended

Summary

Atay Kozlovski, a Postdoctoral Researcher at the University of Zurich's Center for Ethics, discusses the pervasive ethical issues in AI deployment, particularly in high-stakes environments like healthcare, government, and warfare. He highlights common failure modes such as automation bias, where users blindly trust AI recommendations, and algorithmic bias, which can lead to discriminatory outcomes. Kozlovski cites examples like the Israeli Defense Forces' "Lavender" system, which generated a kill list with a 10% false positive rate, and a hospital's sepsis alert system that led to unnecessary invasive procedures due to false alarms and punitive protocols. He also explores the ethical complexities of deepfakes and AI simulations of human beings, distinguishing between generic AI companions and personalized simulations, and examining use cases ranging from educational tools to grief-related recreations and political advocacy, emphasizing the critical role of consent and authenticity.

Key takeaway

For CTOs and AI/ML Directors deploying high-stakes AI systems, your teams must adopt a "slow and steady" approach, prioritizing comprehensive socio-technical analysis and robust human oversight. You should design systems from inception to ensure clear human accountability, answerability, and attributability for all AI-driven decisions, rather than retrofitting these critical ethical safeguards. Actively combat AI hype within your organization and foster a culture of critical thinking to anticipate and mitigate potential ethical disasters before deployment, especially in sensitive domains like healthcare, government, or personalized AI interactions.

Key insights

AI's ethical deployment requires critical thinking, robust oversight, and a deep understanding of socio-technical contexts to prevent harm.

Principles

• Responsibility is solely attributable to moral agents, not AI systems.
• AI systems must be designed for accountability, answerability, and attributability.
• Context-dependent analysis is crucial for ethical AI implementation.

Method

The "Meaningful Human Control" (MHC) framework advocates for indirect control through system design, focusing on socio-technical analysis, tracing responsibility, and tracking the AI's reasoning process to ensure ethical outcomes.

In practice

• Implement protocols for human intervention when AI systems err.
• Design systems to identify human accountability for AI-driven outcomes.
• Be critical of AI hype and prioritize risk aversion in deployment.

Topics

• AI Ethics
• Meaningful Human Control
• Automation Bias
• Deepfake Technology
• Responsible AI Deployment

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Ethicist, AI Researcher, Policy Maker

Related on AIssential

• AI as a Double-Edged Sword: Opportunities and Ethical Challenges — AI's dual nature demands responsible development, balancing innovation with ethical safeguards like transparency and human oversight.
• AI raises profound moral questions—for all of humanity to answer — AI systems are now making moral choices previously reserved for humans, raising profound ethical questions.
• HCLTech: How to Ensure AI Compliance and Responsibility — Agentic AI's proactive capabilities demand integrated governance and ethical principles from design to deployment.
• AI Might Be Conscious!? (They Don't Know) — AI advancements promise significant societal benefits but introduce complex challenges requiring careful governance and human oversight.
• Make AI Boring Again (xpost) — Engaging with AI, despite its harms, is a moral responsibility for tech professionals to shape its disciplined and accountable future.
• Deadline Day for Autonomous AI Weapons & Mass Surveillance — AI reliability and ethical governance are critical for deploying autonomous systems, especially in military and surveillance contexts.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by DataFramed.