Trust by Design: Creating Responsible AI-Powered Services

An AI system falsely accused thousands of families of fraud. Learn how to build responsible systems that prioritize people over algorithms.

#1about 2 minutes

How AI is currently used in public administration

AI applications in government range from low-risk chatbots to high-risk systems for predictive analytics, fraud detection, and automated decision-making.

#2about 2 minutes

Establishing a framework for trustworthy AI development

A three-pronged approach to building trust involves training teams on ethics, adhering to a code of conduct, and embedding ethics enablers in projects.

#3about 2 minutes

OpenAI's framework for responsible AI development

OpenAI uses a 'teach, test, and share' framework that includes curating training data, red teaming models, and incorporating user feedback to ensure responsible development.

#4about 4 minutes

Examples of AI systems causing societal harm

Case studies from Austria and the Netherlands reveal how biased AI systems in government led to discrimination and injustice in job placement and child benefits.

#5about 3 minutes

Practical design patterns for building user trust

Implementing features like transparent automation notifications and opt-in voice bots gives users autonomy and control, which helps build trust in AI systems.

#6about 5 minutes

The critical role of transparency and model cards

True transparency in AI involves not just explaining decisions but also clarifying the system's core purpose and using tools like model cards to document its behavior.

#7about 3 minutes

Navigating the impact of regulation on AI innovation

While regulations like the EU AI Act may introduce process overhead, they provide a necessary risk-based framework for ensuring AI is developed responsibly.

#8about 3 minutes

Core principles for developers building trustworthy AI

Developers should embrace their societal responsibility, adopt an iterative 'build, ship, learn' mindset, and always center the specific use case and affected stakeholders.

#9about 2 minutes

How non-developers can contribute to better AI

Everyone can contribute to more trustworthy AI by educating themselves, actively using the tools, providing feedback, and engaging in public discourse about its use.