Responsible data dimensions

Responsible AI Data Management

Maria Prokofieva

Lead ML Engineer

Responsible data management

Ethical data management
Evaluate models with technical metrics
Responsible AI

Responsible data dimensions

In this course

Introduce responsible AI dimensions and metrics
Apply concepts to the real world
Overview of regulation and licensing
Data governance and acquisition
Validation and bias mitigation

Lawfulness

Compliance with laws and regulations
Ensures data is collected, processed, and used correctly
Some laws and regulations include:
- Data protection laws
- Human rights laws
- Ethical regulations towards stakeholders
- Can differ depending on the governing body or country

Always confirm what applies!

Responsible data dimensions: lawfulness

Fairness

Algorithms and data practices do not create inequalities
Treat everyone fairly, without discrimination

Responsible data dimensions: fairness

Transparency and accountability

How the data is used
How the model is developed
How decisions are made
Explain the AI
Build stakeholders trust

Responsible data dimensions: transparency and accountability

Diversity and inclusion

Data diversity
Diverse perspectives and experiences
Key for bias mitigation

Responsible data dimensions: diversity and inclusion

Privacy and security

Safeguarding of personal and sensitive data
Respecting and protecting individual rights
Protect data and models from unauthorized access

Responsible data dimensions: privacy and security

Amazon AI hiring tool

Amazon: 2015-2017
Automated talent acquisition
Use AI to rate job applicants
Led to scandal and abandoning of the initiative

amazon AI

¹ Reuters: https://www.reuters.com/article/idUSKCN1MK0AG/

Challenges of AI models

What went wrong?

Not gender-neutral
Imbalanced training data
Used only technical metrics for AI evaluation

imbalanced data

Let's practice!

Responsible AI Data Management

Preparing Video For Download...