Microsoft's AI Principles

Questions and Answers

To create AI systems that prioritize efficiency over fairness

To create AI systems that can learn from their mistakes

To develop AI systems that can function without human intervention

To ensure AI applications provide amazing solutions to difficult problems without unintended negative consequences (correct)

Data preprocessing

Automated feature engineering

Model interpretability (correct)

Hyperparameter tuning

Because they were too expensive to maintain

Because they were not compatible with other Microsoft services

Because they were not accurate enough

Because they could be misused to infer emotional states and identity attributes (correct)

Substantial risk to human life Signup and view all the answers

To respect privacy and prevent personal data misuse Signup and view all the answers

Privacy or security issues Signup and view all the answers

To empower everyone and engage people regardless of their differences Signup and view all the answers

Making users fully aware of the system's purpose and limitations Signup and view all the answers

People should be accountable for AI systems, including designers and developers Signup and view all the answers

To provide a framework for creating ethical AI solutions Signup and view all the answers

Study Notes

AI systems should treat all people fairly, without bias or discrimination based on gender, ethnicity, or other factors.
Examples of bias in AI systems include loan approval models that unfairly deny or approve loans based on certain characteristics.
Azure Machine Learning provides tools to interpret models, quantify feature influence, and identify bias.

AI systems should perform reliably and safely to avoid risks to human life.
Examples of high-risk AI systems include autonomous vehicles and medical diagnosis models.
Rigorous testing and deployment management processes are essential to ensure AI systems work as expected.

AI systems should respect privacy and be secure, protecting personal data used in machine learning models.
Privacy and security concerns persist even after model training and deployment, as new data is used for predictions and decision-making.

AI systems should empower everyone and engage people, providing benefits to all parts of society regardless of physical ability, gender, sexual orientation, ethnicity, or other factors.
Examples of inclusive AI systems include those that support people with disabilities or provide access to education and healthcare.

AI systems should be understandable, with users aware of the system's purpose, functionality, and limitations.
Transparency is critical to building trust in AI systems and ensuring users understand how they work.

People should be accountable for AI systems, working within a framework of governance and organizational principles.
Designers and developers of AI-based solutions should ensure their solutions meet ethical and legal standards.