AI Security, Compliance, and Explainability Training

Ethics and Regulation

• What is an AI System?
• View of AI System
• AI System Classifications
• Branches of AI Today
• AI by the numbers
• AI – the Good
• AI – the Bad
• Principles of AI Ethics
• Principles of AI Ethics
• Fairness
• Accountability
• Transparency
• Explainability
• Privacy and autonomy
• Reliable
• Ask ChatGPT 3.5
• AI Ethics in Practice
• Regulatory Compliance in AI Systems
• What are the benefits of AI regulation?
• What are the disadvantages of regulating AI
• Regulations and standards in AI
• GDPR and data protection
• AI in healthcare (HIPAA and other relevant laws)
• AI in healthcare examples
• AI in finance and regulatory compliance
• US FINRA AI Deployment
• AI in US finance examples
• AI in the global finance examples
• Case studies of AI non-compliance
• Addressing Regulatory and Compliance
• Dangers of Discrimination and Bias
• Data Security and Data Privacy
• Control and Security Concerns of AI
• Cooperative Corporate Compliance

Security and Privacy

• What is AI Cybersecurity?
• Threats and challenges in AI security
• Implementing AI in cybersecurity
• Adversarial attacks
• Model inversion and extraction
• Data poisoning
• Best practices for securing AI systems
• Robustness techniques
• Differential privacy
• Federated learning
• Homomorphic encryption

Secure AI Design and Deployment

• Secure Software Development
• Connectivity
• Exploitation of AI Systems (Jailbreaks)
• Infrastructure Concerns
• System Vulnerabilities
• Data Privacy
• Data Leaks via Generating Text
• OpenAI GPT-3/4 Data Location and Storage
• Azure OpenAI
• Adversarial Attacks
• Malicious Use of AI
• Bias and Discrimination
• Regulatory and Ethical Considerations
• Security and Privacy in Chatbots
• Ensuring Security and Privacy
• Data Protection
• Enforcing Data Protection
• Anonymization Techniques
• Best Practices for Security with Generative AI
• Sources of Bias in AI
• Tackling AI Bias
• Real-world Case Studies
• Autonomous Vehicles and the Trolley Problem
• AI in Warfare and Weaponization
• AI in Criminal Justice

AI Auditing and Certification

• Introduction
• Organizational Roles in AI Ethics and Compliance
• Implementing AI Ethics Guidelines and Checklists
• Key Components of an AI Audit
• Steps in the AI Auditing Process
• Post-Deployment Monitoring and Feedback Loops
• Reporting and Recommendations
• AI Certification Process

Explainable AI (XAI)

• Introduction to Machine Learning Interpretability
• Importance of ML interpretability
• Different types of ML interpretability models
• Model-agnostic interpretability methods
• Model-specific interpretability methods
• Limitations of model-specific interpretability
• Limitations of Model-agnostic interpretability
• Global vs. Local interpretability
• Interpretability in Deep Learning
• Techniques and Methods for Explainability
• Layer-wise relevance propagation (LRP)
• Sensitivity analysis
• Gradient-weighted class activation mapping (Grad-CAM)
• Evaluating Interpretability
• Techniques for evaluating interpretability
• Overview of existing evaluation frameworks
• Model-Agnostic Visual Analytics (MAVA)
• Human-AI Collaborated Evaluation (HACE)
• Interpretability in Large Language Models
• Interpretability in Generative LLM’s
• Common evaluation metrics for generative AI models
• Common evaluation metrics – Diversity metrics
• Common evaluation metrics – Likelihood
• Common evaluation metrics – Perplexity
• Common evaluation metrics – Inception Score
• Common evaluation metrics – FID
• Common evaluation metrics – BLEU
• Common evaluation metrics – ROUGE
• Common evaluation metrics – Human evaluation
• Techniques for Interpreting Large Language Models
• Importance of XAI in various sectors
• XAI in Healthcare: Enhancing Care and Transparency
• XAI in Finance: Driving Decisions and Building Trust
• XAI in Legal Systems: Fairness and Accountability

Lab Exercises

• Lab 1. AI Ethics and Regulation
• Lab 2. Understanding security and privacy
• Lab 3. Learning the CoLab Jupyter Notebook Environment
• Lab 4. Guardrails with template manual
• Lab 5. Guardrails with system prompt
• Lab 6. Optional – Implementing Nemo Guardrails for LLM Response Restriction Overview
• Lab 7. Designing an Audit Process for OpenAI’s ChatGPT
• Lab 8. AstroZeneca Ethics-Based AI Audit Framework Design
• Lab 9. Lab 1 – Designing a Gender Bias Test for a Large Language Model (LLM)
• Lab 10. Exploring Machine Learning Interpretability (MLI) with H2O’s Driverless AI Overview