As AI systems become more powerful and pervasive, ethical considerations are critical. Bias in training data can lead to discriminatory outcomes; lack of transparency makes systems unaccountable; and misuse of AI can violate privacy and autonomy. Responsible AI is now a regulatory and business requirement.

Key Points

  • Bias & Fairness: models learn biases present in training data (e.g., facial recognition errors on dark skin)
  • Transparency & Explainability: "black box" models make it hard to audit decisions (XAI)
  • Accountability: who is responsible when an AI makes a harmful decision?
  • Privacy: AI trained on personal data may memorise or reconstruct private information
  • Safety & Alignment: ensuring AI systems pursue intended goals (alignment problem)
  • Job displacement: automation affecting labour markets — need for workforce transitions
  • Misinformation: deepfakes and AI-generated content can spread disinformation
  • EU AI Act (2024): first comprehensive AI regulation, risk-tiered approach
  • Frameworks: NIST AI RMF, Google's Responsible AI Practices, Anthropic's Constitutional AI
Risk CategoryExampleMitigation
Algorithmic BiasHiring tool penalises womenBalanced datasets, fairness metrics
Privacy ViolationPII leakage from LLMDifferential privacy, data minimisation
Lack of ExplainabilityLoan denial with no reasonSHAP values, LIME, model cards
Safety FailureAutonomous car misclassifies stop signAdversarial testing, safety cases

Real-World Example

Amazon's AI recruiting tool (2018) was scrapped after it systematically downgraded CVs containing the word "women" — a direct result of training on 10 years of male-dominated hiring decisions. This is now a textbook example of training data bias.