AI Ethics and Future
As AI systems make increasingly consequential decisions, ethical considerations become critical. AI ethics addresses fairness, transparency, accountability, safety, and societal impact.
Bias and Fairness
AI bias occurs when systems produce discriminatory outcomes. Sources: biased training data, biased labels, biased features, and historical societal biases. Bias affects hiring tools, loan decisions, criminal justice, and healthcare. Fairness metrics and bias auditing help identify and mitigate bias.
Explainability
Explainable AI (XAI) makes AI decisions interpretable. Black-box models are powerful but opaque. Techniques: LIME (local explanations), SHAP (Shapley values), attention visualisation, decision trees. Required in regulated domains (healthcare, finance).
Privacy
AI raises privacy concerns through data collection, facial recognition, behavioural profiling, and surveillance. Federated learning trains without centralising data. Differential privacy adds noise to protect individuals.
Safety and Control
AI safety ensures systems behave as intended. Concerns: specification gaming, distributional shift, reward hacking, autonomous weapons. Alignment research aims to ensure AI goals match human values.
Regulation
The EU AI Act classifies AI by risk level. Approaches: industry self-regulation, ethics boards, algorithmic auditing, impact assessments. Balancing innovation with protection.
Future of AI
Trends: larger models, multimodal AI, AI agents, AI in science (protein folding, drug discovery), personalised AI assistants. The path to AGI remains uncertain but research accelerates.
AI in Nepal
Academic research (KU, TU), startups using ML for agriculture, healthcare, Nepali NLP. Challenges: data availability, computing infrastructure, talent retention.
Summary
AI ethics ensures technology serves humanity fairly and safely. Addressing bias, explainability, privacy, safety, and regulation is essential as AI becomes more powerful.