Plain-language explanation.
Artificial intelligence is transforming society — automating jobs, personalising information, making decisions about people's lives, and changing what it means to be human. Understanding AI's social implications requires not just technical knowledge but also ethical reasoning, policy analysis, and sociological imagination. Who benefits from AI? Who is harmed? Who decides?
Core concepts and standard treatment.
AI capabilities and hype: narrow AI (task-specific systems — computer vision, NLP, game-playing AI) vs. general AI (hypothetical human-level general intelligence); large language models (GPT-4, Gemini, Claude — transformer-based, trained on vast text corpora, capable of in-context learning); foundation models and emergent capabilities; and the challenge of evaluating AI capabilities (benchmark saturation, contamination, and the gap between benchmarks and real-world performance). Automation and labour: task vs. job automation (Autor: cognitive non-routine tasks increasingly automatable — affecting white-collar work previously thought automation-proof); labour market polarisation (hollowing out of middle-skill jobs); and debate on net employment effects (technological unemployment vs. new task creation).
Deeper theory, debates and edge cases.
Algorithmic bias: face recognition systems with higher error rates for darker-skinned women (Buolamwini and Gebru's Gender Shades study); recidivism algorithms (COMPAS) showing racial disparities; and hiring algorithms discriminating against women (Amazon's scrapped recruiting tool). Fairness notions (demographic parity, equalized odds, predictive parity, individual fairness) are mathematically incompatible — no single technical fix. AI in high-stakes decisions: healthcare (diagnostic AI, triage, drug discovery), criminal justice (predictive policing, bail and sentencing recommendations), financial services (credit, insurance, fraud detection), and benefits administration — each with specific accuracy requirements, explainability needs, and regulatory frameworks.
How it is applied in practice.
AI governance: EU AI Act (risk-based regulation: prohibited, high-risk, and minimal-risk categories; conformity assessment, transparency requirements, and fundamental rights impact assessments); US AI Executive Order; UNESCO AI Ethics Recommendation. AI safety: alignment problem (ensuring AI systems pursue intended goals); mesa-optimisation and emergent objectives; AI red-teaming and safety testing. Responsible AI frameworks in organisations (Google, Microsoft, Meta — each with different emphasis on fairness, safety, transparency, accountability, and privacy). AI literacy: the need for non-technical stakeholders (policymakers, executives, educators, journalists) to understand AI well enough to govern, commission, and critique AI systems effectively — creating new educational and professional development imperatives.