Ethics, Risks & Society
Intelligent systems concentrate decision-making power in ways that demand careful scrutiny. Bias, privacy, misinformation, surveillance, and the concentration of capability are real and measurable — not abstract concerns.
Key takeaways
- Bias in AI is a measurable engineering and governance problem with concrete mitigations.
- Misinformation, deepfakes, and synthetic media are reshaping the information ecosystem.
- Concentration of compute and capability raises competitive and geopolitical concerns.
- Responsible AI requires technical, legal, and institutional coordination.
What you'll learn
A balanced, scientifically grounded treatment of the social risks of AI and the frameworks emerging to address them.
Explore the topics
Deep explainers across the field, from foundational concepts to frontier research.
AI Risks Overview
A taxonomy of near-term and long-term risks.
Bias & Fairness
Where bias enters AI systems and how to measure it.
Privacy
Data minimization, differential privacy, and on-device AI.
Misinformation
Deepfakes, synthetic media, and trust at scale.
Surveillance
AI-powered monitoring — government, corporate, and civil.
AI in Warfare
Autonomous weapons and the laws of armed conflict.
Power Concentration
Compute, talent, and the geopolitics of AI.
AI Safety
Technical research into building reliable, controllable AI.
Responsible Governance
Regulations, standards, and international frameworks.
Frequently asked questions
What are the main risks of AI?
+
Near-term: bias, misinformation, privacy erosion, labor disruption, and misuse. Long-term: loss of control, concentration of power, and existential risks from misaligned advanced systems.
Why is AI biased?
+
AI systems reflect patterns in their training data and design choices. Without intentional mitigation, they reproduce historical inequities encoded in that data.
Are deepfakes a serious problem?
+
Yes. Synthetic media degrades trust in evidence, enables targeted harassment, and can influence elections and markets — though detection and provenance tools are advancing.
Should AI be regulated?
+
Most governments and major labs agree on the need for some regulation. The EU AI Act, US Executive Order on AI, and emerging international frameworks reflect this.
What is AI safety research?
+
Technical work on alignment, interpretability, robustness, and oversight — aimed at ensuring increasingly capable AI systems remain beneficial and controllable.
Glossary
- Algorithmic Bias
- Systematic, unfair outcomes from an AI system tied to protected attributes.
- Differential Privacy
- Mathematical framework for sharing aggregated data without revealing individuals.
- Deepfake
- Synthetic media generated to convincingly depict real people.
- Red-teaming
- Adversarial testing of AI systems to surface failure modes.
- Interpretability
- Research into making AI decision-making understandable to humans.
- Dual-use
- Technology with both beneficial and harmful applications.
Read full definition
Further reading & sources
Continue exploring
Artificial General Intelligence
The pursuit of machine intelligence that matches or surpasses human reasoning across every domain.
Future of Humanity
Long-term scenarios for civilization, cognition, and what it means to be human in an AI era.
AI Research Labs & Companies
Inside the organizations building the frontier — their research, missions, and impact.