AI Extinction Fears Are ‘Science Fiction’: Andrew Ng

2026-09-19 · Industry Analysis

Andrew Ng presents a strongly optimistic but not risk-free view of AI. He argues that public discussion has become disproportionately focused on existential threats while paying insufficient attention to AI’s practical benefits and more immediate risks such as cybersecurity. A central theme is control. Ng rejects the idea that technology must be perfectly controllable before society can deploy it. He compares early AI with early aviation: aircraft could never be controlled perfectly, but engineering, operational experience and safety systems progressively made flying much safer. He expects AI safety to develop similarly through deployment, observation, testing and iterative improvement. Ng also argues that responsibility should primarily sit with the people and organisations deploying AI systems. His analogy is a hammer: manufacturers should exercise reasonable care when building the tool, but responsibility for its use largely belongs to its operator. He applies similar reasoning to AI, while acknowledging that providers still need appropriate safeguards. The interview also addresses recent examples of models behaving unexpectedly. Ng says such behaviour can be surprising but argues that frontier laboratories normally test models in contained environments, study failures and strengthen safeguards. He therefore favours continued experimentation with strong sandboxing rather than broad restrictions based on worst-case predictions.