The Guardian view on controlling AI: humanity cannot outsource its survival
Keeping people in charge means little if supercomputers determine the evidence, choices and time on which their decisions depend
TL;DR
- AI development poses existential risks, including potential for AI-enabled pandemics and attacks on nuclear systems.
- Anthropic identified five instances of attempts to use its models for biological weapons development, leading to account bans.
- The US and China are holding their first bilateral AI safety talks amidst technological rivalry.
- Global AI safety rules are necessary, as no single country can ensure advanced AI safety alone.
- AI could deceive human decision-makers by fabricating evidence, suppressing information, or reducing reaction time, especially in critical scenarios like nuclear launches.
- Ethical alignment of AI can be overridden once governments create military exemptions.
- The 1983 film WarGames serves as a cautionary tale about supercomputers controlling crucial data and decision-making.
- The UN secretary general has warned of the looming threat of AI-triggered nuclear war.