AI Leaders Warn of Extinction-Level Risks Post-White House Accord

Leading AI developers recently voiced concerns about severe risks from advanced artificial intelligence. These statements occurred despite earlier commitments made at a White House summit. The summit, held on 21 July 2023, saw seven major AI companies sign voluntary safeguards. These pledges included external security testing and information sharing with the U.S. government.

Sam Altman, CEO of OpenAI, stated that advanced AI could pose an extinction-level threat. He also said that mitigating such risks is impossible without cooperation from multiple parties. Demis Hassabis, CEO of Google DeepMind, echoed these concerns. He described the dangers as ranging from misuse to loss of control, categorising them as a potentially 'extinction-level' risk.

These public warnings indicate a perceived disparity between official agreements and underlying anxieties within the AI community. The accord committed signatories to develop secure systems. They also agreed to prioritize research into societal risks, including bias and privacy. However, the subsequent public pronouncements focused on more extreme outcomes.

The White House initiative sought to establish responsible AI development. It outlined principles such as safety, security, and trust. Companies like Anthropic and Microsoft were also signatories. Their participation aimed to ensure AI benefits society while minimising harm. The expressed concerns now raise questions about the sufficiency of these current measures. Are the agreed safeguards adequate for the most severe predicted risks?