In an interview on The New York Times podcast The Ezra Klein Show, Bill Gates said that AI is the most dangerous thing for humanity, presenting a much graver threat than nuclear weapons. He expressed concern over AI’s rapidly increasing capability in absence of a concrete safety framework.
In recent interviews, Gates has also challenged President Trump’s remarks that AI safety concerns were a “hoax”. He also expressed a desire to meet him and change his position on safety policies for US-based AI models.
‘AI Makes Nuclear Weapons Look Like Nothing’ – Bill Gates
Gates emphasized that AI has surpassed a critical threshold. It can now enable those with malicious intent to design dangerous biological weapons. He further said that AI tools are now capable of designing new molecules, including pathogens that could cause greater damage than any found in nature. These models can also genetically modify existing pathogens and make them more dangerous than smallpox. Traditionally, developing bioweapons required lengthy experiments and expertise. Advanced AI can now provide step-by-step synthesis instructions to untrained individuals. Anyone with access to advanced AI could potentially carry out an act of bioterrorism.
Gates further drew a contrast between nuclear weapons and AI. Developing nuclear weapons is a difficult task requiring rare raw materials, industrial enrichment facilities, specialized engineering and massive amounts of money. Due to these limitations, only nation-states have the capacity to build such weapons.
In comparison, AI is entirely digital. Advanced models can be trained, leaked, copied and accessed remotely. It can be used by anyone. AI could give even a single individual significant operational capabilities because no large-scale industrial infrastructure is required.
The lack of government oversight
Gates stressed the detrimental effects of the lack of governmental oversight over AI. Comparing AI to nuclear weapons, Gates said that nuclear technology was developed as part of government projects such as the Manhattan Project. Therefore, these projects were heavily controlled by the government. Meanwhile, AI is built entirely by private tech giants.
Moreover, because of AI’s rapid development, governments lack the internal technical expertise to understand and regulate it. Gates rejected the idea of companies regulating themselves because the financial incentives against doing so are substantial. In addition, AI is developing at a pace at which where its capabilities are expanding month by month. Legislation takes years to enact, by the time governments draft laws for AI regulation, the nature of the risk may already have changed.
He also noted that establishing a global regulatory framework for AI will be harder than the Cold War-era nuclear non-proliferation treaties. While the expansion of nuclear capabilities can be monitored through satellite imagery by counting the number of missile silos or cooling towers, satellites cannot monitor AI. The AI arms race among geopolitical rivals race also makes international cooperation considerably more difficult.
