A Collective Agreement for AI Safety

The AI Action Summit concluded with a pivotal agreement focusing on the responsible development of AI technologies. Interestingly, while nations came together to sign this important document, the United States and the United Kingdom chose to abstain. Notably, even China, a country recognized for its stringent internet regulations and censorship, took part in the initiative.

A striking embodiment of this censorship is highlighted by DeepSeek, a Chinese AI startup that recently captured global attention. Their AI model, DeepSeek R1, has been designed to self-censor in real time, specifically to avoid engaging with topics that are deemed sensitive by Chinese authorities.

The Implications of Non-Signatories

The refusal of the US and UK to endorse the joint agreement raises significant concerns that will likely resurface in discussions whenever incidents of AI misuse arise. Recent examples—such as the widely circulated deepfake video involving anti-Kanye sentiments—underscore the necessity for collective action in regulating AI technologies, especially as key figures in Western governance express their apprehensions about AI potentially spiraling out of control.

Expert Perspectives: Eric Schmidt’s Warnings

Eric Schmidt, the former CEO of Google and a significant investor in the AI domain, stands out as a critical voice in the dialogue surrounding AI safety. Speaking to the BBC during the summit, Schmidt expressed grave concerns about AI being exploited in perilous ways, likening potential scenarios to those orchestrated by malevolent actors in history, such as Osama Bin Laden.

He articulated, “The real fears that I have are not the ones that most people talk about AI—I talk about extreme risk.” Schmidt pointed to nations like North Korea, Iran, and Russia, suggesting that these regimes could leverage AI technologies for malicious purposes, including the orchestration of devastating biological attacks.

“I’m always worried about the ‘Osama Bin Laden’ scenario, where you have some truly evil person who takes over some aspect of our modern life and uses it to harm innocent people,” Schmidt cautioned, advocating for greater governmental oversight of private companies engaged in developing sophisticated AI solutions.

He emphasized the importance of regulators maintaining transparency and vigilance over advancements in AI, insisting that while innovation is crucial, it must occur within a framework of accountability. Schmidt acknowledged the Biden administration’s stance on export controls regarding microchips fundamental to cutting-edge AI operations.

DeepSeek’s Innovations and Industry Dynamics

DeepSeek has garnered significant attention not just for its products, but for the innovative methods it employed to create reasoning-oriented AI without access to advanced chips. By combining legacy hardware with novel software strategies, the startup was able to produce an AI system that competes with giants like ChatGPT. However, this does not undermine the essential role of next-generation GPU technology in the industry.

Despite urging for regulation, Schmidt also warned against the pitfalls of excessive regulation, which could hamper technological progress. While he is no longer affiliated with Google, he remains deeply invested in AI advancements. A recent report from CNBC revealed that Schmidt has actively supported various AI startups, signaling his firm belief in the sector’s transformative potential.

Since 2019, Hillspire, Schmidt’s family investment office, has backed 22 AI startups, with 13 of those investments occurring in the fast-paced year of 2022. The rise of ChatGPT, which gained immense popularity at the end of 2022, signaled an industry-wide shift towards prioritizing AI development.

Schmidt’s investments in noteworthy companies like Anthropic, Holistic AI, and SandboxAQ exceed $5 billion, showcasing his dedication to nurturing innovative AI technologies while advocating for ethical frameworks surrounding their use.