AI New Era Approaching: DeepMind CEO Calls for Oversight Mechanism

(Central News Agency reporter Lin Hong-han, Los Angeles, 14th) Demis Hassabis, CEO of Google DeepMind, wrote an article stating that 'Artificial General Intelligence' (AGI), capable of thinking and solving problems like humans, may emerge in just a few years, with an impact comparable to humanity's historical discovery of electricity or mastery of fire. He calls for the United States to establish an AI oversight mechanism and set international common standards.

Hassabis (Demis Hassabis) jointly received the Nobel Prize in Chemistry in 2024 with his research partners. They developed artificial intelligence (AI) technology to predict the structures of over 200 million proteins, significantly shortening biomedical research timelines.

Hassabis is a Nobel laureate and CEO of Google's DeepMind, leading teams in developing AI models such as Gemini, making him a pivotal figure in global AI development. Today, he published an article on social media X, offering recommendations on AI development.

'We have found a way to make sand think,' Hassabis wrote, expressing cautious optimism about AI's evolution. The 'sand' he refers to is silicon, used in computer chips. AI performs massive computations using semiconductor chips, developing capabilities in language and problem-solving.

He believes 'Artificial General Intelligence' (AGI) could emerge in just a few years, having a tremendous impact on human civilization—ten times greater than the Industrial Revolution, and advancing at ten times the speed. He points out that AGI is not just another technological breakthrough like the internet or mobile communications; 'it is more like humanity discovering electricity or learning how to use fire.'

'Looking back over the next few decades, this moment may be a pivotal turning point in human civilization,' he said. AGI could accelerate the development of new medicines and clean energy, create new materials, and 'free humanity from resource constraints, ushering in an astonishing age of abundance.'

Current AI tools like ChatGPT, Gemini, and Claude can already write articles, translate, analyze data, and write code. However, the 'Artificial General Intelligence' Hassabis envisions will be much closer to human capabilities—not only following instructions and answering questions but also learning across domains, planning actions, and independently finding solutions to unseen problems.

Hassabis welcomes the prospect of AGI leading human civilization into another golden age and a new economic model. However, he also warns that cutting-edge AI capabilities are rapidly surpassing human understanding, and while moving toward prosperity, society faces cybersecurity, nuclear, and biochemical safety risks.

He notes that advanced AI is already posing challenges to cybersecurity, and as capabilities grow, other threats in nuclear and biological domains may emerge. If AI evolves to autonomously execute tasks, safeguards must be in place to prevent deceptive or misleading behaviors that could endanger human safety.

Hassabis states that AI development is caught in an extremely intense, multi-layered competition involving commercial and geopolitical interests. While competition can accelerate innovation, under conditions of high uncertainty and risk, cautious optimism is the wise and correct strategy. 'Public policy must promote innovation while encouraging responsibility and ensuring safety, considering how AI can benefit society.'

He calls for the establishment of an advanced AI standards body in the United States, modeled after the Financial Industry Regulatory Authority (FINRA). Funded by the industry and staffed by technical experts, it would operate under federal government supervision. Any advanced AI model exceeding a certain capability threshold would be required to undergo testing.

He suggests that AI companies voluntarily submit their models for testing 30 days before release, and only those passing the evaluation could enter the U.S. market. If risks increase, the oversight body could even coordinate with major AI companies to slow down development. He hopes the U.S. will establish such a system to drive international common standards. (Editor: Zhang Zhi-xuan) 1150715