Recently, Demis Hassabis, the founder of DeepMind, admitted that the super artificial intelligence developed by him has the risk of exterminating mankind. The current AI development competition has entered an unstoppable state, and it is difficult for humans to effectively control it through external governance methods.He completely abandoned the previous idea of ​​relying on systems and ethics committees to restrict AI, and admitted that the governance structure would be difficult to play a role in key decisions, and he could only rely on himself to gain the right to speak to control the direction of AI safety.

Hassabis has always been a staunch advocate of AI safety. In his early years, he envisioned creating absolutely safe superintelligence through independent supervision and secret research and development.


When DeepMind was sold to Google in 2014, it also explicitly required the retention of independent decision-making power on AI security, the establishment of an external oversight committee, and the prohibition of military applications.

However, these security concepts failed one after another. After attending relevant meetings, Musk turned to founding OpenAI, breaking the idea of ​​a single subject controlling AI. Subsequent governance negotiations with Google also ended in failure.

The emergence of ChatGPT in 2022 has completely disrupted the pace of AI research and development.Google fell into competition anxiety, DeepMind merged with Google Brain, security regulations gave way to research and development speed, and the previous security defense lines gradually failed.

Hassabis has thus come to understand the reality. Under the pressure of business competition and technological competition, the non-profit governance mechanism that relies on business giants will find it difficult to survive.

Now Hassabis has changed his thinking. His core proposition is to allow himself to occupy the core decision-making position and control the direction of technology at key nodes with awe of AI risks.

While promoting the development of models such as Gemini, he adheres to the bottom line of safety and tries to use his personal influence to reduce the risk of AI runaway.

Industry analysts believe that this change reflects the cruel current situation of the global AI competition. Unified governance is difficult to achieve, the AI ​​safety window continues to narrow, and mankind is facing huge challenges related to the survival of civilization.