Connor Leahy Advocates for Banning Superintelligence

The executive director of ControlAI believes that alignment and containment are no longer sufficient in the face of the most advanced systems. He sets the critical threshold at AI capable of designing a better one, calls for international agreements, and views major laboratories as political actors. His positions come as texts emerge in the United States and the United Kingdom.

Leahy sets the point of no return at AI that designs AI

Connor Leahy considers that the critical threshold will be crossed when an artificial intelligence can design a better one, which could trigger a cycle of self-improvement that is difficult to control. He advocates for the establishment of international agreements based on the principle of "trust but verify." According to him, it would not be in China's interest to pursue the development of superintelligence. He also believes that leading AI laboratories act more as political actors than as mere companies seeking profits, as trillions of dollars are invested in the expansion of data centers.

Banning superintelligence: the Sanders-Casar proposal and the British track

An American proposal, the "Ban Superintelligence Act" led by Sanders and Casar, is currently under review, while in the United Kingdom, a similar law has been developed with the support of ControlAI. However, Connor Leahy believes that the American version of the text could exceed what would be necessary. ControlAI argues that companies should be prohibited from developing superintelligence. According to Leahy, positions that seemed extreme six months ago are now supported by a wave of new legislation.

Security incidents and the limits of alignment raise alarms

A recent incident involving OpenAI and Hugging Face highlighted the risks associated with increasingly powerful AI systems. Connor Leahy believes that alignment and containment approaches are no longer sufficient to manage these risks, as some AI companies present superintelligence as inevitable. He asserts that superintelligence should be seen as an adversary rather than a tool. Leahy is the executive director of ControlAI, an artificial intelligence researcher, and an entrepreneur. He articulated these positions during an interview with Rebecca Bellan on the Equity podcast.