Tech executive cautions about the constrained impact of Sunak’s AI safety summit

Connor Leahy proposes that prominent tech corporations are attempting to sway or manipulate discussions among government leaders

An executive invited to Rishi Sunak’s upcoming international AI safety summit has expressed reservations about the potential effectiveness of the event. Connor Leahy, the CEO of Conjecture, an AI safety research company, is concerned that influential tech companies may be trying to exert undue influence on this crucial gathering, potentially allowing the unbridled development of highly advanced AI with nearly limitless capabilities. Leahy is part of a select group of 100 attendees, including foreign government officials, tech leaders, and civil society representatives, who will participate in the summit scheduled for November at Bletchley Park. The UK government aims for this event to signify a turning point in the development of advanced AI technology.

Authorities have unveiled a summit agenda emphasizing the significance of “responsible capability scaling,” implying that companies should create advanced models while adhering to a set of guidelines. Nonetheless, Leahy and others advocate for a complete cessation in the development of artificial general intelligence—AI models capable of tasks at the level of human or superhuman intelligence.

Leahy remarked, “The main objective of responsible scaling is to establish a framework that gives the impression of action, enabling politicians to assert they have taken measures. In reality, the policy remains inadequate.”

Leahy and his colleagues have launched a movement named “Control AI,” seeking to urge policymakers to take an additional step and leverage the Bletchley Park summit to implement a temporary suspension on the development of the most advanced forms of AI. Similar to certain voices within the industry, Leahy is apprehensive that humanity is in jeopardy if we create an AI system that learns to operate beyond human control.

He stressed, “If we build systems that surpass humans in manipulation, business, politics, science, and every other domain, without effective control, then the future may be shaped by these systems rather than by us.”

AI companies and affiliated organizations are striving to dominate the summit and establish an unregulated status quo, potentially sparking a perilous competition,” he added.

Leahy pointed to a recent interview with Dario Amodei, the CEO of the AI company Anthropic, who indicated that the probability of an AI system causing catastrophic consequences “on the level of human civilization” ranged from 10% to 25%.

In reaction, Leahy suggested, “If you are in the process of developing a machine with such a probability, my recommendation is not to proceed.”

Geoffrey Hinton, commonly known as the “godfather of AI,” recently departed from Google, expressing concerns about what he termed the “existential risk” posed by digital intelligence.

Soon after, a collective of senior executives in the AI industry, including Leahy, released a concise statement asserting, “Mitigating the threat of AI-induced extinction should be a global priority, comparable to other societal-scale risks such as pandemics and nuclear war.”

These precautionary remarks led Sunak to organize the summit scheduled for November 1-2. The purpose of this gathering is to provide a forum for heads of state and technology leaders to engage in in-person discussions on these issues for the first time.

UK officials have been actively reaching out to leaders globally, urging their involvement in the conference, with the aspiration that it will be the inaugural event of a series of similar summits.

Eric Schmidt, former Google CEO, and Mustafa Suleyman, co-founder of Inflection and DeepMind, have recently endorsed the creation of a global panel of AI experts, akin to the Intergovernmental Panel on Climate Change. While UK authorities express their backing for such an initiative, they believe that the specific operational details should be deliberated through UN discussions.

Contrarily, sources within the UK government are optimistic that the AI industry will exhibit any sign of decelerating its quest for artificial general intelligence, a move unprecedented in the history of technology.

Leahy, however, argues that vague commitments to proceed with the responsible development of advanced AI would not be adequate. He asserts that such an approach would signify a triumph for tech companies over regulators.

A spokesperson for the government remarked, “The AI safety summit seeks to bring together a diverse range of participants, including international governments, academia, industry, and civil society, in a collaborative endeavor to push for targeted and prompt international measures for the safe and responsible progression of AI.