Frontier AI,’ Power, and the Public Interest: Who Benefits, Who Decides

As the rapid industrialization of generative AI (GenAI) reached a crescendo in the fall of 2023, a series of international AI policy initiatives, like the UK AI Safety Summit and G7’s Hiroshima Process, cropped up as a response to corresponding global AI governance challenges. The policymakers and government officials who drove these initiatives emphasised that the rise of ‘frontier AI’ technologies was bringing humankind to a historical inflection point—placing humanity at a crossroads and situating the present generation at an axis of choice which would determine whether the evolution of AI innovation moves toward the exponential advancement of the public good or toward possibilities for potentially irreparable harm to people, society, and the planet. And yet, despite the inflationary rhetoric of historical transformation around which these policy initiatives were framed, their actual results (high-level voluntary commitments, non-binding codes of conduct, the formation of light-touch national AI safety institutes, etc.) seem vastly out of sync with the scope and scale of the problem to which such initiatives claimed to respond.

This article, co-authored by David Leslie, Carolyn Ashurst, Natalia Menéndez González, Frances Griffiths, Smera Jayadeva, Mackenzie Jorgensen, Michael Katell, Shyam Krishna, Doschmund Kwiatkowski, Carolina Iglésias Martins, Sabeehah Mahomed, Carlos Mougan, Shachi Pandit, Mark Richey, Joseph W. Sakshaug, Shannon Vallor, and Luke Vilainwe, argues that, if anything, this framing of the GenAI moment as a historical pivot point raises fundamental questions about the thorny relationship of ‘frontier AI,’ power, and the public interest, namely: Who actually has their hands on the wheel? Who defines the agenda of ‘frontier AI’ innovation? Who controls the means of producing it, and thus ultimately its influence on humanity’s broader fate? These questions cut much deeper than those surrounding the risks arising from unforeseen advances in ‘frontier AI’ capability or those around pre-deployment safety testing that took up much of the oxygen at the UK AI Safety Summit and related international AI policy discussions. They have to do with who possesses agenda-setting power, who decides on when, where, and how these technologies are developed and deployed, and who stands to benefit from or be harmed by them.