A Path to Regulating AI

Tejas Narechania
Tejas Narechania

In a recent Politico op-ed, Professor Tejas N. Narechania and co-author Ganesh Sitaraman argue that the same regulatory levers governments already use to govern utilities and other essential services should be a template for new rules for the burgeoning AI sector. 

The biggest companies now pouring money into AI — including Amazon, Google, and Microsoft — have all been sued for anticompetitive tactics, Narechania and Sitarmaran write, and are “poised to control” a broad new swath of innovation. 

The op-ed builds on a forthcoming article by Narechania, a faculty co-director of our Berkeley Center for Law & Technology, and Sitaraman, a Vanderbilt professor. 

“Leaving the development of such a revolutionary technology to a few unregulated mega-corporations is short-sighted at best and dangerous at worst. While AI might be new, the problems that arise from concentration in core technologies are not,” they write. “To keep Big Tech from becoming an unregulated AI oligopoly, we should turn to the playbook regulators have used to address other industries that offer fundamental services, like electricity, telecommunications and banking services.”

Their concerns stem from control over the “AI stack,” which is akin to a supply chain for AI applications. At the bottom are the microprocessing chips, topped by the cloud server farms that provide the massive computing power needed to train an AI foundation model. Those models then power specific products: ChatGPT, for example, is embedded in Microsoft’s Bing search engine. 

Amazon holds a huge swath of the cloud server market, while Microsoft has a 49% stake in Open AI, the parent of ChatGPT. With other companies dominating the chip manufacturing and another design, it’s easy to imagine how smaller competitors at every level of the stack could be squeezed out or have their ideas swiped by the big players. 

One possible regulation Narechania and Sitaraman suggest is “structural separation,” or not allowing companies to hold a piece of multiple layers of the stack. Another is the idea of nondiscrimination, like net neutrality: If Comcast is your internet provider, it’s required to give you equal bandwidth for Netflix and Peacock, a subsidiary. 

The co-authors also envision the possibility of a public option for cloud computing, which could “serve government agencies and researchers who want to solve public problems, rather than using AI to make tech platforms even more addictive.” It also would almost certainly lower costs for new AI developers. 

Lawmakers have a toolbox at the ready, Narechania and Sitaraman conclude. But will they use it? 

“We have spent two decades learning the hard way what happens when tech companies have unchecked, unregulated power to swallow up markets and eliminate competition,” they write. “The rise of AI offers what might be our last, best chance to get this power under control.”

Narechania also attended a recent White House meeting on competition policy and AI.