The other night I attended a press dinner hosted by a company called Box. Other guests included the leaders of two data-oriented companies, Datadog and MongoDB. Typically, executives at these evenings are on their best behavior, especially when the discussion is official, like this one. So I was surprised by an exchange with Box CEO Aaron Levie, who told us that he had a tough stop at dessert because that night he was flying to Washington, DC. He was headed to a special interest marathon called TechNet Day, where Silicon Valley goes on a speed date with dozens of Congressional creatures to determine what the (uninvited) public will have to live with. And what did you want from that legislation? “As little as possible,” Levie replied. “I will be the only one responsible for stopping the government.”
I was joking about that. Something like. He went on to say that while regulating clear abuses of AI, such as deepfakes, makes sense, it is too early to consider restrictions such as forcing companies to submit large language models to government-approved AI police, or scanning chatbots in look for things like prejudice or the ability to hack real-life infrastructure. He pointed to Europe, which has already adopted restrictions on AI, as an example of what No do. “What Europe is doing is quite risky,” she said. “There is a view in the EU that if you regulate first, you create a kind of atmosphere of innovation,” Levie said. “That has been shown empirically to be incorrect.”
Levie’s comments fly in the face of what has become a standard position among Silicon Valley AI elites like Sam Altman. “Yes, regulate us!” they say. But Levie points out that when it comes to what exactly the laws should say, the consensus breaks down. “We, as a tech industry, don’t know what we’re really asking for,” Levie said, “I haven’t been to a dinner with more than five AI people where there is a single agreement on how to regulate AI.” Not that it matters: Levie believes dreams of a comprehensive AI bill are doomed to failure. “The good news is that “There is no way the United States will ever coordinate in this way. There simply will not be an AI Law in the United States.”
Levie is known for his irreverent loquacity. But in this case he is simply more sincere than many of his colleagues, whose please-regulate-us position is a sophisticated form of drug rope. The only public TechNet Day event, at least as far as I could discern, was a live streamed panel discussion on AI innovation that included Google president of global affairs Kent Walker and Michael Kratsios, most recently US CTO and now an executive at Scale AI. The sentiment among the panelists was that the administration should focus on protecting American leadership on the ground. While they admitted that the technology has its risks, they argued that existing laws largely cover potential unpleasant effects.
Google’s Walker seemed particularly alarmed that some states were developing AI legislation on their own. “In California alone, there are 53 different AI bills pending in the legislature today,” he said, and he wasn’t bragging. Walker, of course, knows that this Congress will hardly be able to keep the government afloat, and the prospect of both chambers juggling this hot potato in an election year is as remote as Google rehiring the eight authors of the article on transformers.
The United States Congress has pending legislation. And the bills keep coming, some perhaps less significant than others. This week, Rep. Adam Schiff, D-Calif., introduced a bill called the Generative AI Copyright Disclosure Act of 2024. It requires large language models to submit to the copyright office “a sufficiently detailed summary of any copyrighted work used… in the training data set.” It is not clear what “sufficiently detailed” means. Would it be okay to say, “We just eliminate the open network”? Schiff’s staff explained to me that they were adopting a measure in the EU AI bill.