Controversial Calif. AI Bill Could Have Indirect Impact on Telecom

The telecom sector could make use of AI in various ways. For example, AI could enhance operations, including in customer experience, as well as in call centers and field employee deployment, according to a Feb. 25, 2022, McKinsey & Company article. Meanwhile, Ericsson highlighted potential AI applications for radio access networks and network management in a Nov. 7 blog post. AI may also have applications for spectrum sharing, a recent RCR Wireless webinar indicated (see 2408280045). The California bill doesn't appear to specifically target those possible areas.

The AI bill by Sen. Scott Wiener (D) would authorize California's attorney general to bring civil actions against entities operating in the state for violations that cause death, public safety risks and other harms, according to an Aug. 29 Morgan Lewis blog. It would cover models that cost at least $100 million to create, or at least $10 million if developed by modifying an existing covered model. The California AI bill's critics, including tech entrepreneur Andrew Ng, say holding companies liable for harm that their use of AI causes would stifle innovation (see 2405300064). Musk, however, argues regulating products that pose public risks is appropriate (see 2408290005).

The Morgan Lewis lawyers don’t expect that the proposed law would “have significant direct effects on or create liability for telecom companies given that the bill is targeted to large AI models,” they told us in a joint statement. Leigha Beckman, Minna Lo Naranjo, Dion Bregman and Andrew Gray wrote the blog. “However, telecom companies are likely included in the definition of ‘critical infrastructure’ whereby a large AI model developer would need to account for any risk of disruption or cyberattacks,” the attorneys said. “It’s currently unclear how AI developers would ‘take into account’ the risk to critical infrastructures, but it’s possible that the telecom industry may need to interface with large AI models to ensure that there are possible protocols in place to properly identify disruption risks and ways to mitigate that risk.” Regardless of what SB-1047 requires, the telecom industry should identify “risks of disruption from AI and creat[e] red team exercises to better prepare for a mitigation plan.”

AI's potential use in the telecom sector includes enhanced user experience, network performance and energy consumption, an Ericsson spokesperson said. “These advancements must go hand in hand with progress in AI trustworthiness and safety, with a primary focus on preventing undesirable outcomes, in line with the reasoning behind the California AI bill.” Ericsson has “invested in research and development to ensure AI systems are both trustworthy and safe.”

Tech industry groups have raised concerns about how SB-1047 could influence AI development. The Computer & Communications Industry Associations (CCIA) worries it “may impact the broader AI ecosystem, not just larger developers,” said State Policy Director Khara Boender. “While the bill appears to only target models above certain computing power and cost thresholds, the downstream impacts could be felt by smaller businesses and startups, researchers, other independent labs, and academics who may build on such models.”

CTA and the Software & Information Industry Association urged Newsom to veto the bill. In an Aug. 30 letter, they wrote, “While well-intentioned, we are concerned that SB 1047 proposes requirements that would fundamentally undermine innovation, hurt the California economy, and weaken U.S. national security.” Separately, SIIA and CCIA joined TechNet and other business groups opposing the bill in an Aug. 26 alert addressed to the California Assembly.

However, consumer groups have supported the measure. “We see the liability imposed by SB 1047 as important accountability the bill will bring to AI models,” Consumer Watchdog’s Justin Kloczko emailed. “The bill has already been watered down with reduced civil penalties and narrower testing requirements.” The AG may sue for injunctive relief only after a mass casualty or mass damage event, and not for reasons of negligence, he noted. The bill no longer requires safety tests to be submitted under oath, he added.

“It’s hard to see that this bill will stop California from being a leader in tech innovation,” said Kloczko. “Google and Facebook are still the most powerful companies in the world. And after the passage of data privacy rights such as the CCPA, tech companies are still doing business and making billions in profits.”

In an Aug. 28 statement after the bill passed the legislature, Wiener called it “a light touch, commonsense measure that codifies commitments that the largest AI companies have already voluntarily made.”

“This measure will be evaluated on its merits,” a Newsom spokesperson said Friday.