news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Source: Global Times

[Global Times Special Correspondent in the United States Zhuoran Global Times Special Correspondent Wang Yi] "California Artificial Intelligence (AI) Bill Causes Silicon Valley to Split." According to AFP on the 25th, the California State Legislature is promoting a bill aimed at regulating artificial intelligence models (i.e., "SB 1047 Bill", full name "Frontier Artificial Intelligence Model Safety Innovation Act"). The bill was submitted to the California State Assembly in February this year and caused great controversy. On August 19, the California State Legislature announced a revised version of the bill. The bill was supported by Jeffrey Hinton and Joshua Bengio, who are known as the "godfathers of artificial intelligence", but was also opposed by scientists such as LeCun and Fei-Fei Li, as well as leading artificial intelligence companies.

On August 20, a robot strolled in front of Amazon's artificial intelligence laboratory in San Francisco. (Visual China)

Want to avoid risks and promote innovation

According to US media, SB 1047 aims to establish security standards for developers with training costs exceeding $100 million or reaching a certain computing power to ensure the safe development of large-scale artificial intelligence models. The bill requires developers of large cutting-edge artificial intelligence models to take preventive measures, such as pre-deployment testing, simulated hacker attacks, installation of network security, and protection for whistleblowers.

Scott Wiener, a Democratic U.S. Senator and sponsor of SB 1047, said, "With Congress deadlocked over AI regulation, California must take action to prevent the rapidly developing AI from posing foreseeable risks while also promoting innovation." Wiener believes that AI safety and innovation are not mutually exclusive, and the revised version of the bill has addressed the concerns of some previous critics.

According to Business Insider on the 25th, the revised version released on August 19 contains several changes. First, it limits the scope of civil penalties for violations that do not cause harm or pose an imminent risk. Some key wording of the bill has also become looser. Calvin, a senior policy adviser at the Center for Artificial Intelligence Safety Action Fund, a co-proposing agency of the bill, said this "helps to make it clear that the focus of the bill is on testing and mitigating risks." Secondly, the bill no longer requires the establishment of a cutting-edge model department, but will still establish a cutting-edge model committee and place it within an existing government operating agency. The updated bill also eliminates the penalty for perjury, thereby exempting the relevant companies from criminal liability and only stipulates civil liability.

Some celebrities expressed support

Anthropic, a California artificial intelligence startup, proposed a series of amendments after the initial bill was announced, and expressed "cautious approval" a month later. In a letter to California Governor Gavin Newsom, Anthropic CEO Dario Amodei said that the bill currently has "more benefits than flaws." But he added, "We are not sure about this, and there are still some aspects of the bill that make us worried or ambiguous."

However, the bill still has the support of many well-known figures in the field of artificial intelligence. For example, University of Montreal professor Joshua Bengio and computer scientist Geoffrey Hinton. Hinton wrote in a column in Fortune magazine, "Powerful artificial intelligence brings incredible hope, but the risks are also real and should be taken extremely seriously. SB 1047 takes a very sensible approach to balancing these concerns."

Dan Hendrick, founder of the Center for AI Safety, told The New York Times that the bill would push big tech companies to identify and eliminate hazards contained in their expensive technology.

Opponents: “The disadvantages outweigh the advantages”

While there is widespread agreement in the tech community on the need to curb the risks of new AI models, critics say Wiener’s proposal would stifle startups, benefit U.S. competitors and weaken California’s position in the field.

Fei-Fei Li, a professor of computer science at Stanford University, is one of the well-known scholars who oppose the bill. She said that SB 1047 does not solve the problems it wants to solve, and will seriously damage the AI ​​academic community, open source community, etc., because developers will bear excessive responsibilities, which will limit their innovation. In addition, the bill requires all models that exceed a certain threshold to have a "kill switch", a mechanism that can shut down the program at any time, which will make developers more hesitant in writing code and collaborating, and research in the public sector and academia will be weakened due to lack of collaboration and other reasons. She also said that the most worrying thing is that the bill does not address the potential hazards of the development of artificial intelligence, including bias and deep fakes.

OpenAI, an American artificial intelligence company, recently publicly expressed its opposition to the bill. Jason Quan, the company's chief strategy officer, wrote in a letter to Weiner that the bill threatens "California's unique position as a global leader in artificial intelligence." He added that it could "slow down the pace of innovation and cause world-class engineers and entrepreneurs to leave the state to seek opportunities elsewhere." Another technology company, Meta, believes that the bill "prevents the release of open source artificial intelligence models."

Critics, including U.S. Democratic lawmakers, believe that threatening to take punitive measures against developers in an emerging field could stifle innovation. Former House Speaker Pelosi also issued a statement opposing the bill on the 16th. The statement said, "Many people in Congress believe that SB 1047 is well-intentioned but wrong." She said that senior Democratic Party officials have shared their concerns with Weiner, "Although we hope that California will lead the development of artificial intelligence in terms of protecting consumers, data, intellectual property rights, etc., SB 1047 does more harm than good."

California lawmakers, hoping to get Governor Gavin Newsom to sign a less controversial artificial intelligence bill, made significant changes to SB 1047, but failed to convince Silicon Valley's most important artificial intelligence lab that the bill is worth passing. SB 1047 is currently undergoing a final vote in the California State Assembly and may be submitted to Newsom's desk by the end of this month. Newsom has not yet stated his views on the bill, but his decision is widely watched by Silicon Valley and the US technology community.