Last month, DeGraaf traveled to Sacramento to speak to several state lawmakers key to AI regulation:
The meeting to discuss the bills was at least the sixth visit to Sacramento by de Graaf or other EU officials in two months. EU officials who helped write the AI Act and EU Commission Vice President Josep Fonteles also made trips to Sacramento and Silicon Valley in recent weeks.
This week, EU leaders ended a years-long process with the approval of the AI Act, which regulates the use of artificial intelligence in 27 countries. It bans emotion recognition in schools and the workplace, social credit scores such as those used in China to reward or punish certain types of behavior and bans some examples of predictive policing. . The AI Act applies high-risk labels to AI in healthcare, hiring and issuing government benefits.
There are some notable differences between EU law and California law. The AI Act specifies how law enforcement agencies can use AI, while Bauer-Kahan's bill does not, and Wicks' watermarking bill may be stronger than the AI Act's requirements. But both the California bill and the AI Act take a risk-based approach to regulation, both recommending continuous testing and evaluation of forms of AI considered high risk, and both requiring watermarking of generative AI outputs. demand.
“If you take those three bills together, you're probably 70%–80% of what we put into the AI Act,” DeGraff said. “It's a very strong relationship that benefits both of us.”
At the meeting, de Graaff said he reviewed draft AI bills, AI bias and risk assessment, advanced AI models, the state of watermarking images and videos created by AI, and which issues to prioritize. The San Francisco office works under the authority of the EU Delegation in Washington DC to promote EU tech policy and strengthen cooperation with influential tech and policy figures in the US.
Artificial intelligence can make predictions about people, including what movies they want to watch on Netflix or the next words in a sentence, but without high quality and constant testing, important information about people's lives. Decision-making AI can automate discrimination. AI has a history of harming people of color, such as police using facial recognition to decide whether to apply for an apartment or a home mortgage. Technology has been shown to have a negative impact on the lives of many people, including women, people with disabilities, the young, the elderly and those applying for government benefits.
In a recent interview with KQED, Umberg spoke about the importance of striking a balance, insisting that “we can get it wrong.” Too little regulation could have disastrous consequences for society, and too much could “strangle the AI industry” that calls California home.
The coordination between California and EU authorities seeks to unify regulatory initiatives in two uniquely influential markets.
The majority of top AI companies are based in California, and over the past eight months, San Francisco Bay Area companies have raised more AI investment money than the rest of the world combined, according to startup tracker Crunchbase.
The General Data Protection Regulation, known as the GDPR, is the European Union's best-known piece of legislation to protect privacy. This has also led to the term “Brussels effect”, when the enactment of one law leads to greater influence in other countries. In this case, EU law forced tech companies to adopt tougher consumer protections if they wanted access to the region's 450 million inhabitants. The law went into effect in 2018, the same year California passed a similar law. More than a dozen US states followed suit (PDF).
Definition of AI
Coordination is important, De Graaff said, because technology is a global industry and policies that make it difficult for businesses to comply with laws around the world must be avoided.
One of the first steps in working together is a common definition of how to define artificial intelligence so that you agree on what technology is covered under the law. De Graaf said his office worked with Bauer-Kahan and Umberg on how to define AI “because if you have very different definitions to begin with, then convergence or harmonization is almost impossible. “
Given the recent passage of the AI Act, the absence of federal action, and the complexity of regulating AI, Senate Judiciary staff attorneys have held numerous meetings with EU officials and staff, Amber told CalMatters in a statement. . The definition of AI used by California's Senate Judiciary Committee is informed by multiple voices, including federal agencies, the Organization for Economic Co-operation and Development, and the EU.
“I strongly believe that we can learn from each other's work and manage AI responsibly without harming innovation in this dynamic and rapidly changing environment,” Umberg told CalMatters in a written statement. .
A trio of bills discussed with DeGraaf in April passed their respective chambers this week. He suspects California lawmakers' questions will become more specific as the bills get closer to passage.
California lawmakers proposed more than 100 bills to regulate AI in the current legislative session.
“I think what's important now for the Legislature is to reduce the bills to a more manageable number,” he said. “I mean, there are more than 50 so we focused specifically on the bill for those assembly members or senators.”
The state agency also seeks to protect the privacy of Californians.
Elected officials and their staff are not the only ones who speak to EU officials. The California Privacy Protection Agency — a state agency created to protect people's privacy and require businesses to comply with data deletion requests — also regularly talks with EU officials, including de Graaf.