California lawmakers have passed a comprehensive bill to monitor how employers and industries use automated decision-making tools, from algorithms to filter out job applicants to programs to detect academic misconduct. , trying to lead artificial intelligence surveillance.
Rebecca Bauer-Kahan (Democrat) is sponsoring the only AI-related bill in the California Legislature, the bill (AB 331), which imposes evaluation requirements on the use of such software by the private sector, even across the country. One of the few bills. .
The state bill comes after federal action on the topic has been delayed. AI has been in the public eye amid concerns about its potential abuse, but it was only recently that the Biden administration sought public input on how to build a framework to oversee her AI program. That’s it. Legislators in multiple states have introduced a series of bills, but most either seek to investigate this topic or simply affect how governments use AI.
The Bauer-Kahan Act is broader as it covers discrimination from AI software in employment, education, housing, utilities, healthcare, financial services and other areas.
“Today, 50% of businesses are using these automated decision-making tools to make important decisions. We have to,” said Bauer-Kahan. “We are not ahead of the ball. We are already behind the ball.”
Valuation and protection
Business lobbyists are eyeing a bill that is passing a legislative committee. The move is guided by the framework of the Biden administration’s AI Bill of Rights, he said, Bauer-Kahan. The Biden document outlines broad principles to prevent automated decision-making systems from causing discrimination and other harm.
The law details how creators and users can scrutinize AI systems. Instead of relying on independent third-party audits, Bauer-Kahan’s action requires developers (those who create or code automated tools) and tool users to complete their annual impact assessments with the California Department of Civil Rights by 2025. must be submitted to The bill would be the first bill in the country to divide those responsibilities, observers said.
“It’s an interesting choice. “Some kind of evaluation needs to be done by the people who develop that model. But depending on how it’s used, if it’s used in medicine and if it’s used to write poetry. A completely different assessment would be required.”
Under the bill, the assessment would include how automated decision-making tools are used, what data is collected, what safeguards are in place, and what adverse impacts may occur. and how the tool was evaluated. In addition, companies should implement governance programs that enforce these safeguards.
The measure requires those using AI to create a public policy listing the types of automated tools used and how the company manages the risk of unlawful discrimination. Where decisions are made based solely on AI, affected individuals have the right to opt out if the request is “technically feasible”.
business concerns
Many of the terms in the law are unclear, and industry groups have expressed concerns about compliance, such as how to determine what exactly constitutes a violation, or how to determine what is “technically feasible” in an opt-out request. At a hearing on April 11, California Chamber of Commerce policy advocate Ronak Daylami said the frequency and scope of assessments could also be burdensome for small businesses. Said there was.
The bill would partially address such concerns by extending the requirement to businesses that use technology and have 25 or more employees, as long as the tool does not affect at least 1,000 people annually. I’m here.
The biggest concern for businesses is that the penalties under this bill could adversely affect innovation in this area because of potential financial penalties.
Under this action, state civil rights offices can impose fines of $10,000 each day if an impact assessment is not submitted. But more of a concern among his business lobbyists is the private lawsuit right that would allow state residents to file lawsuits, which his other AI proposals include. Friedler said it is a provision that is not covered by the law. The company or developer is given 45 days to rectify the violation to avoid a lawsuit seeking injunctive relief, but business groups say that isn’t enough time.
Bauer-Kahan defended these provisions, but sometimes compromised.
“So the impact assessment itself is a remediation mechanism, right? – Kahan said. “There is currently an off-ramp related to injunctive relief, but we know the opposition would like to see more of it, and we are talking about it.”
Rep. Bill Essayli (Republican) noted that California law already outlaws discrimination, whether by humans or algorithms, and the hearing called for the creation of a new private action right. I asked if it was necessary.
Future work
The California Department of Civil Rights Council is in the process of adapting existing employment discrimination regulations regarding the use of AI. There are also rules that have yet to be made by the California Privacy Protection Agency regarding privacy protections for automated systems. Industry groups said they were concerned that the Bauer-Kahan bill would conflict with those regulatory efforts.
But privacy advocates argue that the bill would also complement future rulemaking. would also hold employers accountable for discrimination by automated tools. This provision places an additional burden on employers to understand how other companies’ automated systems work.
The Bauer-Kahan bill strengthens the proposed language because it requires developers of AI tools to send certain information to deployers of the technology. The bill also clarifies that trade secrets need not be disclosed as part of compliance. This is a concern about the council’s draft regulations.
More importantly, Bauer-Kahan’s bill echoes Biden’s proposal in that it would require advance notice for those targeted by automated decision tools. This provision has implications for potential state regulation, as it is difficult to enforce protections against AI abuse if residents do not know that AI is being used.
Bauer-Kahan said he still expects the bill to be amended, but fellow Democrats on the state legislature’s Privacy and Consumer Protection Committee voted 8 to 3 at a hearing on April 11. California should take the lead in AI, despite opposition from business, lawmakers debate said inside.
“It will never be perfect. On regulating AI, Rep. Josh Lowenthal (Democrat) said: “This will forever be a work in progress. It will never be completed in
