When exploring the specifics of California's recent disclosure on high-risk AI systems, it becomes clear that the state has shifted its transparency practices. Initially claiming not to use any high-risk AI tools, California's state government acknowledged the existence of six such systems after a comprehensive inventory was conducted by the Department of Technology on June 15. This inventory, mandated by the recently passed AB 302 legislation, highlights how automated decision-making tools are employed in critical areas like criminal justice and public benefits.
What systems are being used in California? The disclosed AI systems are aimed at predicting recidivism rates among incarcerated individuals and assessing potential unemployment fraud. This revelation comes not because California has adopted new technology overnight, but rather due to improvements in the evaluation and reporting processes of existing systems. Notably, six other systems initially flagged as high-risk during this review were reassigned to a lower risk status, indicating a more nuanced understanding of the state's AI landscape.
What drove this change in disclosure practices? The push for transparency stems from AB 302, a law enacted late in 2023. This law requires state agencies to compile an annual inventory of their automated systems and report on any high-risk applications. Importantly, this requirement continues through 2029, aimed at ensuring ongoing scrutiny from legislators and the public alike. The crucial element of this change is that it does not ban the use of high-risk AI technologies but requires that their use be disclosed for scrutiny and accountability.
How does this impact the future of AI governance? Acknowledging these systems positions California at the forefront of discussing both the opportunities and the risks associated with AI. However, the focus now shifts to whether the state will enhance its transparency further by providing robust performance metrics and impact analyses related to these systems. While AB 302 marks a significant step toward accountability, it remains to be seen if more comprehensive measures will follow to ensure that these technologies operate equitably and effectively within the state's social fabric.