New York’s AI Bill: Transparency or Barrier?

by Kay Lee

In recent months, a bold legislative move in New York sought to redefine the landscape of artificial intelligence development. The proposed Assembly Bill 8595, better known as the Artificial Intelligence Transparency for Journalism Act, aimed to open the black box of AI but inadvertently paved the way for unprecedented compliance challenges that have left its future uncertain.

The Bill’s Bold Ambitions

State Sen. Kristen Gonzalez brought AB 8595 to the forefront as a means to enforce transparency among AI developers. A core requirement mandated these developers to publish exhaustive lists of URLs and data sources accessed during model creation, veering towards an unprecedented level of detail. The objective? To ensure transparency in how AI systems gather and process information.

The Fine Line Between Transparency and Burden

However, as with many regulatory efforts, the bill straddles a delicate line. While aimed at fostering transparency, its requirements threaten to become barriers. Imagine the monumental task of tracing every digital footprint across billions of documents—an endeavor more taxing than it seems. The potential for such demands to shift from transparency enablers to innovation stiflers is a major concern. According to Reason Foundation, the legislation raised questions about its practical implications.

AI’s interaction with copyright law remains a hotbed of legal debate. The bill’s focus on data transparency echoes ongoing legal battles, like the landmark case alongside Anthropic, where a $1.5 billion settlement over unauthorized AI training set the tone. The legal field continues to grapple with defining ‘fair use’ in an AI-driven world, as seen in cases involving notable figures and companies.

Challenges Faced by AI Developers

Developers of large language models (LLMs) wrestle with the technical nuances of the bill’s demands. Tracking each URL accessed by automated crawlers goes against standard industry practices. These models do not store URL pointers but instead learn through complex numerical models—adding an extra layer to the compliance conundrum.

As the bill languishes in legislative limbo, the conversation is far from over. It embodies the ongoing tension between fostering innovation and safeguarding interests. AI developers and publishers remain at loggerheads, and a reappearance of such legislative efforts seems imminent, whether in New York or beyond.

In the evolving narrative of technology’s coexistence with regulation, New York’s experience serves as a critical reference point. The debates surrounding transparency, innovation, and legality continue to chart the uncertain future of artificial intelligence in the legislative realm.