Google Shareholders Call for More Transparency in AI Algorithms

• Google has defied shareholder demands for greater transparency about its algorithm.
• The proposal was presented at parent company Alphabet’s 2023 shareholder meeting.
• CEO Sundar Pichai emphasized the imperative of enhancing the potential and safety of new generative AI.

Google’s parent company Alphabet has voted against shareholder proposals seeking greater transparency around its algorithms.

Trillium Asset Management announced the proposal at Alphabet’s 2023 Annual Meeting of Shareholders. Trillium made a similar request last year, before the ChatGPT craze swept the tech industry.

In the proposal, the investor expressed concern about how algorithmic systems could have detrimental consequences in areas such as criminal justice and healthcare. Trillium also cited a New Zealand Royal Commission report that revealed that YouTube content helped radicalize those behind the 2019 Christchurch shootings. He argued that accountability and transparency in artificial intelligence are necessary in order for artificial intelligence technology to remain safe for society.

Google opposed the proposal, saying it already provided meaningful disclosures about its algorithm, such as through a website that provides an overview of how YouTube’s algorithm classifies content. He added that the company’s proprietary algorithms are the foundation of its business and that they “could be abused.”

In a statement to the contrary, Google said, “When considering the algorithm transparency rule, we do not want information to be misused by malicious parties, affect user privacy, or expose commercially sensitive information. “We have to consider the serious risk that it could happen.”

Trillium requested further information about “how Alphabet uses algorithmic systems for targeting and serving ads, error rates, and the impact these systems have on users’ voices and experience.” The company also suggested considering recommendations from researchers at the Mozilla Foundation and New York University on standards for algorithmic and advertising transparency.

During the Q&A session at the end of the shareholder meeting, Google CEO Sundar Pichai was asked how to maximize the potential of AI while mitigating risks.

“It’s been seven years since we started our journey as an AI-first company,” Pichai said at the conference. “We have long worked to incorporate AI into our products to make them more convenient. We believe it can have an impact, and we are taking it boldly and doing it responsibly.”

Google advised shareholders to vote against all 11 proposals put forward by shareholders. Other proposals included asking Google to reconsider its plans to build a data center in Saudi Arabia, citing privacy and human rights concerns. Others suggested that Google do more to prevent users’ information from being shared with law enforcement when searching for abortion-related services in states with restricted access.

Trillium has long been a major shareholder of Google, owning about $135 million worth of the company’s stock, according to recent SEC filings. Trillium did not immediately respond to a request for comment.

The call for greater transparency comes shortly after Google vice president Jeffrey Hinton left the company and publicly warned about the dangers of new AI-based chatbots, saying the potential was “very scary.” was broken And Timnit Gebrew, who previously led Google’s AI team in 2020 and found in research that the company’s algorithms could perpetuate sexism and racism, clashed with the company over a paper he co-authored. He said he was fired for doing so.

Algorithmic bias is not a new concern. His 2016 report in ProPublica found that algorithmic software used by U.S. judges disproportionately weighted people of color, perpetuating racial inequality . Still, the generative AI boom has improved AI’s ability to write and even write code with near-human capabilities.

Any Google tips? You can reach out to this reporter via email at tmaxwell@insider.com, Signal at 540.955.7134, or Twitter at Twitter. @tomaxwell.