Google has acknowledged that it reserves key search data exclusively for its Gemini AI service, limiting access to rival artificial intelligence applications.
The admission came from Elizabeth Reid, Google’s head of search, who testified in a historic antitrust trial that certain features and data are withheld from competitors, effectively prioritizing Gemini over other AI platforms. This revelation has raised concerns about Google’s data-sharing practices with artificial intelligence companies.
In her testimony Tuesday, Reid described how Google has been giving select artificial intelligence companies access to some search results for use in training their models.
The service is accessible through Google’s Vertex AI platform, which sits in the company’s Google Cloud. However, Reid’s comments showed that what third-party companies can access vs. what Google’s internal AI service, Gemini, can access is worlds apart.
Reid acknowledged that while outside companies using Google’s search results via Vertex AI can create chatbot responses, Gemini is granted access to more tools that only Google can provide. These include the Knowledge Graph, which helps answer questions or analyze the relationships between entities, or OneBox, which can return immediate answers to questions or factoids like sports scores or flight information.
These additional data layers mean that Gemini is better equipped to paint more accurate, nuanced responses and represents a clear point of difference for the company compared to competing models that can only provide basic search results.
DOJ demands broader search access for rivals
The revelations of Google’s favoritism for Gemini also come as the U.S. Department of Justice (DOJ) calls for more stringent regulation to foster fair competition in the tech space.
The DOJ proposes that Google may need to open up a significant chunk of the data that the search giant uses to create its search results for competitors.
The DOJ has proposed that Google should grant competitors access to a substantial portion of the data it uses to generate search results. It is part of an effort to tamp down accusations that the company has an iron grip on online search and its underlying technology.
Reid punched back at the DOJ’s proposal, calling it “far-reaching and intrusive.” She said compelling Google to share its search data with rivals would mean sharing a “sensitive proprietary commodity” prone to abuse, like data breaches.
Along with the privacy issues, Reid also raised logistical challenges related to implementing the DOJ’s proposal. She added that Google’s search infrastructure is complicated, with different models used for different features.
The idea from the DOJ may be untenable to implement, but given the way Google has built its search and AI, the idea could be insinuated into Maverick’s proposal.
The proposal has ignited a heated debate in the tech industry, with some saying it would allow fair competition because smaller companies and startups would have access to the same data that Google uses to improve its search algorithms.
However, others, including Google, worry that it could introduce major security vulnerabilities and compromise the legitimacy of Google’s search system.
AI companies demand fair access to search data
Google’s data-sharing practices have drawn reactions from other key players in the AI space. Meta CEO Mark Zuckerberg highlighted the potential benefits of integrating Meta AI with Google, stating that it would significantly improve real-time access to information, underscoring the value of building effective AI for access to Google search data.
Likewise, OpenAI exec Nick Turley said, in a trial, that the company had contacted Google last year about a potential partnership that would allow ChatGPT to use Google’s search technology. Still, Google chose not to partner with OpenAI, and Turley said they have no partnership with the firm today.
These disclosures suggest that Google’s unwillingness to share search data with other AI firms may be a central issue to the trial and a broader industry theme.
Even companies with close ties to Google, like the AI startup Anthropic, would want some of that same access to Google’s search results to train better models. Despite Google’s hefty ownership stake in the company, Reid was unsure whether Anthropic’s request for additional data had been approved.
KEY Difference Wire helps crypto brands break through and dominate headlines fast
