Thursday, 8 February 2024, Bengaluru, India
Anything that could be construed as damning or that you wouldn’t want another person to see should not be entered into Google’s Gemini family of GenAI apps.
(Image Source: https://tech.hindustantimes.com/)
That’s the PSA that Google is putting out today in a new support page detailing how it gathers user data from its Gemini chatbot applications on iOS, Android, and the web.
To enhance the service, Google reports that human annotators regularly read, tag, and process conversations with Gemini, even when those chats are “disconnected” from Google Accounts. (Google is silent on whether these annotators are employed inside or externally, which may be significant regarding data security.) Together with “related data” such as the user’s location and the languages and devices they utilize, these conversations are kept on file for three years.
Google now gives consumers some choice about what Gemini-related data is kept around and for how long.
Gemini Apps Activity is on by default; to avoid having future conversations with Gemini saved to a Google Account for review, turn it off in Google’s My Activity dashboard. This will eliminate the three-year review timeframe. On the Gemini Apps Activity panel, however, you can remove specific prompts and discussions with Gemini.
However, to “maintain the safety and security of Gemini apps and improve Gemini apps,” Google states that Gemini discussions will be kept to a Google Account for up to 72 hours even when Gemini Apps Activity is turned off.
Google notes that it is essential that users refrain from entering confidential information or any other type of data in their conversations that they wouldn’t want reviewers to see or use by the company to enhance its machine learning systems.
In all fairness, Google’s policies on gathering and keeping GenAI data don’t differ from those of its competitors. For example, OpenAI retains all of its talks with ChatGPT for 30 days unless the user has paid to an enterprise-level plan with a specific data retention policy. ChatGPT’s conversation history function is not saved.
However, Google’s policy highlights the difficulties in balancing privacy and developing GenAI models that use user data to learn and adapt.
In the recent past, vendors have found themselves in hot water with regulators due to liberal GenAI data retention rules.
Last summer, the FTC asked OpenAI to provide comprehensive information on how it verifies data—including customer data—used to train its algorithms and safeguards that data from unauthorized access. The Italian Data Protection Authority, which oversees data privacy in Italy, declared that OpenAI did not have a “legal basis” for gathering and storing large amounts of personal data in order to train its GenAI models.
Organizations are becoming more cautious about privacy issues as GenAI products become more widespread.
According to a recent Cisco poll, 63% of businesses have restricted the data that may be entered into GenAI platforms. At the same time, 27% have outright prohibited GenAI. According to the same survey, 45% of workers have entered “problematic” data, such as personnel details and confidential company files, onto GenAI technologies.
Businesses that specifically don’t save data for long periods—whether for model training or any other reason—are the target market for GenAI products from OpenAI, Microsoft, Amazon, Google, and other companies. However, as is frequently the case, consumers wind up with the short end of the stick.
To summarize, Google’s default practice of retaining conversations with Gemini for an extended period highlights the continuous struggle to balance convenience and privacy. This situation emphasizes the need for heightened transparency, user education, and empowering individuals to actively control and make informed choices regarding the duration of storage for their digital communications. The changing landscape of digital privacy necessitates a thorough reconsideration of default settings, underscoring the significance of granting users agency in the digital domain.
(Information Source: Techcrunch.com)
